Medmain

• CAS-Nummer: 576-11-4
• Molekularformel: C13H18N2
• Molekulargewicht: 202.30 g/mol
• IUPAC-Name: 3-ethyl-N,N,2-trimethyl-1H-indol-5-amine
• InChI: InChI=1S/C13H18N2/c1-5-11-9(2)14-13-7-6-10(15(3)4)8-12(11)13/h6-8,14H,5H2,1-4H3
• InChI-Schlüssel: IOYNGCZNYGEZRO-UHFFFAOYSA-N
• Kanonische SMILES: CCC1=C(NC2=C1C=C(C=C2)N(C)C)C

Beschreibung

Medmain is a useful research compound. Its molecular formula is C13H18N2 and its molecular weight is 202.30 g/mol. The purity is usually 95%.
BenchChem offers high-quality Medmain suitable for many research applications. Different packaging options are available to accommodate customers' requirements. Please inquire for more information about Medmain including the price, delivery time, and more detailed information at info@benchchem.com.

Foundational & Exploratory

Medmain: A Technical Deep Dive into AI-Powered Digital Pathology

The PidPort Platform: An Overview

PidPort is a comprehensive digital pathology platform designed to streamline the pathology workflow.[1][3] Its core functionalities can be categorized into three main pillars:

• Cloud-Based Storage and Management: PidPort offers secure cloud storage for digitized whole slide images (WSIs), eliminating the need for physical slide archives and enabling easy access and sharing of pathological data.[3][4] This feature is crucial for facilitating remote collaboration, telepathology, and the creation of large-scale databases for research.

• Telepathology and Collaboration: The platform provides a high-speed WSI viewer, enabling pathologists and researchers to remotely view, annotate, and share cases for consultations, conferences, and educational purposes.[3] This is particularly beneficial in addressing the geographical disparity of pathology expertise.

Core Technology: Deep Learning in Pathological Image Analysis

Transfer Learning: This approach leverages pre-trained CNN architectures, such as EfficientNet, which have been trained on massive datasets like ImageNet.[6] By fine-tuning these models on specific pathology datasets, Medmain can achieve high performance with faster convergence, even with limited annotated medical images.[6]

Weakly-Supervised Learning: A significant challenge in computational pathology is the need for vast amounts of meticulously annotated data. To overcome this, Medmain employs weakly-supervised learning techniques. This method allows the AI model to be trained on WSIs with only slide-level labels (e.g., "cancer" or "no cancer"), without requiring precise annotations of cancerous regions. This dramatically reduces the annotation burden on pathologists and enables the use of larger datasets.

The general workflow for Medmain's AI model development can be visualized as follows:

References

• 1. Medmain inc. [en.medmain.com]
• 2. Japanese medtech startup Medmain raises $3.3m to grow its digital pathology product - Startup Weekly [startup-weekly.com]
• 3. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 4. Medmain inc. [en.medmain.com]
• 5. pidport.medmain.com [pidport.medmain.com]
• 6. Medmain inc. [en.medmain.com]

Medmain's AI in Pathological Image Analysis: A Technical Deep Dive

Fukuoka, Japan - Medmain Inc., a leading provider of digital pathology solutions, has developed a powerful suite of artificial intelligence (AI) technologies designed to revolutionize pathological image analysis. Centered around their cloud-based platform, "PidPort," these AI models leverage deep learning and transfer learning to assist pathologists in the rapid and accurate diagnosis of various cancers. This technical guide provides an in-depth overview of the core technology, experimental validation, and performance metrics for researchers, scientists, and drug development professionals.

Medmain's AI algorithms are trained on vast datasets of digitized whole slide images (WSIs), annotated by expert pathologists.[1] The company has published numerous studies in peer-reviewed journals, demonstrating the efficacy of their models in identifying malignant lesions in a range of tissues, including the stomach, colon, breast, lung, prostate, and urothelium.[2][3][4][5]

Core Technology: Deep Learning and Transfer Learning

Medmain's core technology is built upon convolutional neural networks (CNNs), a class of deep learning models particularly adept at image recognition tasks. The development process typically involves:

• Weakly Supervised Learning: In many instances, the models are trained using a weakly supervised approach. This means that instead of requiring pathologists to manually outline every cancerous region on a slide (a time-consuming process), the AI can learn from slides that are simply labeled as cancerous or benign. This allows for the utilization of larger datasets.

• Transfer Learning: To enhance performance and reduce training time, Medmain employs transfer learning.[5] Pre-trained neural networks, which have already learned to recognize a vast array of features from large image datasets like ImageNet, are fine-tuned using Medmain's specific pathological image data. This approach is particularly effective, even with smaller, specialized datasets.

Experimental Workflow and Validation

The development and validation of Medmain's AI models follow a rigorous, multi-stage process, as depicted in the workflow diagram below. This process ensures the robustness and generalizability of the models across different patient populations and laboratory settings.

A generalized experimental workflow for the development and validation of Medmain's AI models.

Performance Metrics of Medmain's AI Models

The following tables summarize the performance of Medmain's AI models across various cancer types, as reported in their publications. The primary metric used for evaluation is the Area Under the Receiver Operating Characteristic Curve (AUC), which provides a comprehensive measure of a model's diagnostic ability.

Table 1: Performance in Gastrointestinal Cancers

Cancer Type	Specimen Type	Key Findings	AUC
Gastric Diffuse-Type Adenocarcinoma	Endoscopic Biopsy	High accuracy in classifying diffuse-type adenocarcinoma from WSIs.[6]	0.95 - 0.99
Poorly Differentiated Colorectal Adenocarcinoma	Endoscopic Biopsy	Effective classification using transfer learning with hard mining of false positives.[7]	Up to 0.977

Table 2: Performance in Urological and Lung Cancers

Cancer Type	Specimen Type	Key Findings	AUC
Prostate Adenocarcinoma	Needle Biopsy & TURP	High performance on needle biopsy and TCGA datasets.[4]	Up to 0.987
Urothelial Carcinoma	Liquid-Based Cytology	Accurate screening of neoplastic urothelial cells in urine specimens.[2]	0.984 - 0.990
Indeterminate Lung Carcinoma	Transbronchial Lung Biopsy	Differentiates between major histological types (ADC, SCC, SCLC) in challenging cases.[3]	0.94 - 0.99

Table 3: Performance in Breast and Cervical Cancers

Cancer Type	Specimen Type	Key Findings	AUC
Breast Invasive Ductal Carcinoma (IDC)	Surgical & Biopsy	Effective classification using weakly-supervised and transfer learning.[8]	0.96 - 0.98
Cervical Cancer	Liquid-Based Cytology	Aiding in the screening process for neoplastic and non-neoplastic cases.[9]	0.89 - 0.96

Detailed Experimental Protocols

The methodologies employed in Medmain's validation studies are crucial for understanding the reliability of their AI models. Below are summaries of the experimental protocols for key studies.

Gastric Diffuse-Type Adenocarcinoma Classification

• Objective: To develop a deep learning model to classify gastric diffuse-type adenocarcinoma from WSIs.

• Methodology: The study utilized deep learning models trained to classify this specific type of gastric cancer. The models were evaluated on five distinct test sets to ensure robustness.[6]

• Data: The training and validation datasets consisted of a large number of WSIs of endoscopic biopsy specimens.

Prostate Adenocarcinoma Classification

• Objective: To train a WSI prostate adenocarcinoma classification model using transfer learning and weakly supervised learning.

• Methodology: The models were evaluated on needle biopsy, transurethral resection of the prostate (TUR-P), and The Cancer Genome Atlas (TCGA) public dataset test sets.[4] This multi-dataset validation was designed to confirm the algorithm's applicability to different types of specimens.

• Data: The study included a substantial number of WSIs from both internal and public sources.

Lung Cancer Sub-classification

• Objective: To develop an AI model to classify lung cancer subtypes in lung biopsy specimens.

• Methodology: A deep learning model was trained to classify H&E-stained WSIs of transbronchial lung biopsy specimens into adenocarcinoma, squamous cell carcinoma, small-cell lung cancer, and non-neoplastic categories. The model's performance was validated on an independent test set of challenging indeterminate cases, with diagnoses confirmed by immunohistochemistry or surgical resection.[3]

• Data: The training set comprised 579 WSIs, and the model was tested on multiple independent sets, including 83 indeterminate cases and a combined total of 2407 WSIs from TBLB and surgical specimens.

Logical Relationships in Model Development

The development of Medmain's specialized AI models often involves a logical progression, starting from a foundational model and adapting it for more specific tasks. This is particularly evident in their work with adenocarcinoma classification.

References

• 1. A deep learning model for the classification of indeterminate lung carcinoma in biopsy whole slide images - PMC [pmc.ncbi.nlm.nih.gov]
• 2. Deep Learning-Based Screening of Urothelial Carcinoma in Whole Slide Images of Liquid-Based Cytology Urine Specimens - ProQuest [proquest.com]
• 3. A deep learning model for the classification of indeterminate lung carcinoma in biopsy whole slide images - PubMed [pubmed.ncbi.nlm.nih.gov]
• 4. A Deep Learning Model for Prostate Adenocarcinoma Classification in Needle Biopsy Whole-Slide Images Using Transfer Learning - PMC [pmc.ncbi.nlm.nih.gov]
• 5. mdpi.com [mdpi.com]
• 6. A deep learning model for gastric diffuse-type adenocarcinoma classification in whole slide images - PMC [pmc.ncbi.nlm.nih.gov]
• 7. Deep Learning Models for Poorly Differentiated Colorectal Adenocarcinoma Classification in Whole Slide Images Using Transfer Learning - PMC [pmc.ncbi.nlm.nih.gov]
• 8. mdpi.com [mdpi.com]
• 9. mdpi.com [mdpi.com]

PidPort: A Technical Guide to the Digital Pathology Platform for Researchers and Drug Development Professionals

Core Platform Architecture and Features

PidPort operates as a cloud-native platform, leveraging Amazon Web Services (AWS) for secure data storage and Auth0 for robust user authentication.[2] This architecture ensures scalability, accessibility, and data security, which are critical for research and clinical environments. The platform is accessible via a web browser, eliminating the need for local software installation and reducing the initial setup costs.[1][3]

Key Platform Features

The PidPort platform is built around three core pillars: Storage and Management, Browsing and Sharing, and AI Analytics.[2]

Feature	Description	Technical Details
Cloud Storage & Management	Securely stores and manages virtual slides and associated case information in the cloud.[2][3] This facilitates the creation of large, centralized databases for research and development.[2]	- Storage Backend: Amazon Web Services (AWS)[2] - Supported File Formats: SVS (Leica), NDPI (Hamamatsu Photonics), JPEG, TIFF[1] - Data Upload: Bulk upload functionality for pathology images and case data.[1]
Image Viewing & Collaboration	A high-speed, proprietary web viewer enables smooth visualization and interaction with large WSI files.[1] The platform supports real-time collaboration among multiple users.	- Viewer: Installation-free, web-based viewer[1][3] - Collaboration Tools: Annotation and comment functions for case discussions.[1] - Multi-view: Split view and sync function to compare up to four virtual slides simultaneously.[1]
AI-Powered Analysis	Integrates proprietary artificial intelligence models to assist in the analysis of pathological images. These models are developed using deep learning techniques.[1][4]	- AI Development: Utilizes deep learning and transfer learning on a dataset of hundreds of thousands of pathological specimens.[1] - Deployment: AI analysis is presented as a partner to pathologists, providing rapid and accurate image analysis.[1][3]

System Requirements

Component	Recommendation
Web Browser	Google Chrome[3]
PC Environment	Windows and macOS compatible[1]
Memory	4GB or more[1]

Experimental Protocols and AI Model Development

While detailed, proprietary experimental protocols for the development of all of Medmain's AI models are not publicly available, published research provides insight into their methodology. Medmain has successfully developed and validated AI models for various cancer types.

AI Model for Gastric Poorly Differentiated Adenocarcinoma Classification

A peer-reviewed paper published in Technology in Cancer Research & Treatment details the development of a pathological AI for classifying gastric poorly differentiated adenocarcinoma in endoscopic submucosal dissection (ESD) whole slide images.[3] The development was supported by a grant from the New Energy and Industrial Technology Development Organization (NEDO).[3]

Methodology (Based on available information):

• Data Collection: A large dataset of whole-slide images from ESD specimens of gastric poorly differentiated adenocarcinoma was collected.

• Annotation: Pathologists annotated the images to create a ground truth dataset for model training and validation.

• Model Training: A deep learning model, likely a convolutional neural network (CNN), was trained on the annotated dataset using weakly supervised learning.[3]

• Validation: The model's performance was rigorously validated against a separate test set to evaluate its accuracy in classifying the specified cancer type.

AI Model for Malignant Melanoma Differentiation

Medmain has also published a paper in Cancers on the development of an AI model to differentiate malignant melanoma in digital dermatopathology histology specimens.[5]

Methodology (Inferred):

• Data Acquisition: A substantial number of digital histology slides of skin biopsies, including malignant melanoma and benign nevi, were acquired.

• Expert Annotation: Dermatopathologists would have annotated regions of interest to train the AI model.

• Deep Learning: A deep learning framework was employed to train the AI to distinguish between malignant and benign lesions.

• Performance Evaluation: The model's diagnostic accuracy, sensitivity, and specificity were likely evaluated against the consensus diagnosis of expert pathologists.

Platform Workflow and Logical Relationships

The following diagrams illustrate the core workflows and logical relationships within the PidPort platform.

Digital Pathology Workflow with PidPort

This diagram outlines the end-to-end process from slide digitization to collaborative diagnosis and research within the PidPort ecosystem.

PidPort Digital Pathology Workflow

AI Model Development and Integration Logic

This diagram illustrates the logical flow of developing and integrating an AI model into the PidPort platform for automated analysis.

References

• 1. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 2. pidport.medmain.com [pidport.medmain.com]
• 3. Medmain inc. [en.medmain.com]
• 4. Medmain inc. [en.medmain.com]
• 5. PidPort｜Digital Pathology AI Analysis Solution [us.medmain.com]

Medmain's Deep Learning Models for Cancer Diagnosis: A Technical Overview

For Researchers, Scientists, and Drug Development Professionals

Introduction

Core Technology: Deep Learning in Digital Pathology

Medmain's approach to cancer diagnosis leverages the power of deep learning, a subset of machine learning, to analyze whole-slide images (WSIs) of pathological specimens.[4][6] The company has developed AI screening models for a variety of cancers, including but not limited to, gastric, colon, breast, lung, prostate, cervical, and urothelial cancers.[7][8] These models are trained on vast datasets, reportedly comprising hundreds of thousands of WSIs, to learn the intricate patterns and features indicative of malignancy.[3][5]

The development and training of these models are facilitated by Medmain's cloud-based "PidPort" platform, which includes an in-house annotation tool allowing specialist pathologists to create the necessary training data.[2] The platform's high-speed viewer and extensive annotation capabilities are crucial for the creation of high-quality datasets that form the foundation of their deep learning models.[2][3]

Deep Learning Methodologies

• Transfer Learning: This technique involves using a pre-trained neural network, typically one that has been trained on a large, general-purpose image dataset like ImageNet, and fine-tuning it for a specific medical imaging task. This approach is particularly effective as it leverages the feature extraction capabilities learned from a massive dataset, which can then be adapted to the nuances of pathological images. Medmain's research indicates that fine-tuning only the trainable weights of batch normalization layers can lead to performance comparable to fine-tuning all weights, but with the benefit of faster convergence.

• Fully Supervised Learning: In this standard approach, the deep learning model is trained on a dataset where every image tile is explicitly labeled by a pathologist (e.g., as "cancer" or "non-cancer"). This method requires detailed and labor-intensive annotation but can lead to high accuracy.

• Weakly Supervised Learning: Given the time-consuming nature of detailed annotation, Medmain also utilizes weakly supervised learning. In this paradigm, the model is trained on WSIs that are labeled at the slide level (e.g., the entire slide is labeled as containing cancer), without specific annotations for every cancerous region. The model then learns to identify the relevant features and regions corresponding to the slide-level label.

Experimental Workflows

The development and validation of Medmain's deep learning models follow a structured workflow, which can be generalized from their publications. The key stages are illustrated in the diagrams below.

Figure 1: Generalized Experimental Workflow.

Figure 2: WSI Tiling and Inference Workflow.

Model Architectures

• EfficientNet: A family of models that achieves high accuracy with fewer parameters and faster inference times by systematically scaling the network's depth, width, and resolution. The publications often refer to the use of EfficientNet-B1.

• ResNet (Residual Network): A deep neural network architecture that utilizes "skip connections" or "shortcuts" to allow the training of much deeper networks. ResNet-50 is a commonly used variant.

The general architecture of these models involves a CNN for feature extraction from image tiles, followed by a classification head to predict the probability of malignancy.

Figure 3: High-Level Model Architecture.

Performance of Deep Learning Models

Prostate Adenocarcinoma Classification

Dataset	Number of WSIs (Training)	Number of WSIs (Validation)	Number of WSIs (Test)	Model Architecture	Performance (ROC-AUC)
Needle Biopsy	1122	60	60	TL-colon poorly ADC-2 (20x, 512)	0.967 - 0.978
TUR-P	Not specified in detail	Not specified in detail	Not specified in detail	TL-colon poorly ADC-2 (20x, 512)	0.7377 - 0.9098
TCGA	Not specified in detail	Not specified in detail	Not specified in detail	TL-colon poorly ADC-2 (20x, 512)	0.9873

Gastric Signet Ring Cell Carcinoma Classification

Dataset	Number of WSIs (Training)	Number of WSIs (Test Set 1)	Number of WSIs (Test Set 2)	Model Architecture	Performance (ROC-AUC)
Internal & DigestPath2019	1765	999	455	EfficientNet-B1 (Weakly Supervised)	≥ 0.99

Cervical Cancer Screening (Neoplastic vs. Non-neoplastic)

Dataset	Number of WSIs (Training)	Number of WSIs (Test)	Model Architecture	Performance (ROC-AUC)
LBC Specimens	1605	1468 (3 sets)	CNN + RNN	0.89 - 0.96

Urothelial Carcinoma Screening (Neoplastic vs. Non-neoplastic)

Dataset	Number of WSIs (Training)	Number of WSIs (Test)	Model Architecture	Learning Approaches	Performance (ROC-AUC)
Urine LBC Specimens	786	750 (2 sets)	CNN-based	Transfer, Fully Supervised, Weakly Supervised	0.984 - 0.990

Experimental Protocols

The following sections detail the generalized experimental protocols based on the methodologies described in Medmain's publications.

Data Collection and Preparation

• Specimen Selection: Pathological specimens (e.g., needle biopsies, liquid-based cytology) are collected from multiple medical institutions.

• WSI Scanning: Glass slides are digitized into high-resolution WSIs using a scanner.

• Quality Control: WSIs are reviewed for quality, and those with significant artifacts may be excluded.

• Annotation: Pathologists use Medmain's PidPort annotation tool to label WSIs. For fully supervised learning, specific regions of interest (e.g., cancerous cells) are annotated. For weakly supervised learning, slide-level diagnoses are provided.

• Dataset Splitting: The annotated dataset is split into training, validation, and test sets.

WSI Tiling and Preprocessing

• Tissue Detection: An algorithm is used to identify tissue regions within the WSI to avoid processing the background.

• Tiling: The identified tissue regions are divided into smaller, fixed-size tiles (e.g., 512x512 pixels) at a specific magnification (e.g., 20x).

• Tile Selection: A strategy is employed to select tiles for training, which may involve filtering out tiles with insufficient tissue or other criteria.

• Data Augmentation: To increase the diversity of the training data and prevent overfitting, various data augmentation techniques are applied to the tiles. These may include:

  • Flipping (horizontal and vertical)

  • Rotation

  • Scaling

  • Color augmentation (e.g., brightness, contrast, saturation adjustments)

Model Training and Validation

• Model Selection: A base CNN architecture (e.g., EfficientNet-B1) is chosen.

• Training Strategy: The model is trained using one of the aforementioned strategies:

  • Transfer Learning: A pre-trained model is fine-tuned on the pathology dataset.

  • Fully Supervised: The model is trained on meticulously annotated tiles.

  • Weakly Supervised: The model is trained using slide-level labels.

• Hyperparameter Tuning: Key hyperparameters (e.g., learning rate, batch size, number of epochs) are optimized using the validation set to achieve the best performance.

• Model Evaluation: The trained model's performance is evaluated on the held-out test set using metrics such as the Area Under the Receiver Operating Characteristic Curve (ROC-AUC), sensitivity, and specificity.

Signaling Pathways

Conclusion

Medmain has demonstrated the development of high-performing deep learning models for the diagnosis of a range of cancers from whole-slide images. Their use of advanced techniques such as transfer learning and weakly supervised learning, coupled with their PidPort platform for data annotation and management, positions them as a significant contributor to the field of computational pathology. The quantitative results from their published studies indicate that their AI models can achieve high accuracy, sensitivity, and specificity, suggesting their potential as valuable tools to support pathologists in their diagnostic workflow. Future research and development will likely see the expansion of their AI models to a wider array of cancer types and the continued refinement of their methodologies to further enhance diagnostic accuracy and efficiency.

References

• 1. ijrpr.com [ijrpr.com]
• 2. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 3. arxiv.org [arxiv.org]
• 4. Medmain inc. [en.medmain.com]
• 5. Medmain inc. [en.medmain.com]
• 6. 製品情報 | メドメイン株式会社 [medmain.com]
• 7. researchgate.net [researchgate.net]
• 8. Medmain inc. [en.medmain.com]

The Technology Behind Medmain's Whole-Slide Imaging: A Technical Guide

For Researchers, Scientists, and Drug Development Professionals

This guide provides an in-depth look at the core technologies behind Medmain's whole-slide imaging (WSI) and artificial intelligence-driven digital pathology platform, PidPort™. Medmain is at the forefront of applying deep learning to pathological diagnostics, aiming to enhance the accuracy and efficiency of cancer detection and classification. This document synthesizes available technical data from their publications and public communications to offer a comprehensive overview for research and development professionals.

The PidPort™ Digital Pathology Ecosystem

The general workflow within the Medmain ecosystem begins with the digitization of pathological specimens at the Medmain Imaging Center.[1] These whole-slide images are then uploaded to the PidPort™ cloud infrastructure, where they can be viewed, shared, and analyzed by pathologists and researchers.[2] The platform's AI capabilities provide screening and analytical support for various cancer types.[4][5]

Caption: A high-level overview of the Medmain PidPort™ digital pathology workflow.

Whole-Slide Imaging Technology

General specifications for high-throughput whole-slide scanners typically used in such applications are provided in the table below for context.

Parameter	Typical Specification	Significance in Digital Pathology
Objective Lenses	20x, 40x magnification	Determines the level of detail captured in the WSI. 40x is common for diagnostic purposes.
Numerical Aperture (NA)	High NA objectives (e.g., >0.75)	A higher NA provides better resolution, allowing for clearer visualization of cellular and subcellular structures.
Image Resolution	~0.25 µm/pixel at 40x	Defines the smallest resolvable detail in the image, critical for accurate morphological assessment.
Scan Speed	Varies (e.g., <60 seconds per slide at 20x)	High speed is essential for high-throughput digitization in large-scale studies and clinical practice.
Slide Capacity	1 to >400 slides	Batch processing capabilities are crucial for efficiency in a centralized imaging center.
Focusing Mechanism	Automated with Z-stacking capability	Ensures the entire tissue section is in focus, which is particularly important for thick specimens and cytology.

AI-Powered Pathological Analysis

Medmain has developed a suite of deep learning models for the detection and classification of various cancers. These models are trained on extensive datasets of hundreds of thousands of whole-slide images, annotated by specialist pathologists.[2][4] The primary AI methodology employed is deep learning, including the use of transfer learning and weakly-supervised learning techniques.[4]

Experimental Protocol for AI Model Development

While detailed, step-by-step protocols are proprietary, the general workflow for developing their deep learning models can be inferred from their publications and industry best practices.

Caption: A generalized workflow for the development of Medmain's pathology AI models.

Methodology Details:

• Data Acquisition: Whole-slide images are sourced from multiple medical institutions to ensure a diverse and robust dataset.[4]

• Annotation: A large number of specialist pathologists create the "teacher data" by annotating the WSIs.[2]

• Model Architecture: Medmain utilizes Convolutional Neural Networks (CNNs), a standard in image recognition tasks. Their publications mention the use of architectures like EfficientNet-B3 with transfer learning.

• Training: Models are trained using supercomputers, leveraging techniques like weakly-supervised learning, which allows the model to learn from slide-level labels rather than requiring precise annotations of all pathological findings.[6]

• Validation: The models are rigorously tested on independent datasets from different sources to ensure their generalizability and robustness.

Performance of AI Models

Medmain has published performance data for several of their AI models. The following tables summarize the reported metrics for various cancer types.

Table 1: Performance in Gastrointestinal Cancers

Cancer Type	Task	Model Performance	Publication
Gastric & Colonic Epithelial Tumors	Classification	Not specified in abstract	Scientific Reports (2020)
Gastric Poorly Differentiated Adenocarcinoma	Classification	Not specified in announcement	Technology in Cancer Research & Treatment (2022)

Table 2: Performance in Lung Cancer

Task	Training Set Size	Test Set Size	Performance (AUC)	Publication
Indeterminate Lung Carcinoma Classification	579 WSIs	83 challenging cases + 4 independent sets (2407 WSIs total)	0.99 on indeterminate cases; 0.94-0.99 on independent sets	Scientific Reports (2020)
Lung Carcinoma vs. Non-neoplastic	3,554 WSIs	4 independent test sets	High AUCs reported	Scientific Reports (2020)

Table 3: Performance in Other Cancers

Cancer Type	Task	Training Set Size	Test Set Size	Performance	Publication
Pancreatic Ductal Adenocarcinoma	Detection on EUS-FNB	Not specified	Not specified	AUC: 0.984, Accuracy: 0.9417, Sensitivity: 0.9302, Specificity: 0.9706	Scientific Reports (2021)
Prostate Adenocarcinoma	Classification on Needle Biopsy	Not specified	Not specified	High performance reported	MDPI (2022)
Suspected Urothelial Carcinoma	Classification on LBC Urine Specimens	Not specified	Not specified	Successful development announced	Cancers (2023)

AUC: Area Under the Receiver Operating Characteristic Curve; EUS-FNB: Endoscopic Ultrasound-Guided Fine-Needle Biopsy; LBC: Liquid-Based Cytology.

Applications in Drug Development

Digital pathology is a transformative tool in preclinical and clinical drug development, offering quantitative, reproducible, and efficient analysis of tissue-based endpoints. While Medmain's primary focus has been on diagnostic support, their technology has direct applications in the pharmaceutical industry.

Potential Applications:

• Preclinical Toxicology Studies: Automated and standardized analysis of tissue slides from toxicology studies can improve the accuracy and efficiency of identifying and grading drug-induced tissue injuries.

• Clinical Trial Pathology: Centralized digital pathology platforms like PidPort™ can streamline the pathological review of clinical trial samples, enabling remote collaboration and standardized analysis across multiple trial sites.

Caption: Application of Medmain's technology in the drug development pipeline.

Conclusion

Medmain's whole-slide imaging and AI technology, centered around the PidPort™ platform, represent a significant advancement in the field of digital pathology. By leveraging large, well-annotated datasets and sophisticated deep learning models, they have demonstrated high performance in the classification and detection of various cancers. While specific technical details regarding their imaging hardware and detailed experimental protocols remain proprietary, the available data indicates a robust and scalable system. For researchers and professionals in drug development, this technology offers powerful tools to enhance preclinical and clinical research through quantitative, reproducible, and efficient tissue analysis. Further publications with more detailed methodologies and expanded applications are anticipated to provide deeper insights into their innovative approach.

References

• 1. Medmain inc. [en.medmain.com]
• 2. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 3. Medmain inc. [en.medmain.com]
• 4. Product Details | Medmain Inc. [en.medmain.com]
• 5. pidport.medmain.com [pidport.medmain.com]
• 6. blog.medmain.com [blog.medmain.com]

Medmain's Pioneering Role in Computational Pathology: A Technical Deep Dive

Fukuoka, Japan - Medmain Inc., a Japanese health-tech startup, is at the forefront of revolutionizing histopathology through the application of deep learning and artificial intelligence.[1] Facing a global shortage of pathologists and the increasing demand for rapid and accurate diagnostics, Medmain has developed "PidPort," a cloud-based digital pathology platform designed to support healthcare professionals and researchers.[1][2] This in-depth technical guide explores the core technologies, experimental validation, and workflows that underpin Medmain's contributions to the field of computational pathology, with a particular focus on its implications for researchers, scientists, and drug development professionals.

Core Technology: The PidPort Platform

The AI models are developed using a combination of deep learning techniques, including transfer learning and weakly supervised learning, and are trained on high-performance supercomputers.[7] These models are designed to screen for and classify various neoplastic and non-neoplastic lesions across a range of tissues.

Key AI Model Applications and Performance

Medmain has developed and validated AI screening models for a variety of histopathological and cytopathological evaluations. The company has published several peer-reviewed papers detailing the performance of these models.

Histopathology AI Models:

Medmain's histopathology AI is engineered to differentiate between malignant epithelial tumors, benign epithelial tumors, and non-neoplastic lesions in several key organs.[1][2]

Organ/Tissue	Target Lesions	Key Performance Metrics (ROC-AUC)
Stomach & Colon	Adenocarcinoma, Adenoma, Non-neoplastic	Not explicitly stated in snippets
Prostate	Adenocarcinoma vs. Benign (Needle Biopsy)	Up to 0.978
Prostate	Indolent vs. Aggressive Adenocarcinoma	Indolent: 0.846, Aggressive: 0.980
Breast	Ductal Carcinoma in Situ (DCIS), Invasive Ductal Carcinoma (IDC), Benign	DCIS: Up to 0.960, IDC: Up to 0.977
Stomach	Poorly Differentiated Adenocarcinoma (ESD specimens)	Up to 0.975
Stomach	Signet Ring Cell Carcinoma	At least 0.99 on all four test sets
Colorectal	Poorly Differentiated Adenocarcinoma	Approximately 0.95
Lung	Malignant Epithelial Tumor vs. Non-neoplastic	Not explicitly stated in snippets
Pancreas	Adenocarcinoma (EUS-FNA specimens)	Not explicitly stated in snippets

Cytopathology AI Models:

The company has also extended its AI capabilities to the analysis of cytological specimens.

Specimen Type	Target	Key Performance Metrics
Uterine Cervix	Neoplastic Cells	Not explicitly stated in snippets
Urine	Neoplastic Cells	Not explicitly stated in snippets

Experimental Protocols and Methodologies

Medmain's research publications provide insight into their rigorous experimental protocols for model development and validation. A generalized workflow can be summarized as follows:

1. Data Acquisition and Digitization:

• Histopathological slides are collected from multiple medical institutions through collaborative research agreements.

• These glass slides are digitized into whole-slide images (WSIs) using high-resolution scanners. Supported formats include SVS (Leica) and NDPI (Hamamatsu Photonics), as well as JPEG and TIFF.[3]

2. Annotation and Ground Truth Establishment:

• A large number of board-certified pathologists create "teacher data" by annotating the WSIs. This process involves delineating regions of interest and assigning diagnostic labels, which serve as the ground truth for training the AI models.

3. Model Training and Architecture:

• Medmain employs deep learning models, frequently utilizing the EfficientNet architecture (e.g., EfficientNetB1).

• Transfer Learning: A key strategy involves using models pre-trained on large, non-medical image datasets (e.g., ImageNet) and fine-tuning them on the histopathology data.[1] This approach is particularly effective when dealing with limited numbers of specific pathological cases. Medmain has explored a "partial fine-tuning" approach, where only the affine parameters of the batch normalization layers and the final classification layer are fine-tuned, leading to faster convergence.

• Weakly Supervised Learning: For some tasks, weakly supervised learning is employed, where the model is trained on slide-level labels without detailed annotations of specific regions. This is particularly useful for large-scale screening tasks.

• Fully Supervised Learning: In other cases, fully supervised learning is used, leveraging the detailed annotations provided by pathologists.

4. Model Validation:

• The trained models are rigorously evaluated on independent test sets, often sourced from different medical institutions than the training data to ensure generalizability.

• Performance is measured using standard metrics such as the Receiver Operating Characteristic Area Under the Curve (ROC-AUC), which assesses the model's ability to distinguish between different classes.

Visualizing Medmain's Workflows

To better illustrate the processes involved in Medmain's computational pathology ecosystem, the following diagrams are provided.

References

• 1. Medmain inc. [en.medmain.com]
• 2. Medmain Inc. Set to Begin Hospital Testing Trials for PidPort [prnewswire.com]
• 3. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 4. pidport.medmain.com [pidport.medmain.com]
• 5. Medmain inc. [en.medmain.com]
• 6. blog.medmain.com [blog.medmain.com]
• 7. blog.medmain.com [blog.medmain.com]

Medmain's AI in Histopathology: A Technical Deep Dive

Fukuoka, Japan - Medmain Inc., a prominent player in the health technology sector, is advancing the field of pathology with its artificial intelligence (AI)-driven solutions. At the core of their innovation is the "PidPort" digital pathology platform, which leverages deep learning to assist pathologists in the analysis of histopathological images. This technical guide synthesizes available data from Medmain's publications to provide researchers, scientists, and drug development professionals with an in-depth understanding of their methodologies and the performance of their AI models.

Core Technology: Deep Learning on Whole Slide Images

Medmain's approach centers on the application of deep learning, primarily using Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), to analyze whole slide images (WSIs) of tissue specimens.[1] Their process involves a comprehensive workflow that begins with the digitization of glass slides to create high-resolution WSIs. These images are then used to train and validate their AI models. The company has developed a suite of AI models for various tissues, including the stomach, colon, breast, lung, and pancreas, as well as for cytological analysis of the uterine cervix and urine.[2][3][4][5][6]

A key aspect of their methodology is the use of "weakly supervised learning." This approach allows the AI to learn from WSIs that are labeled at the image level (e.g., adenocarcinoma present) rather than requiring meticulous, pixel-level annotations by pathologists. This significantly accelerates the training process and reduces the annotation burden on specialists.[7][8] Transfer learning is another integral technique, where models pre-trained on large, general image datasets are fine-tuned for specific histopathological tasks.[3][9]

Experimental Workflow and Protocols

Medmain collaborates with multiple medical institutions to amass large datasets of hundreds of thousands of WSIs for training and validation.[3][5][10] The general experimental workflow can be summarized as follows:

• Data Acquisition and Digitization: Histopathological glass slides are digitized to create WSIs.

• Dataset Curation and Annotation: Pathologists provide diagnostic labels for the WSIs, which serve as the ground truth for training the AI models.

• Model Training: Deep learning models, often based on established architectures, are trained on a large, annotated dataset. Techniques like transfer learning and weakly supervised learning are employed.

• Model Validation: The trained models are rigorously evaluated on independent test sets from different medical institutions to ensure their generalizability and robustness.[1][11]

The following diagram illustrates this general experimental workflow:

General Experimental Workflow for Medmain's AI Model Development.

Performance of AI Models

Medmain has published the performance of its AI models for several types of cancer in peer-reviewed journals. The primary metric reported is the Area Under the Receiver Operating Characteristic Curve (AUC), which measures the model's ability to distinguish between classes.

Gastric and Colonic Epithelial Tumors

In a study published in Scientific Reports, Medmain detailed the performance of their CNN and RNN-based models for classifying gastric and colonic epithelial tumors into adenocarcinoma, adenoma, and non-neoplastic categories. The models were evaluated on three independent test sets.[1]

Organ	Classification	Highest Achieved AUC
Gastric	Adenocarcinoma vs. All	0.97
Gastric	Adenoma vs. All	0.99
Colon	Adenocarcinoma vs. All	0.96
Colon	Adenoma vs. All	0.99

Gastric Diffuse-Type and Poorly Differentiated Adenocarcinoma

Further research has focused on specific, challenging subtypes of gastric cancer. For diffuse-type adenocarcinoma, models achieved AUCs between 0.95 and 0.99 on five distinct test sets, demonstrating high generalizability.[12] For poorly differentiated adenocarcinoma, a model trained on endoscopic submucosal dissection WSIs reached an ROC-AUC of up to 0.975.[8][13]

The logical flow for the classification of gastric epithelial tumors is depicted below:

Logical flow for AI-based classification of gastric epithelial tumors.

Applications in Drug Development

The ability of AI to provide rapid, accurate, and reproducible analysis of histopathological images has significant implications for drug development. By automating the assessment of tissue samples, Medmain's technology can potentially accelerate preclinical studies and clinical trials. For instance, AI can be used for high-throughput screening of tissue responses to novel therapeutics, identification of biomarkers, and patient stratification. While Medmain's publications primarily focus on diagnostic applications, the underlying technology is directly applicable to the research and development needs of the pharmaceutical industry. The platform's capacity to handle large volumes of data and provide quantitative insights can enhance the efficiency and objectivity of pathological evaluations in a drug development context.

The PidPort Platform

The relationship between the user, the PidPort platform, and the AI analysis is illustrated in the following diagram:

Ecosystem of the PidPort platform for AI-driven histopathology.

Conclusion

Medmain's approach to AI in histopathology, centered around its PidPort platform and validated deep learning models, represents a significant step forward in computational pathology. By leveraging large datasets and advanced machine learning techniques, the company has demonstrated high accuracy in the classification of various cancers. For researchers and professionals in drug development, this technology offers a powerful tool to enhance the speed, scale, and objectivity of tissue analysis, ultimately contributing to the acceleration of therapeutic innovation. As Medmain continues to expand its portfolio of AI models and publish further validation studies, the impact of its technology on both clinical practice and pharmaceutical research is expected to grow.

References

• 1. Deep Learning Models for Histopathological Classification of Gastric and Colonic Epithelial Tumours - PubMed [pubmed.ncbi.nlm.nih.gov]
• 2. Medmain inc. [en.medmain.com]
• 3. Product Details | Medmain Inc. [en.medmain.com]
• 4. Medmain inc. [en.medmain.com]
• 5. 製品情報 | メドメイン株式会社 [medmain.com]
• 6. pidport.medmain.com [pidport.medmain.com]
• 7. blog.medmain.com [blog.medmain.com]
• 8. Medmain inc. [en.medmain.com]
• 9. Deep learning approaches for pathological image classification - PubMed [pubmed.ncbi.nlm.nih.gov]
• 10. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 11. blog.medmain.com [blog.medmain.com]
• 12. Medmain inc. [en.medmain.com]
• 13. Medmain inc. [en.medmain.com]
• 14. Medmain inc. [en.medmain.com]
• 15. PidPort｜Digital Pathology AI Analysis Solution [us.medmain.com]

Medmain's Pioneering Research in AI-Powered Digital Pathology: A Technical Overview

Fukuoka, Japan - Medmain Inc., a leading Japanese health-tech startup, is at the forefront of revolutionizing pathological diagnosis through its advanced artificial intelligence (AI) and deep learning technologies. This technical guide provides an in-depth overview of Medmain's research and development, focusing on their methodologies for creating and validating AI models for the histopathological classification of various cancers. This document is intended for researchers, scientists, and drug development professionals seeking a comprehensive understanding of Medmain's innovative approach to digital pathology.

Core Technology: The "PidPort" Digital Pathology Platform

Research Focus and Key Publications

Medmain's research is centered on developing highly accurate AI models for the classification of various epithelial tumors. Their work has been published in several peer-reviewed scientific journals, demonstrating the efficacy of their deep learning approaches. Key areas of research include the analysis of gastric, colonic, urothelial, prostate, and skin cancers.

Quantitative Performance of Medmain's AI Models

Medmain's publications report high performance metrics for their AI models across a range of cancer types. The following tables summarize the key quantitative data from their studies, primarily focusing on the Area Under the Receiver Operating Characteristic Curve (AUC), a common metric for evaluating the performance of binary classification models.

Study Focus	Tissue Type	Classification Task	Model Architecture	Key Performance Metric (AUC)
Gastric & Colonic Epithelial Tumors	Stomach, Colon	Adenocarcinoma vs. Adenoma vs. Non-neoplastic	InceptionV3 (CNN) & RNN	Gastric Adenocarcinoma: 0.97, Gastric Adenoma: 0.99, Colonic Adenocarcinoma: 0.96, Colonic Adenoma: 0.99
Gastric Diffuse-type Adenocarcinoma	Stomach	Diffuse-type Adenocarcinoma vs. Other	Not Specified	Not Specified in available abstracts
Urothelial Carcinoma	Urine Cytology	Neoplastic vs. Non-neoplastic	Not Specified	0.984 - 0.990
Prostate Cancer	Prostate	Adenocarcinoma vs. Benign	EfficientNetB1	Up to 0.984
Malignant Melanoma	Skin	Malignant Melanoma vs. Other	Not Specified	Not Specified in available abstracts
Poorly Differentiated Gastric Adenocarcinoma	Stomach	Poorly Differentiated Adenocarcinoma vs. Other	Not Specified	Up to 0.975

Experimental Protocols and Methodologies

Medmain employs a rigorous and standardized workflow for the development and validation of their deep learning models. This process involves several key stages, from data acquisition and preparation to model training and evaluation.

Data Acquisition and Digitization

• Specimen Collection: Histopathological slides are collected from multiple medical institutions through collaborative research agreements.

• Whole-Slide Imaging (WSI): Glass slides are digitized at high resolution using whole-slide scanners to create gigapixel-sized digital images. This process is managed by the "Medmain Imaging Center".[1]

Data Annotation and Preparation

• Pathologist Annotation: Board-certified pathologists meticulously annotate the WSIs, delineating regions of interest (ROIs) corresponding to different pathological findings (e.g., adenocarcinoma, adenoma, non-neoplastic tissue).

• Tiling: Due to the large size of WSIs, they are divided into smaller, manageable image patches or "tiles." These tiles serve as the input for the deep learning models.

• Data Augmentation: To increase the diversity of the training data and prevent model overfitting, various data augmentation techniques are applied to the tiles. These techniques include:

  • Geometric Transformations: Rotation, flipping, and scaling.

  • Color Augmentation: Adjustments to brightness, contrast, and saturation to account for variations in staining.

Deep Learning Model Training

• Model Architectures: Medmain primarily utilizes Convolutional Neural Networks (CNNs), which are well-suited for image analysis tasks. Specific architectures mentioned in their publications include:

  • InceptionV3: A deep CNN architecture known for its efficiency and performance.

  • EfficientNet: A family of models that systematically scale network width, depth, and resolution to achieve high accuracy with fewer parameters.

  • Recurrent Neural Networks (RNNs): Used in conjunction with CNNs to analyze sequences of image patches, capturing contextual information across a WSI.

• Training Process: The models are trained on large, annotated datasets of image tiles. The training process involves iteratively adjusting the model's parameters to minimize the difference between its predictions and the ground-truth labels provided by pathologists.

• Weakly Supervised Learning: In some cases, Medmain employs weakly supervised learning approaches. This means the model is trained on WSIs with slide-level labels (e.g., "cancer" or "no cancer") without detailed annotations of cancerous regions. The model learns to identify the relevant features within the WSI that correspond to the slide-level label.

Model Evaluation and Validation

• Independent Test Sets: The performance of the trained models is evaluated on independent test sets of WSIs that were not used during the training process. This ensures an unbiased assessment of the model's generalization capabilities.

• Performance Metrics: Key performance metrics include:

  • Area Under the Curve (AUC): A measure of the model's ability to distinguish between different classes.

  • Sensitivity and Specificity: Measures of the model's ability to correctly identify positive and negative cases, respectively.

  • Accuracy: The overall proportion of correct classifications.

• Heatmap Generation: The models generate heatmaps that visualize the probability of malignancy across the entire WSI, providing an intuitive visual aid for pathologists.

Visualization of Workflows and Concepts

To further elucidate Medmain's methodologies, the following diagrams, generated using the DOT language, illustrate key workflows and relationships.

Conclusion

Medmain's research and development efforts represent a significant advancement in the field of digital pathology. Through the rigorous application of deep learning methodologies, the company has developed a suite of powerful AI models capable of accurately classifying various types of cancer from histopathological images. The integration of these models into the "PidPort" platform provides a valuable tool for pathologists, with the potential to enhance diagnostic accuracy, improve efficiency, and ultimately, advance patient care. As Medmain continues to expand its research to other organs and disease types, the impact of its technology on the future of pathology is poised to grow.

References

• 1. Medmain inc. [en.medmain.com]
• 2. Medmain inc. [en.medmain.com]

The Impact of Medmain on Telepathology and Consultations: A Technical Guide

This technical guide provides an in-depth analysis of Medmain's core technologies, focusing on the quantitative performance of its AI models, the experimental protocols employed in their development, and the logical workflows that underpin the PidPort platform.

Quantitative Performance of Medmain's AI Models

Medmain has developed a suite of deep learning models for the histopathological classification of various epithelial tumors. The performance of these models has been rigorously validated and published in several peer-reviewed scientific journals. The following tables summarize the key performance metrics of their AI models for different cancer types.

Table 1: Performance in Gastric and Colonic Epithelial Tumors

Tissue	Classification Task	Test Set	Accuracy	Sensitivity	Specificity	AUC
Stomach	Adenocarcinoma vs. Adenoma vs. Non-neoplastic	Internal	0.97	0.96	0.98	0.99
External 1	0.92	0.91	0.93	0.97
External 2	0.89	0.88	0.90	0.95
Colon	Adenocarcinoma vs. Adenoma vs. Non-neoplastic	Internal	0.98	0.97	0.99	0.99
External 1	0.94	0.93	0.95	0.98
External 2	0.91	0.90	0.92	0.97

Data sourced from a study published in Scientific Reports.

Table 2: Performance in Pancreatic Adenocarcinoma Detection

Specimen Type	Classification Task	Accuracy	Sensitivity	Specificity
EUS-FNB	Adenocarcinoma Detection	0.9417	0.9302	0.9706

EUS-FNB: Endoscopic Ultrasound-Guided Fine-Needle Aspiration Biopsy. Data indicates the model's ability to accurately detect challenging cases with isolated and low-volume cancer cells.[4]

Table 3: Performance in Indeterminate Lung Carcinoma Classification

Specimen Type	Classification Task	Test Set	AUC
TBLB	ADC vs. SCC vs. SCLC vs. Non-neoplastic	Independent (Indeterminate Cases)	0.99
Independent (TBLB)	0.94
Independent (Surgical 1)	0.95
Independent (Surgical 2)	0.96
Independent (Surgical 3)	0.97

TBLB: Transbronchial Lung Biopsy; ADC: Adenocarcinoma; SCC: Squamous Cell Carcinoma; SCLC: Small Cell Lung Carcinoma. The model demonstrates high performance on challenging, indeterminate cases.[4][5]

Experimental Protocols

Medmain employs a rigorous and systematic approach to the development and validation of its deep learning models. The core methodology revolves around the use of convolutional neural networks (CNNs) and, in some cases, recurrent neural networks (RNNs), trained on extensive datasets of whole-slide images (WSIs).[4]

Data Acquisition and Preparation

• Collaborative Data Sourcing : Medmain collaborates with multiple medical institutions to source hundreds of thousands of histopathological glass slides.[2]

• Digitization : These glass slides are digitized into high-resolution WSIs.

• Annotation and Teacher Data Creation : A large team of highly specialized pathologists utilizes Medmain's proprietary annotation tools to create detailed ground-truth data for model training.[2] This process involves delineating regions of interest (e.g., tumors, benign lesions) on the WSIs.

Deep Learning Model Training

• Network Architecture : The primary architectures employed are CNNs, which are well-suited for image recognition tasks. For certain applications, RNNs are also utilized to capture sequential information within the WSIs.[4]

• Training Methodology : Medmain leverages both fully-supervised and weakly-supervised learning approaches.[2]

  • Fully-Supervised Learning : The models are trained on meticulously annotated data where specific regions are labeled.

  • Weakly-Supervised Learning : The models are trained using slide-level labels, which is a more scalable approach as it does not require detailed, pixel-level annotations. This method is particularly useful for large-scale analysis of WSIs.

• Transfer Learning : To accelerate the development and improve the performance of their AI models, Medmain utilizes transfer learning. This technique involves taking a pre-trained model (trained on a large, general-purpose image dataset) and fine-tuning it on the specific task of histopathological image analysis.

Model Validation

• Internal Validation : The trained models are first validated on an internal test set, which is a portion of the initial dataset that was held out from the training process.

• External Validation : To ensure the generalizability and robustness of the models, they are further evaluated on multiple independent test sets from different medical institutions. This process helps to confirm that the model's performance is not specific to the data it was trained on.

Visualized Workflows and Relationships

The following diagrams illustrate the core logical and experimental workflows within the Medmain ecosystem.

References

• 1. 研究論文一覧｜PidPort メドメイン株式会社 [service.medmain.com]
• 2. blog.medmain.com [blog.medmain.com]
• 3. Medmain inc. [en.medmain.com]
• 4. Product Details | Medmain Inc. [en.medmain.com]
• 5. 製品情報 | メドメイン株式会社 [medmain.com]

Methodological & Application

Application Notes and Protocols for Utilizing PidPort in Oncology Research

For Researchers, Scientists, and Drug Development Professionals

Introduction to PidPort in Oncology Research

PidPort is a cloud-based digital pathology solution designed to support pathologists and researchers by providing a suite of tools for efficient and collaborative analysis of pathological images.[4] Its core functionalities relevant to oncology research include:

• Centralized Cloud Storage: Securely store, manage, and access large volumes of digital slide data from anywhere, eliminating the logistical challenges of physical slide management.[2][3]

• High-Speed Whole Slide Image Viewer: A user-friendly interface for smooth and detailed examination of histopathology slides.[3]

• Collaborative Tools: Features for commenting, annotating, and sharing slides facilitate seamless collaboration among research teams and with external partners.[3]

Key Applications in Oncology Research

PidPort can be instrumental in various stages of oncology research, from basic research to clinical trials.

• Quantitative Biomarker Analysis: The platform's AI capabilities can be trained to quantify the expression of protein biomarkers (e.g., from immunohistochemistry-stained slides) and other histological features. This quantitative data is crucial for understanding tumor biology and stratifying patient populations.[7][8][9]

• Tumor Microenvironment (TME) Analysis: Researchers can utilize PidPort to analyze the spatial relationships between different cell types within the TME, providing insights into immune infiltration and other critical aspects of cancer progression.

• High-Throughput Screening: The digital nature of the platform allows for the efficient analysis of large numbers of slides, making it suitable for screening studies in drug discovery and development.

• Standardization and Reproducibility: By automating parts of the analysis process, PidPort can help reduce inter-observer variability and improve the reproducibility of research findings.[10]

Experimental Protocols

While specific protocols may vary depending on the research question and sample type, the following provides a general framework for using PidPort in an oncology research project.

Protocol 1: Digitization and Upload of Histopathology Slides

This protocol outlines the initial step of converting glass slides into a digital format compatible with PidPort.

Objective: To create high-resolution whole slide images (WSI) for analysis in PidPort.

Materials:

• Glass slides with stained tissue sections (e.g., H&E, IHC)

• A whole slide scanner (compatible formats include JPEG, TIFF, SVS (Leica), and NDPI (Hamamatsu))[11]

• A computer with a stable internet connection

• PidPort account credentials

Procedure:

• Slide Preparation: Ensure slides are clean and properly labeled.

• Scanning:

  • Follow the manufacturer's instructions for the whole slide scanner.

  • Select the appropriate scanning resolution (e.g., 20x or 40x magnification) based on the research requirements.

  • Save the scanned images in a compatible format.[11]

• Logging into PidPort: Access the PidPort platform via a web browser and log in with your credentials.

• Uploading WSI:

  • Navigate to the upload section of the platform.

  • Select the WSI files to be uploaded. PidPort offers a bulk upload function for efficiency.[3]

  • Enter relevant case information and metadata for each slide to ensure proper organization and searchability.

Protocol 2: AI-Assisted Tumor Detection and Classification (Example: Prostate Cancer)

This protocol describes the use of PidPort's AI capabilities for the analysis of prostate cancer specimens. Medmain has developed and published research on an AI model for classifying prostate cancer in transurethral resection and core needle biopsy specimens.[5][12][13][14]

Objective: To utilize PidPort's AI to identify and classify cancerous regions in prostate tissue WSI.

Materials:

• Digitized WSI of prostate tissue uploaded to PidPort.

• PidPort's prostate cancer analysis AI module.

Procedure:

• Select WSI: From your PidPort library, select the prostate tissue WSI you wish to analyze.

• Initiate AI Analysis:

  • Access the AI analysis tools within the PidPort interface.

  • Select the appropriate AI model for prostate cancer analysis.

• Review AI Results:

  • The AI will process the WSI and highlight regions suspected of being cancerous.

  • The platform may provide a classification of the tumor grade (e.g., based on the Gleason grading system).

• Annotation and Collaboration:

  • Use PidPort's annotation tools to add your own comments, measurements, or drawings to the WSI.[3]

  • Share the analyzed slides with collaborators for second opinions or further discussion using the platform's sharing features.

Quantitative Data Presentation

The quantitative outputs from PidPort's AI analysis can be summarized in tables for easy comparison and further statistical analysis.

Case ID	WSI File Name	Total Tissue Area (mm²)	Tumor Area (mm²)	% Tumor Area	AI-Predicted Gleason Grade
PCa-001	slide_001.svs	15.2	4.8	31.6%	3+4=7
PCa-002	slide_002.svs	12.5	2.1	16.8%	3+3=6
PCa-003	slide_003.svs	18.9	7.3	38.6%	4+5=9

Visualization of Workflows and Signaling Pathways

Experimental Workflow

The following diagram illustrates a typical workflow for using PidPort in oncology research, from sample preparation to data analysis.

Caption: General workflow for oncology research using the PidPort platform.

Linking Quantitative Data to Signaling Pathway Analysis

PidPort's primary function is not direct signaling pathway analysis. However, the quantitative data it generates on biomarker expression can be a critical input for such analyses. For example, quantifying the expression of key proteins in a specific cancer-related pathway can help researchers understand its activation state.

The following diagram illustrates this logical relationship.

Caption: Linking PidPort's quantitative output to signaling pathway analysis.

Conclusion

PidPort offers a powerful and versatile platform for oncology researchers, streamlining workflows, enabling robust quantitative analysis, and fostering collaboration. By integrating digital pathology and AI into the research pipeline, scientists and drug development professionals can accelerate the pace of discovery and contribute to the advancement of personalized medicine in oncology. For more detailed information on specific AI models and their performance, researchers are encouraged to consult the peer-reviewed publications by Medmain Inc.[1][5]

References

• 1. blog.medmain.com [blog.medmain.com]
• 2. Medmain inc. [en.medmain.com]
• 3. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 4. pidport.medmain.com [pidport.medmain.com]
• 5. Medmain inc. [en.medmain.com]
• 6. mdpi.com [mdpi.com]
• 7. blog.crownbio.com [blog.crownbio.com]
• 8. Quantitative Image Analysis for Tissue Biomarker Use: A White Paper From the Digital Pathology Association - PMC [pmc.ncbi.nlm.nih.gov]
• 9. sonraianalytics.com [sonraianalytics.com]
• 10. The use of digital pathology and image analysis in clinical trials - PMC [pmc.ncbi.nlm.nih.gov]
• 11. pidport.medmain.com [pidport.medmain.com]
• 12. mdpi.com [mdpi.com]
• 13. Artificial Intelligence for Clinical Diagnosis and Treatment of Prostate Cancer [mdpi.com]
• 14. mdpi.com [mdpi.com]

Application Notes & Protocols for PidPort in Collaborative Research

Audience: Researchers, scientists, and drug development professionals specializing in pathology, oncology, and computational medicine.

Application Notes

Introduction: PidPort is a cloud-based digital pathology system designed for the storage, viewing, sharing, and analysis of pathological images.[1] Its integrated features, including a high-speed viewer and AI-powered analytical tools, make it an ideal platform for facilitating collaborative research projects among geographically dispersed teams.[1][2] These notes outline the application of PidPort in a multi-institutional collaborative research project aimed at validating a novel AI algorithm for identifying malignant melanoma in dermatopathology histology specimens.[3]

Core Applications in Collaborative Research:

• Centralized & Secure Data Repository: PidPort serves as the central repository for all whole slide images (WSI) and associated case data for the project. This eliminates the need for physical shipment of glass slides, reducing costs and preventing potential damage or loss.[2] The platform's cloud-based infrastructure ensures secure access for authorized researchers from any location.[1]

• Remote Consultation and Annotation: Researchers from different institutions can simultaneously view and annotate WSIs.[2] PidPort’s comment and drawing functions allow for interactive discussions directly on the images, facilitating consensus-building on diagnoses and feature identification.[2] This is particularly useful for training and validating AI models, where precise annotations are critical.

• Standardized Viewing Environment: The platform provides a consistent, high-quality viewing experience for all collaborators, which is crucial for reducing inter-observer variability in pathological assessment.[1] This ensures that all researchers are analyzing the images under the same conditions, leading to more reliable and reproducible results.

• Efficient Project Management and Workflow: The platform can streamline the research workflow by providing features for organizing cases, tracking progress, and managing user access.[2] This helps to ensure that the project stays on schedule and that all collaborators are working with the most up-to-date information.

Benefits in a Multi-Institutional Setting:

• Accelerated Timelines: By removing the logistical barriers of sharing physical slides, PidPort can significantly reduce the time required for data collection and review in a multi-center study.[3]

• Reduced Costs: The elimination of slide shipping and travel for in-person review meetings results in substantial cost savings.[2][4]

• Enhanced Collaboration and Knowledge Sharing: The platform fosters a more dynamic and interactive collaborative environment, allowing for real-time knowledge exchange and problem-solving.[4]

• Improved Diagnostic Accuracy and Consensus: By facilitating easy access to second opinions and expert consultations, PidPort can help to improve the accuracy and consistency of pathological diagnoses.[3]

Quantitative Data Summary

The following table summarizes the projected efficiencies and improvements for a multi-institutional study using PidPort compared to traditional methods involving the physical shipment of slides.

Metric	Traditional Method (Physical Slides)	Using PidPort	Projected Improvement
Average Time for Multi-Site Slide Review	2-3 weeks per batch	1-2 days per batch	>90% reduction
Cost per Case for Shipping & Handling	$50 - $100	$0	100% reduction
Travel Costs for Consensus Meetings	$5,000 - $10,000 per meeting	$0 (Virtual Meetings)	100% reduction
Risk of Slide Damage/Loss	Low to Moderate	Negligible	>99% reduction
Time to Reach Diagnostic Consensus	Days to Weeks	Hours to Days	75-80% reduction

Experimental Protocols

1.0 Objective: To validate the performance of a novel AI algorithm for the detection of malignant melanoma on digitized dermatopathology slides by comparing its output to the consensus diagnosis of a panel of expert pathologists from three independent institutions.

2.0 Materials & Platform:

• Whole Slide Scanner

• PidPort Cloud Platform Account

• Web Browser (PC or Tablet)

• Anonymized dermatopathology slides (n=500) from three participating institutions (A, B, C)

3.0 Methodology:

3.1 Slide Preparation and Digitization:

• Each institution will select a cohort of anonymized dermatopathology cases with a known diagnosis of malignant melanoma or benign nevi.

• Glass slides will be digitized at 40x magnification using a calibrated whole slide scanner to create high-resolution WSIs.

• A standardized naming convention will be used for all WSI files to ensure proper tracking (e.g., InstA_Case001_Melanoma.svs).

3.2 Data Upload and Organization in PidPort:

• The research coordinator will create a new project folder in PidPort titled "Melanoma AI Validation Study."

• Subfolders will be created for each institution (Institution_A, Institution_B, Institution_C).

• Each institution's designated data manager will use the bulk upload function to upload their WSIs into their respective folders.[2]

• Relevant clinical data (e.g., age, sex, lesion location) will be attached as a PDF to each case file within PidPort.[2]

3.3 Pathologist Review and Annotation (Phase 1):

• A panel of three expert pathologists (one from each institution) will be granted access to the project folder.

• Each pathologist will independently review all 500 WSIs using the PidPort high-speed web viewer.[2]

• For each case, the pathologist will provide a diagnosis (Malignant Melanoma or Benign) using the comment function.

• Any disagreements in diagnosis will be flagged for a virtual consensus meeting.

3.4 Consensus Meeting Protocol:

• The research coordinator will schedule a video conference call.

• During the meeting, the flagged cases will be reviewed in real-time by the panel using PidPort's screen sharing capabilities.

• The drawing and annotation tools will be used to highlight specific features of interest to reach a consensus diagnosis.[2]

• The final consensus diagnosis for each case will be recorded in the comment section of the respective PidPort file.

3.5 AI Model Analysis (Phase 2):

• The validated AI melanoma detection model will be run on all 500 WSIs within the PidPort platform.

• The AI will generate a classification for each case (Malignant Melanoma or Benign) and a confidence score.

• The AI's output will be automatically logged for each case.

3.6 Data Analysis and Validation:

• The following performance metrics for the AI model will be calculated:

  • Sensitivity

  • Specificity

  • Positive Predictive Value (PPV)

  • Negative Predictive Value (NPV)

  • Overall Accuracy

• The results will be compiled into a final report for publication.

Visualizations

Caption: Collaborative workflow for AI model validation using PidPort.

Caption: Logical relationships between collaborators and PidPort.

References

• 1. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 2. pidport.medmain.com [pidport.medmain.com]
• 3. PidPort｜Digital Pathology AI Analysis Solution [us.medmain.com]
• 4. PidPort｜Digital Pathology AI Analysis Solution [us.medmain.com]

Revolutionizing Immunohistochemistry: Automated Quantitative Analysis with Medmain's PidPort™ Platform

Application Note & Protocol

Audience: Researchers, scientists, and drug development professionals.

Introduction

Immunohistochemistry (IHC) is a cornerstone technique in research and clinical diagnostics for visualizing the distribution and abundance of specific proteins within tissue samples. While traditional IHC provides qualitative and semi-quantitative data, the increasing demand for objective, reproducible, and high-throughput analysis has driven the adoption of digital pathology and automated image analysis. Medmain's PidPort™ is a cloud-based digital pathology platform that leverages artificial intelligence to provide powerful tools for the storage, management, sharing, and analysis of whole-slide images (WSI).[1][2][3][4] This document provides a detailed application note and protocol for the quantitative analysis of IHC staining using a specialized module within the PidPort™ platform, designed to streamline research and drug development workflows.

Application: Quantitative Analysis of PD-L1 Expression in Non-Small Cell Lung Carcinoma (NSCLC) Tissue

Programmed death-ligand 1 (PD-L1) is a critical immune checkpoint protein and a key biomarker for predicting the response to immunotherapy in various cancers, including NSCLC. Accurate and reproducible quantification of PD-L1 expression on tumor cells (TCs) and tumor-infiltrating immune cells (ICs) is crucial for patient stratification and drug development. This application note describes the use of the Medmain PidPort™ IHC Quantification Module to automate the analysis of PD-L1 IHC slides.

Key Features of the PidPort™ IHC Quantification Module:

• Customizable Staining Thresholds: User-defined parameters for classifying staining intensity as negative, low, medium, or high.

• Comprehensive Data Output: Generates detailed quantitative data, including H-score, percentage of positive cells, and cell density.

• Batch Analysis: High-throughput analysis of multiple slides simultaneously.

• Secure Cloud-Based Platform: Enables easy data access, sharing, and collaboration.[3][4]

Experimental Workflow

The overall workflow for IHC quantification using the PidPort™ platform involves several key stages, from sample preparation to data analysis and interpretation.

Caption: General workflow for IHC quantification using the PidPort™ platform.

Protocol for PD-L1 IHC Staining and Quantification

This protocol provides a generalized procedure for the immunohistochemical staining of PD-L1 in formalin-fixed, paraffin-embedded (FFPE) NSCLC tissue sections and subsequent analysis using the PidPort™ platform.

Part 1: Immunohistochemistry Staining Protocol

Materials:

• FFPE NSCLC tissue sections (4-5 µm) on charged slides

• Xylene and Ethanol (B145695) series for deparaffinization and rehydration

• Antigen Retrieval Solution (e.g., Citrate Buffer, pH 6.0)

• Peroxidase Block

• Protein Block (e.g., Normal Goat Serum)

• Primary Antibody: Rabbit Anti-PD-L1 Monoclonal Antibody

• HRP-conjugated Secondary Antibody (Anti-Rabbit)

• DAB Chromogen Kit

• Hematoxylin Counterstain

• Mounting Medium

Procedure:

• Deparaffinization and Rehydration:

  • Incubate slides in Xylene (2 x 5 minutes).

  • Rehydrate through a graded series of ethanol (100%, 95%, 70%) for 3 minutes each.

  • Rinse with distilled water.

• Antigen Retrieval:

  • Perform heat-induced epitope retrieval (HIER) using a steamer or water bath at 95-100°C for 20-30 minutes in Antigen Retrieval Solution.

  • Allow slides to cool to room temperature.

• Staining:

  • Incubate with Peroxidase Block for 10 minutes to quench endogenous peroxidase activity.

  • Rinse with wash buffer (e.g., PBS).

  • Apply Protein Block and incubate for 20 minutes.

  • Incubate with the primary PD-L1 antibody at the recommended dilution for 60 minutes at room temperature.

  • Rinse with wash buffer.

  • Apply the HRP-conjugated secondary antibody and incubate for 30 minutes.

  • Rinse with wash buffer.

  • Apply DAB chromogen and incubate until the desired brown stain intensity develops.

  • Rinse with distilled water.

• Counterstaining and Mounting:

  • Counterstain with Hematoxylin for 1-2 minutes.

  • Rinse with water.

  • Dehydrate through a graded ethanol series and clear with xylene.

  • Coverslip with a permanent mounting medium.

Part 2: Digital Slide Scanning and Analysis with PidPort™

Procedure:

• Whole-Slide Scanning:

  • Scan the stained slides at 20x or 40x magnification using a whole-slide scanner to generate digital WSI files.

• Image Upload and Project Creation:

  • Log in to the PidPort™ platform.[1]

  • Create a new project and upload the WSI files.

• Running the IHC Quantification Module:

  • Select the uploaded WSI for analysis.

  • Launch the "IHC Quantification Module".

  • Define Regions of Interest (ROIs): Use the annotation tools to delineate the tumor areas to be analyzed. Multiple ROIs can be selected per slide.

  • Set Analysis Parameters:

    • Stain: Select "DAB (Brown)" and "Hematoxylin (Blue)".

    • Cell Type: Select "Tumor Cells" and "Immune Cells" for separate quantification. The AI will assist in differentiating these based on morphology.

    • Staining Thresholds: Define the intensity thresholds for Negative, Low, Medium, and High staining. The platform will provide a default setting that can be adjusted by the user based on a visual review of the staining.

• Data Generation and Review:

  • Initiate the analysis. The platform will perform cell segmentation and classification based on the defined parameters.

  • Upon completion, the results will be displayed, including markup images showing the classification of each cell and a summary data table.

Data Presentation

The PidPort™ IHC Quantification Module provides a comprehensive quantitative data summary for each analyzed ROI. The data can be exported in various formats (e.g., CSV, Excel) for further statistical analysis.

Table 1: Example Quantitative Data for PD-L1 Expression in Tumor Cells

Sample ID	ROI Area (mm²)	Total Tumor Cells	PD-L1 Positive Cells (%)	H-Score
NSCLC-001	2.5	15,234	75.6	180
NSCLC-002	3.1	18,987	15.2	45
NSCLC-003	2.8	16,543	90.1	250
Control-001	2.9	17,890	1.2	5

H-Score is calculated as: H-Score = (% of cells with low intensity * 1) + (% of cells with medium intensity * 2) + (% of cells with high intensity * 3)

Table 2: Example Quantitative Data for PD-L1 Expression in Immune Cells

Sample ID	Total Immune Cells	PD-L1 Positive ICs (%)
NSCLC-001	3,456	45.8
NSCLC-002	4,123	5.6
NSCLC-003	2,987	65.2
Control-001	3,876	0.5

Signaling Pathway Diagram

The PD-1/PD-L1 signaling pathway is a key target for cancer immunotherapy. Understanding this pathway is essential for interpreting the results of PD-L1 IHC quantification.

Caption: The PD-1/PD-L1 immune checkpoint signaling pathway.

Conclusion

References

• 1. Medmain inc. [en.medmain.com]
• 2. Medmain inc. [en.medmain.com]
• 3. pidport.medmain.com [pidport.medmain.com]
• 4. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 5. Product Details | Medmain Inc. [en.medmain.com]

Revolutionizing Research: A Workflow for Medmain's AI-Powered Digital Pathology Platform

Application Notes and Protocols for Researchers, Scientists, and Drug Development Professionals

Introduction to Medmain and PidPort

Medmain is a leading health technology company dedicated to advancing pathological diagnosis through the power of artificial intelligence.[1] Their flagship product, PidPort, is a cloud-based digital pathology support system that offers a comprehensive solution for managing, viewing, sharing, and analyzing whole-slide images (WSIs).[4][5] PidPort's integrated AI analytics provide researchers with highly accurate and instantaneous analysis of tissues and cells, accelerating research and discovery.[4] The platform is designed to reduce the workload of pathologists and researchers by offering a centralized and efficient system for managing large-scale pathological data.[5]

The Medmain Research Workflow

A high-level overview of the research workflow using the Medmain PidPort platform.

Key Experiments and Protocols

Medmain's AI models have been validated in several research studies, demonstrating their high performance in classifying various types of cancer. Below are detailed protocols for two key experimental applications.

Experiment 1: Classification of Gastric Adenocarcinoma

This protocol is based on the methodology used in the development and validation of Medmain's deep learning model for classifying gastric diffuse-type adenocarcinoma.

Objective: To accurately classify whole-slide images of gastric biopsies into diffuse-type adenocarcinoma or non-neoplastic tissue.

Experimental Protocol:

• Dataset Preparation:

  • Collect a large dataset of formalin-fixed paraffin-embedded (FFPE) gastric biopsy specimens.

  • Digitize the glass slides at high resolution (e.g., 20x magnification) to create WSIs.

  • Anonymize all patient data associated with the slides.

  • Obtain ethical approval from the relevant institutional review boards.[6][7][8][9][10]

• Image Preprocessing and Tiling:

  • Upload the WSIs to the PidPort platform.

  • Utilize a tiling approach to break down the large WSIs into smaller, manageable image patches (e.g., 512x512 pixels).

  • Apply stain normalization techniques to account for variations in staining intensity across different slides.

• AI Model Training and Validation:

  • Employ a deep learning model, such as a convolutional neural network (CNN) like EfficientNetB1, pre-trained on a large image dataset (e.g., ImageNet).[8]

  • Use a partial fine-tuning approach where only the affine parameters of the batch normalization layers and the final classification layer are trained.[8]

  • Train the model on a dedicated training set of tiled images with corresponding pathological diagnoses.

  • Validate the model's performance on a separate, unseen test set of WSIs.

• Performance Evaluation:

  • Calculate the Receiver Operating Characteristic (ROC) Area Under the Curve (AUC) to assess the model's classification performance.

  • Generate heatmaps to visualize the areas within the WSI that the AI model identifies as cancerous.

Experiment 2: Classification of Colorectal Adenocarcinoma

This protocol outlines the methodology for classifying poorly differentiated colorectal adenocarcinoma using a transfer learning approach with Medmain's AI models.[8]

Objective: To accurately classify whole-slide images of colorectal biopsies into poorly differentiated adenocarcinoma or other tissue types.

Experimental Protocol:

• Dataset Acquisition:

  • Assemble a dataset of colorectal biopsy WSIs from multiple institutions to ensure data diversity.

  • Include cases of poorly differentiated adenocarcinoma as well as other relevant tissue classifications.

  • Ensure all necessary patient consent and ethical approvals are in place.[8]

• Model Training using Transfer Learning:

  • Select a pre-trained CNN model (e.g., EfficientNetB1) as the base for the classifier.[8]

  • Given the often-limited number of poorly differentiated adenocarcinoma cases for training, employ a transfer learning method.[8]

  • Fine-tune the model on the colorectal WSI dataset.

• WSI-level Prediction:

  • Apply the trained model in a sliding window fashion across each WSI in the test set to generate predictions for individual tiles.

  • Aggregate the tile-level predictions to obtain a single prediction for the entire WSI, typically by taking the maximum prediction score.[8]

• Performance Assessment:

  • Evaluate the model's performance using ROC-AUC on the test set.

  • Analyze the model's ability to generalize across datasets from different institutions.

Quantitative Data Summary

The performance of Medmain's AI models has been rigorously evaluated in multiple studies. The following tables summarize the quantitative results from key publications.

Table 1: Performance of Gastric Poorly Differentiated Adenocarcinoma Classification Model [6][7][11]

Test Set	Specimen Type	Number of WSIs	ROC-AUC
Test Set 1	ESD	200	0.975
Test Set 2	Endoscopic Biopsy	500	0.960
Test Set 3	Surgical Specimen	150	0.929

Table 2: Performance of Multi-Organ Adenocarcinoma Classification Model [9][10]

Primary Organ	Number of WSIs in Test Set	ROC-AUC
Stomach	1,200	0.98
Colon	1,500	0.97
Lung	800	0.95
Breast	1,000	0.91

Table 3: Performance of Poorly Differentiated Colorectal Adenocarcinoma Classification Model [8]

Test Set Source	Number of WSIs	ROC-AUC
Combined (5 Institutions)	1799	up to 0.95

Signaling Pathway Visualization

Understanding the underlying molecular pathways is crucial for cancer research and drug development. The following diagram illustrates a simplified view of key signaling pathways commonly dysregulated in colorectal cancer, a target for Medmain's AI models.

Key signaling pathways implicated in colorectal cancer development and progression.

Conclusion

Medmain's PidPort platform and its integrated AI analytics offer a powerful and efficient workflow for researchers in pathology and drug development. By providing tools for centralized data management, rapid and accurate image analysis, and collaborative review, Medmain is accelerating the pace of discovery. The high performance of their validated AI models in cancer classification demonstrates the potential of this technology to revolutionize how pathological data is utilized in research, ultimately leading to new insights and therapeutic strategies.

References

• 1. Medmain inc. [en.medmain.com]
• 2. healthtechalpha.com [healthtechalpha.com]
• 3. Medmain inc. [en.medmain.com]
• 4. pidport.medmain.com [pidport.medmain.com]
• 5. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 6. medrxiv.org [medrxiv.org]
• 7. Weakly Supervised Learning for Poorly Differentiated Adenocarcinoma Classification in GastricEndoscopic Submucosal Dissection Whole Slide Images - PMC [pmc.ncbi.nlm.nih.gov]
• 8. mdpi.com [mdpi.com]
• 9. Weakly supervised learning for multi-organ adenocarcinoma classification in whole slide images | PLOS One [journals.plos.org]
• 10. Weakly supervised learning for multi-organ adenocarcinoma classification in whole slide images - PMC [pmc.ncbi.nlm.nih.gov]
• 11. Medmain inc. [en.medmain.com]

Application Notes and Protocols for Integrating PidPort with Existing Laboratory Information Systems

Introduction

These application notes provide a comprehensive guide for the integration of PidPort, a next-generation data aggregation and analysis platform, with existing Laboratory Information Systems (LIS). The successful integration of PidPort is crucial for creating a seamless digital ecosystem within the laboratory, enabling automated data flow, enhancing data integrity, and accelerating research and development cycles. This document outlines various integration strategies, provides detailed protocols for implementation, and offers recommendations for best practices. The target audience for these notes includes research scientists, laboratory managers, and IT professionals involved in the drug development process.

It is important to note that "PidPort" is used here as a representative example of a modern data integration and analysis platform. The principles and protocols described can be adapted for similar platforms that require integration with an LIS.

Core Concepts: The Role of PidPort and LIS

A Laboratory Information System (LIS) is a software system that records, manages, and stores data for clinical and research laboratories. Its primary functions include managing patient or sample demographics, tracking samples, and managing test results.

PidPort is conceptualized as a platform that aggregates data from various sources, including the LIS, analytical instruments, and other databases. It provides advanced tools for data analysis, visualization, and collaborative research, offering deeper insights into experimental data. The integration of these two systems is paramount for a holistic view of laboratory operations and research data.

Integration Strategies

The choice of integration strategy depends on the capabilities of the existing LIS, the volume and velocity of data, and the specific needs of the laboratory. The following table summarizes the most common integration methods.

Table 1: Comparison of LIS Integration Methods

Integration Method	Description	Data Transfer	Complexity	Scalability
API-Based	Direct communication between PidPort and the LIS via Application Programming Interfaces (APIs).	Real-time or near real-time.	High	High
File-Based	PidPort and the LIS exchange data through structured files (e.g., CSV, XML, HL7) in a shared location.	Batch	Low	Low to Medium
Middleware	A third-party software acts as an intermediary, translating and routing data between PidPort and the LIS.	Real-time or batch.	Medium to High	High
Database Direct	PidPort directly queries and writes to the LIS database.	Real-time.	Very High	Medium

Experimental Protocols

Protocol 1: API-Based Integration

This protocol outlines the steps for establishing a real-time, bidirectional data flow between PidPort and an LIS using APIs.

Methodology:

• Pre-Integration Assessment:

  • Identify the API capabilities of the existing LIS. Does it have a well-documented RESTful API?

  • Define the specific data points to be exchanged (e.g., sample IDs, test results, patient data).

  • Consult with the LIS vendor to obtain API documentation, access credentials, and understand any rate limits.

• Environment Setup:

  • Establish a secure network connection between PidPort and the LIS.

  • Configure firewalls and security groups to allow traffic on the designated ports.

• API Endpoint Configuration in PidPort:

  • Navigate to the "Data Integrations" section in the PidPort administrative panel.

  • Select "Add New Integration" and choose "LIS via API."

  • Enter the LIS API base URL, authentication credentials (e.g., API key, OAuth token), and other required parameters.

• Data Mapping:

  • Utilize the PidPort interface to map data fields from the LIS API response to the corresponding fields in PidPort. Refer to Table 2 for a sample data mapping configuration.

  • Define data transformation rules if necessary (e.g., converting units, reformatting dates).

• Testing and Validation:

  • Initiate a test data transfer from a sandboxed LIS environment.

  • Verify that data is accurately and completely transferred to PidPort.

  • Test the bidirectional data flow by pushing a data update from PidPort back to the LIS.

  • Review logs in both systems for any errors.

• Deployment and Monitoring:

  • Once validation is successful, deploy the integration in the production environment.

  • Set up monitoring and alerting in PidPort to be notified of any integration failures or performance issues.

Table 2: Sample Data Mapping for API Integration

LIS API Field	PidPort Field	Data Type	Transformation Rule
sample_id	SampleID	String	None
patient.name	PatientName	String	Concatenate patient.first_name and patient.last_name
test_result.value	ResultValue	Float	Convert to standard units
test_result.timestamp	ResultTimestamp	Datetime	Parse ISO 8601 format

Protocol 2: File-Based Integration

This protocol details the steps for setting up a batch data transfer using structured files.

Methodology:

• Pre-Integration Assessment:

  • Determine the file formats supported by the LIS for data export (e.g., CSV, XML, HL7).

  • Establish a secure, shared network location (e.g., an SFTP server) accessible by both PidPort and the LIS.

  • Define the file naming conventions and the frequency of file generation.

• LIS Configuration:

  • Configure the LIS to automatically export the required data to the shared location at scheduled intervals.

  • Ensure the exported files adhere to the agreed-upon format and naming convention.

• PidPort Configuration:

  • In the PidPort administrative panel, navigate to "Data Integrations" and select "File-Based Integration."

  • Provide the connection details for the shared network location (e.g., SFTP server address, username, password).

  • Specify the file path, file naming pattern, and the polling frequency for PidPort to check for new files.

• Data Parsing and Mapping:

  • Configure the data parser in PidPort based on the file format (e.g., specify delimiters for CSV files).

  • Map the columns or fields in the source file to the corresponding fields in PidPort.

• Testing and Validation:

  • Manually place a test file in the shared location.

  • Trigger the PidPort file listener and verify that the data is correctly ingested and mapped.

  • Check for error handling by providing a malformed test file.

• Deployment and Monitoring:

  • Enable the automated file transfer schedule in the LIS and the file listener in PidPort.

  • Monitor the integration for any file processing errors or delays.

Visualizations

Caption: Overall data flow from instruments to LIS and PidPort.

Caption: Workflow for a typical API-based data request.

Caption: Decision tree for selecting an integration method.

Application Notes and Protocols for Quantitative Analysis of Histopathological Slides using Medmain's PidPort™

Audience: Researchers, scientists, and drug development professionals.

Introduction

Medmain's PidPort™ is a cloud-based digital pathology platform that leverages powerful, deep-learning artificial intelligence (AI) to provide quantitative analysis of histopathological slides. This technology offers a streamlined workflow for researchers and drug development professionals, enabling objective, reproducible, and highly accurate analysis of tissue samples. PidPort™ assists in identifying and quantifying cellular and tissue features, which is critical for efficacy and toxicity studies in preclinical and clinical research.

The platform's AI models are trained on extensive datasets of whole-slide images (WSIs) to perform tasks such as tumor detection, classification, and quantification of biomarkers. These capabilities aim to reduce the burden on pathologists, increase throughput, and provide robust quantitative data for research and development.[1][2][3]

Key Features and Applications

Medmain's PidPort™ platform offers a suite of tools to facilitate the quantitative analysis of histopathological slides:

• Cloud-Based Platform: Enables easy upload, storage, viewing, and sharing of high-resolution whole-slide images from anywhere with an internet connection.

• Quantitative Data Output: Provides objective and reproducible quantitative data, such as tumor area, cell counts, and biomarker expression levels.

• Collaborative Tools: Features annotation and commenting functionalities to facilitate collaboration among researchers and pathologists.

Applications in Research and Drug Development:

• Efficacy Studies: Quantify tumor response to novel therapeutics by measuring changes in tumor size, cellularity, and proliferation markers.

• Toxicity and Safety Assessment: Objectively assess tissue damage and pathological changes in toxicology studies.

• Biomarker Discovery and Validation: Identify and quantify the expression of novel biomarkers for patient stratification and predicting treatment response.

• High-Throughput Screening: Accelerate the analysis of large-scale histopathology studies in preclinical research.

Quantitative Data Summary

The following tables summarize the performance of Medmain's AI models in various cancer types, as reported in peer-reviewed publications. The primary metric shown is the Area Under the Receiver Operating Characteristic Curve (AUC), which represents the model's ability to distinguish between different classes (e.g., tumor vs. normal tissue).

Gastric Cancer AI Model Performance

Application	Tissue Type	Model	Test Set	AUC	Journal
Adenocarcinoma Classification	Endoscopic Biopsy	EfficientNetB1	Hospital 1	0.97	Scientific Reports
Adenocarcinoma Classification	Endoscopic Biopsy	EfficientNetB1	Hospital 2	0.99	Scientific Reports
Signet Ring Cell Carcinoma	Biopsy	Ensemble	Test Set 1	0.99	Technology in Cancer Research & Treatment[8]
Signet Ring Cell Carcinoma	Biopsy	Ensemble	Test Set 2	0.99	Technology in Cancer Research & Treatment[8]

Colorectal Cancer AI Model Performance

Application	Tissue Type	Model	Test Set	AUC	Journal
Poorly Differentiated Adenocarcinoma	Endoscopic Biopsy	Ensemble	Combined Test Set	0.95	Diagnostics[3][9]

Breast Cancer AI Model Performance

Application	Tissue Type	Model	Test Set	AUC	Journal
Invasive Ductal Carcinoma	Core Needle Biopsy	TL model x10	Hospital 1 (biopsy)	0.98	Cancers[10][11]
Invasive Ductal Carcinoma	Surgical	TL model x10	Hospital 1 (surgical)	0.95	Cancers[10][11]
Invasive Ductal Carcinoma	Surgical	TL model x10	TCGA (surgical)	0.97	Cancers[10][11]

Prostate Cancer AI Model Performance

Application	Tissue Type	Model	Test Set	AUC	Journal
Adenocarcinoma Classification	Needle Biopsy	Ensemble	Needle Biopsy	0.978	Diagnostics[7]
Adenocarcinoma Classification	Needle Biopsy	Ensemble	TCGA	0.987	Diagnostics[7]
Adenocarcinoma Classification	TUR-P	Ensemble	TUR-P	0.984	Cancers[6]

Multi-Organ Adenocarcinoma AI Model Performance

Application	Tissue Type	Model	Test Set	AUC	Journal
Adenocarcinoma Classification	Stomach Biopsy	WS-4K: 224, x10 EfficientNetB1	Stomach	0.98	PLOS ONE[1][5]
Adenocarcinoma Classification	Colon Biopsy	WS-4K: 224, x10 EfficientNetB1	Colon	0.97	PLOS ONE[1][5]
Adenocarcinoma Classification	Lung Biopsy	WS-4K: 224, x10 EfficientNetB1	Lung	0.91	PLOS ONE[1][5]
Adenocarcinoma Classification	Breast Biopsy	WS-4K: 224, x10 EfficientNetB1	Breast	0.96	PLOS ONE[1][5]
Metastatic Adenocarcinoma	Lymph Node Dissection	WS-4K: 224, x10 EfficientNetB1	Lymph Node	0.97	PLOS ONE[1][5]

Experimental Protocols

The following protocols are generalized from methodologies reported in publications utilizing Medmain's AI technology. For specific research applications, optimization of these protocols may be necessary.

Histopathological Slide Preparation and Digitization

This protocol outlines the steps for preparing tissue samples and digitizing them into whole-slide images compatible with the PidPort™ platform.

Materials:

• Tissue samples

• 10% neutral buffered formalin

• Paraffin (B1166041) wax

• Microtome

• Glass slides

• Hematoxylin and Eosin (H&E) stains

• Automated slide scanner

Procedure:

• Fixation: Immediately fix fresh tissue samples in 10% neutral buffered formalin for 24-48 hours.

• Processing and Embedding: Dehydrate the fixed tissues through a series of graded alcohols, clear with xylene, and embed in paraffin wax.

• Sectioning: Cut 3-5 µm thick sections from the paraffin blocks using a microtome.

• Staining: Mount the sections on glass slides and perform standard Hematoxylin and Eosin (H&E) staining.

• Scanning: Digitize the H&E-stained slides at 20x or 40x magnification using a high-resolution whole-slide scanner to create WSI files.

• Data Upload: Upload the WSI files to the Medmain PidPort™ cloud platform.

Quantitative Analysis Workflow using PidPort™

This protocol describes the general workflow for performing quantitative analysis of digitized slides using the PidPort™ platform.

Procedure:

• Case Creation: Create a new case within the PidPort™ platform and upload the corresponding WSI(s).

• AI Model Selection: Select the appropriate pre-trained AI model for the cancer type and tissue under investigation (e.g., Gastric Adenocarcinoma, Breast Invasive Ductal Carcinoma).

• AI Analysis Execution: Initiate the AI analysis on the selected WSI(s). The platform's AI will process the image to identify and classify regions of interest.

• Quantitative Data Extraction: The platform provides quantitative data based on the AI analysis, which may include:

  • Tumor area (mm²)

  • Percentage of tumor in the tissue

  • Cell counts (e.g., tumor cells, inflammatory cells)

• Data Export: Export the quantitative data in a suitable format (e.g., CSV) for further statistical analysis.

Visualizations

Digital Pathology Workflow with Medmain PidPort™

The following diagram illustrates the typical workflow for quantitative histopathological analysis using Medmain's PidPort™ platform.

Caption: Workflow for quantitative analysis using Medmain PidPort™.

Weakly Supervised Learning for AI Model Training

Medmain utilizes a weakly supervised learning approach for training its AI models, where the model learns from whole-slide images with only slide-level labels (e.g., "tumor" or "normal"), rather than requiring detailed, pixel-level annotations.[1][12][13]

Caption: Weakly supervised learning workflow for AI model training.

Conclusion

Medmain's PidPort™ platform provides a robust and efficient solution for the quantitative analysis of histopathological slides. By integrating advanced AI with a user-friendly cloud-based system, PidPort™ empowers researchers, scientists, and drug development professionals to obtain objective and reproducible data, thereby accelerating research and enhancing decision-making in the development of new therapies. The platform's validated performance across a range of cancer types demonstrates its potential to become an indispensable tool in modern pathology research.

References

• 1. medrxiv.org [medrxiv.org]
• 2. researchgate.net [researchgate.net]
• 3. researchgate.net [researchgate.net]
• 4. Medmain inc. [en.medmain.com]
• 5. Weakly supervised learning for multi-organ adenocarcinoma classification in whole slide images - PMC [pmc.ncbi.nlm.nih.gov]
• 6. mdpi.com [mdpi.com]
• 7. A Deep Learning Model for Prostate Adenocarcinoma Classification in Needle Biopsy Whole-Slide Images Using Transfer Learning - PMC [pmc.ncbi.nlm.nih.gov]
• 8. Product Details | Medmain Inc. [en.medmain.com]
• 9. mdpi.com [mdpi.com]
• 10. mdpi.com [mdpi.com]
• 11. researchgate.net [researchgate.net]
• 12. medrxiv.org [medrxiv.org]
• 13. Deep Learning Models for Gastric Signet Ring Cell Carcinoma Classification in Whole Slide Images - PMC [pmc.ncbi.nlm.nih.gov]

Application Notes and Protocols for Medmain's AI in Breast Cancer Analysis

For Researchers, Scientists, and Drug Development Professionals

These application notes provide a comprehensive overview of the research applications of Medmain's artificial intelligence (AI) platform, PidPort, in the analysis of breast cancer. This document details the underlying technology, performance metrics, and experimental protocols for utilizing this AI in research and drug development settings.

Introduction to Medmain's AI for Breast Cancer Analysis

Medmain Inc. has developed a suite of deep learning models integrated into their cloud-based digital pathology platform, PidPort, for the histopathological analysis of breast cancer.[1][2][3] The primary application focuses on the classification of breast cancer subtypes from whole slide images (WSIs) of tissue samples. The AI models are designed to assist pathologists in making rapid and accurate diagnoses, thereby streamlining the workflow for both clinical and research purposes.[1][2]

The core technology leverages deep learning, specifically transfer learning and weakly-supervised learning, to train models on extensive datasets of digitized histopathological slides.[1][4] This approach allows the AI to learn and identify key features associated with different breast cancer pathologies, such as Invasive Ductal Carcinoma (IDC) and Ductal Carcinoma in Situ (DCIS).[1][5][6]

Key Applications in Breast Cancer Research

Medmain's AI technology offers several applications for researchers and drug development professionals:

• High-throughput screening of histopathology slides: Automated analysis of large cohorts of digitized breast tissue samples for identifying specific morphologies.

• Quantitative analysis of tumor characteristics: Objective and reproducible classification of breast cancer subtypes, aiding in cohort stratification for clinical trials or preclinical studies.

• Identification of regions of interest (ROIs): The AI can highlight areas of potential malignancy within a WSI, enabling more focused analysis by researchers.

Performance of Medmain's AI Models

The performance of Medmain's AI models has been validated in peer-reviewed studies. The primary metric used to evaluate the classification models is the Area Under the Receiver Operating Characteristic Curve (AUC).

Invasive Ductal Carcinoma (IDC) Classification

A study published in Cancers detailed the development and validation of a deep learning model for the classification of IDC in both core needle biopsy and surgical specimens.[1][4] The models were evaluated on four independent test sets.

Table 1: Performance of IDC Classification Model

Test Set	Specimen Type	Number of WSIs	Model	Magnification	Tile Size (pixels)	AUC
Biopsy Test Set	Core Needle Biopsy	522	EfficientNetB1	10x	512 x 512	0.98
Surgical Test Set 1	Surgical Excision	378	EfficientNetB1	10x	512 x 512	0.97
Surgical Test Set 2	Surgical Excision	451	EfficientNetB1	10x	512 x 512	0.96
Surgical Test Set 3	Surgical Excision	300	EfficientNetB1	10x	512 x 512	0.96

Source: "Breast Invasive Ductal Carcinoma Classification on Whole Slide Images with Weakly-Supervised and Transfer Learning", Cancers (MDPI)[1][4]

Ductal Carcinoma in Situ (DCIS) and IDC Classification

A subsequent study focused on the differential classification of DCIS, IDC, and benign tissue from WSIs. The models were evaluated on two independent test sets.

Table 2: Performance of DCIS and IDC Classification Model

Test Set	Number of WSIs	Target	Model	AUC
Test Set 1	1,382	DCIS	CNN + RNN	0.960
Test Set 1	1,382	IDC	CNN + RNN	0.977
Test Set 2	548	DCIS	CNN + RNN	0.955
Test Set 2	548	IDC	CNN + RNN	0.971

Source: "A deep learning model for breast ductal carcinoma in situ classification in whole slide images", Virchows Archiv[5][7]

Experimental Protocols

The following protocols are derived from the methodologies published by Medmain for the development and application of their breast cancer classification models.[4][7]

Protocol for Whole Slide Image (WSI) Preparation and Digitization

• Tissue Preparation: Formalin-fixed, paraffin-embedded (FFPE) breast tissue blocks are sectioned at 3-5 µm.

• Staining: Sections are stained with Hematoxylin and Eosin (H&E).

• Scanning: Stained slides are digitized at high resolution (e.g., 20x or 40x magnification) using a whole slide scanner to create WSIs in a standard format (e.g., SVS, TIFF).

• Data Management: WSIs are uploaded to a secure server or a digital pathology platform like Medmain's PidPort for storage and analysis.

Protocol for AI-based Classification of Breast Cancer Subtypes

• Image Pre-processing:

  • WSIs are tiled into smaller, manageable image patches (e.g., 224x224 or 512x512 pixels).

  • Patches are normalized for color and brightness variations.

  • Non-tissue background areas are excluded from the analysis.

• Model Training (for model development):

  • A large dataset of annotated WSIs (classified as benign, DCIS, or IDC by pathologists) is assembled.

  • A deep learning model architecture (e.g., EfficientNetB1) is chosen.

  • Transfer Learning: The model is pre-trained on a large, general image dataset (e.g., ImageNet) and then fine-tuned using the breast histopathology dataset.

  • Weakly-Supervised Learning: The model is trained using slide-level labels, where the model learns to identify relevant features from the entire slide without requiring detailed annotation of every region.

• Inference (for analysis):

  • The trained model is applied to the pre-processed tiles of a new WSI.

  • The model assigns a probability score for each class (e.g., benign, DCIS, IDC) to each tile.

  • An aggregation method (e.g., averaging or max-pooling) is used to generate a slide-level prediction.

• Output Visualization:

  • The AI generates a heatmap overlaid on the WSI, indicating the probability of malignancy in different regions.

  • A final slide-level classification is provided.

Visualizations

Experimental Workflow for AI Model Development and Application

The following diagram illustrates the general workflow for developing and utilizing Medmain's AI for breast cancer analysis.

References

• 1. Medmain inc. [en.medmain.com]
• 2. pidport.medmain.com [pidport.medmain.com]
• 3. Medmain inc. [en.medmain.com]
• 4. Breast Invasive Ductal Carcinoma Classification on Whole Slide Images with Weakly-Supervised and Transfer Learning - PMC [pmc.ncbi.nlm.nih.gov]
• 5. A deep learning model for breast ductal carcinoma in situ classification in whole slide images - PubMed [pubmed.ncbi.nlm.nih.gov]
• 6. Product Details | Medmain Inc. [en.medmain.com]
• 7. medrxiv.org [medrxiv.org]

Application Notes and Protocols: Leveraging Medmain for Large-Scale Data Analysis in Pathology

For Researchers, Scientists, and Drug Development Professionals

Introduction

PidPort's core functionalities include a high-speed WSI viewer, secure cloud storage, and a suite of deep learning-based AI models for the histopathological analysis of various cancers, including those of the stomach, colon, breast, lung, and pancreas.[5][7] These AI models are developed through collaborative research with multiple medical institutions, utilizing hundreds of thousands of pathological specimens to train and validate the algorithms.[3][4]

These application notes provide detailed protocols for leveraging Medmain's PidPort platform for large-scale data analysis in pathology research and drug development. The protocols are based on methodologies outlined in Medmain's peer-reviewed publications and are intended to guide users in harnessing the full potential of this powerful tool.

Quantitative Data Presentation

The performance of Medmain's AI models has been validated in several studies, demonstrating high accuracy in the classification of various tumor types. The following tables summarize the key performance metrics of these models, providing a clear comparison for researchers selecting the appropriate tools for their studies.

Table 1: Performance of Deep Learning Models for Histopathological Classification of Gastric and Colonic Epithelial Tumors [8][9][10]

Tissue	Classification	Test Set	Area Under the Curve (AUC)
Gastric	Adenocarcinoma vs. Non-neoplastic	Independent Test Set 1	0.97
Gastric	Adenoma vs. Non-neoplastic	Independent Test Set 1	0.99
Colonic	Adenocarcinoma vs. Non-neoplastic	Independent Test Set 1	0.96
Colonic	Adenoma vs. Non-neoplastic	Independent Test Set 1	0.99

Table 2: Performance of Deep Learning Model for Detection of Pancreatic Ductal Adenocarcinoma (PDAC) on Endoscopic Ultrasound-Guided Fine-Needle Biopsy (EUS-FNB) [1][3][7]

Metric	Value
Receiver Operator Curve Area Under the Curve (ROC AUC)	0.984
Accuracy	0.9417
Sensitivity	0.9302
Specificity	0.9706

Table 3: Performance of Deep Learning Model for Classification of Cutaneous Melanoma [11]

Evaluation Level	Metric	Value
Whole Slide Image (WSI)	ROC AUC	0.821
Tile Level	ROC AUC	0.936

Experimental Protocols

The following protocols provide a step-by-step guide for utilizing the Medmain PidPort platform for large-scale pathology data analysis. These are generalized protocols derived from published research and may require optimization for specific research questions.

Protocol 1: Large-Scale Histopathological Screening of Gastrointestinal Tumors

This protocol outlines the workflow for the automated classification of gastric and colonic epithelial tumors from whole slide images using Medmain's AI models.

Methodology:

• Data Acquisition and Digitization:

  • Collect formalin-fixed, paraffin-embedded (FFPE) tissue blocks from gastric or colonic biopsies.

  • Prepare 3-5 µm thick sections and stain with Hematoxylin and Eosin (H&E).

  • Digitize the glass slides at high resolution (e.g., 20x or 40x magnification) to generate whole slide images (WSIs) in a compatible format (e.g., SVS, NDPI, JPEG, TIFF).[3]

• Data Upload and Management in PidPort:

  • Log in to the PidPort cloud platform.

  • Utilize the bulk upload function to upload the WSIs and associated case data.[3][5]

  • Organize the cases into a new project folder and apply relevant tags for easy retrieval.

• • Select the cohort of WSIs for analysis.

  • Initiate the appropriate AI analysis pipeline for either gastric or colonic tumors. The platform's AI will automatically classify the WSIs into categories such as adenocarcinoma, adenoma, and non-neoplastic.[8][9]

  • The AI analysis is based on trained convolutional neural networks (CNNs) and recurrent neural networks (RNNs).[8][9][10]

• Review and Verification of AI Results:

• Data Export and Statistical Analysis:

  • Export the quantitative results from PidPort, including the classification for each WSI.

  • Perform statistical analysis to correlate the histopathological classifications with other clinical or molecular data.

Protocol 2: Automated Detection of Pancreatic Ductal Adenocarcinoma (PDAC)

This protocol describes the use of Medmain's AI model to detect PDAC in WSIs from endoscopic ultrasound-guided fine-needle biopsies.

Methodology:

• Specimen Collection and Slide Preparation:

  • Obtain EUS-FNB specimens from the pancreas.

  • Process the specimens into FFPE blocks and prepare H&E stained slides.

  • Digitize the slides to create WSIs.

• Project Setup in PidPort:

  • Create a new project in PidPort specifically for the PDAC detection study.

  • Upload the WSIs and any relevant clinical information for each case.

• Execution of the PDAC Detection AI Model:

  • Select the uploaded WSIs for analysis.

  • Run the specialized AI model for PDAC detection. The model will analyze the complex cytological and architectural features within the biopsy specimens.[1][3][7]

• Visualization and Interpretation of Results:

  • The PidPort viewer will display the results, highlighting regions of interest identified by the AI as suspicious for PDAC.

• Quantitative Analysis and Reporting:

  • Extract the binary classification results (PDAC present/absent) for each WSI.

  • Calculate performance metrics such as sensitivity, specificity, and accuracy against a ground truth established by expert pathologists.

Mandatory Visualizations

Workflow and Logical Relationship Diagrams

The following diagrams, generated using Graphviz (DOT language), illustrate key workflows and logical relationships in leveraging the Medmain platform.

Caption: General workflow for large-scale data analysis using Medmain's PidPort platform.

Caption: Conceptual diagram of Medmain's AI model development process.

Caption: Application of Medmain's platform in the drug development pipeline.

Disclaimer: The provided protocols are for informational purposes and should be adapted to specific research needs. Users should consult Medmain's official documentation and support for detailed guidance on using the PidPort platform. The performance of AI models may vary depending on the specific dataset and experimental conditions.

References

• 1. (PDF) A deep learning model to detect pancreatic ductal adenocarcinoma on endoscopic ultrasound-guided fine-needle biopsy (2021) | Yoshiki Naito | 50 Citations [scispace.com]
• 2. Medmain inc. [en.medmain.com]
• 3. A deep learning model to detect pancreatic ductal adenocarcinoma on endoscopic ultrasound-guided fine-needle biopsy - PubMed [pubmed.ncbi.nlm.nih.gov]
• 4. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 5. pidport.medmain.com [pidport.medmain.com]
• 6. pidport.medmain.com [pidport.medmain.com]
• 7. A deep learning model to detect pancreatic ductal adenocarcinoma on endoscopic ultrasound-guided fine-needle biopsy. [ncbiobank.org]
• 8. Deep Learning Models for Histopathological Classification of Gastric and Colonic Epithelial Tumours - PubMed [pubmed.ncbi.nlm.nih.gov]
• 9. [PDF] Deep Learning Models for Histopathological Classification of Gastric and Colonic Epithelial Tumours | Semantic Scholar [semanticscholar.org]
• 10. researchgate.net [researchgate.net]
• 11. Deep Learning Approach to Classify Cutaneous Melanoma in a Whole Slide Image - PMC [pmc.ncbi.nlm.nih.gov]

Revolutionizing Research: Generating High-Quality Data with PidPort's Annotation Tools

In the fast-paced world of drug discovery and development, the ability to efficiently generate accurate and reproducible research data is paramount. PidPort, a cloud-based digital pathology platform, offers a suite of powerful annotation tools designed to streamline the analysis of whole slide images (WSIs) and facilitate the generation of robust quantitative data for research applications. These tools empower researchers, scientists, and drug development professionals to unlock critical insights from pathological specimens, accelerating the journey from preclinical research to clinical application.

This document provides detailed application notes and protocols for leveraging PidPort's annotation tools for research data generation, with a focus on oncology and drug development.

Application Notes

The annotation tools within the PidPort platform are instrumental in the development of artificial intelligence (AI) models for pathological diagnosis and analysis. By enabling precise delineation and classification of various tissue types and cellular features on whole slide images (WSIs), researchers can generate the high-quality, annotated datasets required for training and validating deep learning algorithms.

A key application is the development of AI models for cancer diagnosis and grading. For instance, in a study focused on prostate cancer, PidPort's annotation tools were utilized to classify prostate adenocarcinoma on core needle biopsy WSIs into "indolent" (suitable for active surveillance) and "aggressive" (requiring definitive therapy) categories. This classification was based on the Gleason scoring system, a critical prognostic indicator.

The performance of the resulting AI model demonstrates the power of high-quality annotated data in training accurate diagnostic tools. The model achieved a high level of accuracy in distinguishing between indolent and aggressive prostate cancer, showcasing the potential of this technology to assist pathologists in making critical treatment decisions.

Quantitative Data Summary

The following table summarizes the performance of an AI model developed using data annotated within the PidPort platform for the classification of prostate adenocarcinoma.

Performance Metric	Indolent Classification	Aggressive Classification
ROC-AUC	0.846	0.980

This data is derived from a study on the classification of prostate adenocarcinoma using deep learning models trained on annotated whole slide images.[1][2]

Experimental Protocols

The generation of high-quality research data using PidPort's annotation tools involves a systematic workflow. The following protocols outline the key steps for annotating WSIs for the development of a pathology AI model.

Protocol 1: Annotation of Histopathological Features for AI Model Training

Objective: To create a dataset of annotated WSIs for training a deep learning model to classify specific pathological features (e.g., tumor regions, cell types).

Materials:

• Whole Slide Images (WSIs) of tissue specimens (e.g., formalin-fixed, paraffin-embedded tissue sections stained with hematoxylin (B73222) and eosin (B541160) - H&E)

• PidPort cloud-based digital pathology platform

• Workstation with a high-resolution monitor

Procedure:

• Case and Image Selection:

  • Select a cohort of cases with definitive pathological diagnoses relevant to the research question.

  • Upload the corresponding WSIs to the PidPort platform. For the prostate cancer study, a dataset of 1300 core needle biopsy WSIs was utilized for training.[1][2]

• Annotation by Pathologists:

  • Experienced pathologists access the WSIs through the PidPort viewer.

  • Utilizing PidPort's drawing and annotation tools, pathologists manually delineate regions of interest (ROIs) on the WSIs.

  • For the prostate cancer model, annotations were made to identify and classify Gleason patterns 3, 4, and 5 of adenocarcinoma.[3]

  • Each annotation is assigned a specific label corresponding to the pathological feature (e.g., "Gleason Pattern 3," "Tumor Stroma," "Normal Epithelium").

• Weakly Supervised Learning Approach (Optional but Recommended):

  • In cases with large datasets, a weakly supervised learning approach can be employed to expedite the annotation process.

  • Initially, pathologists provide slide-level labels (e.g., "indolent," "aggressive") without detailed annotations.

  • The AI model is first trained on these slide-level labels to identify potential regions of interest.

• Data Curation and Quality Control:

  • A second pathologist or a quality control team reviews a subset of the annotated slides to ensure consistency and accuracy of the annotations.

  • Any discrepancies are resolved through consensus.

• Export of Annotated Data:

  • The annotated data, including the image coordinates of the delineated regions and their corresponding labels, is exported from the PidPort platform.

  • This structured data is then used as the ground truth for training and validating the deep learning model.

Visualizing the Research Workflow

The following diagrams illustrate the key workflows in utilizing PidPort for research data generation and AI model development.

Caption: Experimental workflow for AI model development using PidPort.

Caption: Logical flow of the annotation process within the PidPort platform.

By providing a robust and user-friendly environment for the annotation of digital pathology images, PidPort empowers research teams to generate the high-quality, structured data necessary to drive innovation in drug discovery and development. The ability to create large, accurately annotated datasets is a critical step in harnessing the power of artificial intelligence to unlock new frontiers in medicine.

References

• 1. Inference of core needle biopsy whole slide images requiring definitive therapy for prostate cancer - PMC [pmc.ncbi.nlm.nih.gov]
• 2. medrxiv.org [medrxiv.org]
• 3. researchgate.net [researchgate.net]

Troubleshooting & Optimization

PidPort Technical Support Center: Troubleshooting Image Uploads

This guide provides solutions to common image upload issues encountered on the PidPort platform. For optimal experiment documentation and data integrity, it is crucial to ensure all visual data is uploaded correctly.

Frequently Asked Questions (FAQs) - Image Uploads

Q1: Why is my image upload failing repeatedly?

A: Image upload failures can stem from several factors. The most common reasons include incorrect file format, exceeding the maximum file size, or network connectivity issues.[1][2][3] Start by verifying that your image meets the platform's specifications outlined in the table below. Also, ensure you have a stable internet connection.[1]

Q2: What are the accepted image formats and size limits on PidPort?

A: PidPort supports a range of standard image formats to accommodate various types of research data, from microscopy images to data plots. Please refer to the PidPort Image Upload Specifications table for detailed information on supported formats and size limitations.

Q3: I received a "403 Forbidden" error while uploading. What does this mean?

A: A "403 Forbidden" error typically indicates a permission issue.[4] This could be due to your user account not having the necessary permissions for the specific project or dataset you are trying to upload to. Contact your project administrator to verify your user privileges. In some cases, institutional firewalls may also block uploads; please check with your IT department.[1]

Q4: My upload is extremely slow. How can I speed it up?

A: Slow upload speeds are generally related to your internet connection or large file sizes.[2] Consider the following:

• Compressing Images: For very large images, such as high-resolution microscopy scans, consider using image compression software to reduce the file size without significant loss of quality.

• Wired Connection: Use a wired ethernet connection instead of Wi-Fi for a more stable and faster connection.[2]

• Off-Peak Hours: Try uploading during off-peak network hours at your institution.

• Single File Upload: Avoid bulk uploading large files; upload them one at a time to prevent timeouts.[2]

Q5: Can I upload images with special characters in the filename?

A: No, it is strongly recommended to avoid special characters (e.g., $, %, @, !, *) in filenames as they can cause upload errors.[2][3] Use only alphanumeric characters, underscores, and hyphens for your filenames.

PidPort Image Upload Specifications

Parameter	Specification	Notes for Researchers
Supported File Formats	.PNG, .JPEG, .TIFF, .BMP	For microscopy images, .TIFF is often preferred to preserve metadata. For plots and graphs, .PNG is recommended for lossless compression.
Maximum File Size	500 MB	High-resolution images from confocal microscopy or whole-slide scans may exceed this. Consider image tiling or compression.
Recommended Resolution	300 DPI	This ensures high-quality images suitable for publication and detailed analysis.
Filename Conventions	Alphanumeric, underscores, hyphens	Avoid special characters and spaces to prevent upload failures. A consistent naming convention is advised for better data management (e.g., ProjectID_SampleID_Date.tiff).

Experimental Protocol: Uploading Western Blot Images for Analysis

This protocol outlines the standardized procedure for uploading Western Blot images to PidPort for densitometric analysis and record-keeping.

1.0 Objective: To correctly digitize and upload Western Blot images to the PidPort platform for quantitative analysis.

2.0 Materials:

• Chemiluminescence imager (e.g., Bio-Rad ChemiDoc)

• Image analysis software (e.g., ImageJ, FIJI)

• Computer with stable internet access

3.0 Procedure:

• Image Acquisition:

  • Acquire the Western Blot image using a chemiluminescence imager.

  • Save the initial high-resolution image as a .TIFF file to preserve the full dynamic range.

• Image Pre-processing (if necessary):

  • Open the .TIFF file in image analysis software.

  • Crop the image to the relevant blot area.

  • If required, adjust brightness and contrast to visualize bands clearly. Do not use non-linear adjustments. Document all adjustments.

• File Naming:

  • Rename the processed image file according to your lab's established naming convention (e.g., YYYYMMDD_ExperimentID_TargetProtein_SampleID.tiff).

• Image Upload to PidPort:

  • Log in to your PidPort account and navigate to the relevant project and experiment.

  • Locate the "Upload Image" or "Attach File" section.

  • Select the prepared .TIFF file for upload.

  • Add relevant metadata in the provided fields (e.g., primary antibody used, exposure time).

  • Click "Submit" and wait for the confirmation message.

• Troubleshooting:

  • If the upload fails, first check the file size and format against the PidPort Image Upload Specifications.

  • Verify your internet connection.

  • Clear your browser cache or try a different browser.[1][3]

  • If the issue persists, refer to the troubleshooting workflow diagram below.

Image Upload Troubleshooting Workflow

Caption: A flowchart for troubleshooting image upload failures.

References

• 1. Image Upload Keeps Failing - Microsoft Q&A [learn.microsoft.com]
• 2. Troubleshooting upload issues | Podia Help Center [help.podia.com]
• 3. help.pixieset.com [help.pixieset.com]
• 4. Solved: HubSpot Community - Can not upload images to when trying to submit images for App listings - HubSpot Community [community.hubspot.com]

Technical Support Center: Optimizing Whole-Slide Image Quality for Medmain's AI

This guide provides researchers, scientists, and drug development professionals with detailed information on preparing and scanning whole-slide images (WSI) for optimal analysis by Medmain's AI platforms, such as PidPort.[1][2][3] Adherence to these guidelines is crucial for generating high-quality, reliable data.

Frequently Asked Questions (FAQs)

Q1: What are the most critical pre-analytical factors that affect the quality of a whole-slide image for AI analysis?

A: The most critical factors occur before the slide is ever scanned.[4][5] These include proper tissue fixation, standardized tissue processing and thickness, and consistent, high-quality staining.[6][7] Inadequate or delayed fixation can lead to cellular artifacts, while variations in staining intensity can confuse AI algorithms, leading to inaccurate segmentation and analysis.[4][8]

Q2: Why is consistent H&E staining so important, and how can we achieve it?

A: Medmain's AI, like most pathology AI, is trained on vast datasets of H&E images.[1] Significant deviations in color and intensity can degrade the AI's performance.[4] To achieve consistency, it is essential to use standardized, automated staining protocols, regularly monitor reagent quality, and implement quantitative quality control measures.[9][10][11] Using "high-definition" H&E staining protocols adapted for digital imaging can enhance contrast and detail, which is preferable for AI analysis.[10]

Q3: What scanner magnification should be used for Medmain's AI?

A: For most standard histopathology analyses, scanning at ×20 magnification is generally acceptable.[12] However, for applications requiring the resolution of fine cellular details, such as in-situ hybridization or complex cytological analysis, scanning at ×40 magnification is recommended to ensure all necessary information is captured.[12] The choice can be application-dependent, and different magnifications have been shown to affect the performance of machine learning algorithms.[13]

Q4: How do image file formats and compression affect AI analysis?

A: While lossless formats like PNG retain all image data, they result in very large file sizes.[12][14] Many WSI systems use compression formats like JPEG or JPEG 2000 to manage file sizes.[12] Studies have shown no meaningful difference in AI model accuracy when trained on lossy (JPG) versus lossless (PNG) images for certain classification tasks.[14] However, excessive compression can introduce artifacts. It is crucial to use a compression level that balances file size and image integrity.

Q5: Can the AI differentiate between tissue and artifacts?

A: Modern AI models are trained to identify and exclude common artifacts like tissue folds, air bubbles, dust, and ink marks.[6][15][16] However, severe or unusual artifacts can still interfere with analysis.[4] Therefore, the primary goal should always be to produce the cleanest slide possible through best laboratory practices.[17] Automated quality control tools can help flag slides with significant artifacts before they are submitted for analysis.[15]

Troubleshooting Guides

Issue 1: AI Analysis Fails or Returns a "Poor Quality" Error

This often points to pre-analytical or scanning issues that prevent the AI from accurately segmenting and interpreting the tissue.

Potential Cause	Troubleshooting Steps	Recommended Action
Out-of-Focus Regions	Review the WSI in a viewer. Check for large blurry patches. This is a common problem, especially in cytology where the sample is not flat.[18]	Adjust scanner focus points and rescan the slide. For uneven samples, utilize the scanner's Z-stacking capability if available.[4]
Poor Staining	The image appears too pale, too dark, or has inconsistent coloration. This can be due to depleted reagents or incorrect timing.[9][11]	Review and optimize your H&E staining protocol. Replace staining reagents and perform a quality control check with a control slide before rescanning.
Tissue Folds/Tears	Significant folds or tears are visible on the slide, which can be misinterpreted by the AI.[6]	This is a microtomy artifact. The tissue block may need to be re-embedded and re-sectioned by an experienced histotechnologist.
Debris/Contaminants	Dark spots, dust, or other debris are visible on the image, often trapped under the coverslip.[6]	Ensure a clean working environment during slide preparation and coverslipping. If possible, remove the coverslip, clean the slide, and re-coverslip before rescanning.

Issue 2: Inconsistent AI Results Across a Batch of Slides

When the AI provides variable results for slides that should be similar, the issue often lies with a lack of standardization in the workflow.

Potential Cause	Troubleshooting Steps	Recommended Action
Fixation Variability	Review fixation times and protocols for the batch. Delays in fixation or inconsistent durations (>24-48 hours) can alter tissue morphology and staining uptake.[19]	Implement a strict, documented protocol for fixation, ensuring a fixative-to-tissue volume ratio of at least 15-20:1 and a standardized duration (e.g., 6-24 hours).[20][21]
Scanner Calibration Drift	Colors appear different between slides scanned at the beginning and end of a session.	Perform daily color calibration on the slide scanner using a standardized calibration slide. Check for and correct illumination issues that can cause "scanning lines".[18]
Inconsistent Section Thickness	Some sections may be thicker or thinner, affecting stain penetration and cellular overlap.[7]	Use an automated or high-precision microtome to ensure a consistent tissue thickness (typically 3-5 µm) for all slides in the study.

Experimental Protocols

Protocol 1: Optimal Formalin Fixation and Tissue Processing

• Specimen Collection : Immediately place fresh tissue specimens (≤5 mm thickness) into a container with 10% Neutral Buffered Formalin (NBF). The volume of NBF should be at least 15-20 times the volume of the tissue.[21]

• Fixation Duration : Fix tissues for 12-24 hours at room temperature.[20][22] Avoid under-fixation (<6 hours) and over-fixation (>48 hours), as this can negatively impact tissue morphology and antigenicity.[19]

• Dehydration : Process the fixed tissue through a series of graded ethanol (B145695) solutions (e.g., 70%, 95%, 100%) to remove water.

• Clearing : Following dehydration, use a clearing agent like xylene to remove the ethanol.

• Paraffin (B1166041) Infiltration : Infiltrate the tissue with melted paraffin wax in a tissue processor.

• Embedding : Embed the paraffin-infiltrated tissue into a block for sectioning. Ensure the tissue is oriented correctly.

Protocol 2: Standardized H&E Staining for Digital Analysis

This protocol aims to produce crisp, well-differentiated staining with high contrast, ideal for digital imaging.

• Deparaffinization and Rehydration :

  • Immerse slides in Xylene (2 changes, 5-10 minutes each).[8]

  • Immerse in 100% Ethanol (2 changes, 3-5 minutes each).[8]

  • Immerse in 95% Ethanol (3 minutes).[8]

  • Immerse in 70% Ethanol (3 minutes).[8]

  • Rinse gently in running tap water.[8]

• Hematoxylin (B73222) Staining :

  • Immerse in a filtered hematoxylin solution (e.g., Harris's or Mayer's) for 3-5 minutes.[8]

  • Rinse in running tap water.

• Differentiation :

  • Quickly dip slides 1-3 times in 0.5% acid-alcohol to remove excess stain from the cytoplasm.[8]

  • Immediately rinse in running tap water.

• Bluing :

  • Immerse in a bluing agent (e.g., Scott's Tap Water Substitute) for 30-60 seconds until nuclei turn blue.[8]

  • Rinse in running tap water.

• Eosin (B541160) Staining :

  • Immerse in Eosin Y solution for 30 seconds to 2 minutes.[8]

  • Rinse in tap water.

• Dehydration, Clearing, and Mounting :

  • Dehydrate through graded ethanols (95%, 100%).[8]

  • Clear in xylene (2 changes, 5 minutes each).[8]

  • Apply a coverslip using a permanent mounting medium, avoiding air bubbles.[8]

Visualized Workflows

Caption: End-to-end workflow for preparing slides for Medmain AI analysis.

Caption: Troubleshooting decision tree for poor quality WSI analysis results.

References

• 1. Product Details | Medmain Inc. [en.medmain.com]
• 2. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 3. pidport.medmain.com [pidport.medmain.com]
• 4. Whole Slide Images in Artificial Intelligence Applications in Digital Pathology: Challenges and Pitfalls - PMC [pmc.ncbi.nlm.nih.gov]
• 5. turkjpath.org [turkjpath.org]
• 6. documents.cap.org [documents.cap.org]
• 7. Digital pathology and pre-analytics | Sakura Finetek Europe [sakura.eu]
• 8. clyte.tech [clyte.tech]
• 9. How Digital Image Analysis Strengthens H&E Staining Quality Control [nsh.org]
• 10. High-definition hematoxylin and eosin staining in a transition to digital pathology - PMC [pmc.ncbi.nlm.nih.gov]
• 11. Hematoxylin & Eosin (H&E) Staining Intro: Procedures & More [leicabiosystems.com]
• 12. meridian.allenpress.com [meridian.allenpress.com]
• 13. Preanalytic variable effects on segmentation and quantification machine learning algorithms for amyloid-β analyses on digitized human brain slides - PMC [pmc.ncbi.nlm.nih.gov]
• 14. Impact of pre-analytic variables on deep learning accuracy in histopathology - PMC [pmc.ncbi.nlm.nih.gov]
• 15. How Automated Quality Control Of Whole Slide Images Can Increase The Efficiency Of Clinical And Research Pathology Workflows – Pathology News [pathologynews.com]
• 16. researchgate.net [researchgate.net]
• 17. mdpi.com [mdpi.com]
• 18. youtube.com [youtube.com]
• 19. Effect of formalin tissue fixation and processing on immunohistochemistry - PubMed [pubmed.ncbi.nlm.nih.gov]
• 20. Tissue Processing Overview: Steps & Techniques for Histopathology [leicabiosystems.com]
• 21. ihisto.io [ihisto.io]
• 22. lidoc.paginas.ufsc.br [lidoc.paginas.ufsc.br]

Medmain Digital Pathology Implementation: Technical Support Center

This technical support center provides troubleshooting guidance and frequently asked questions to assist researchers, scientists, and drug development professionals in successfully implementing and utilizing Medmain's digital pathology solutions.

Frequently Asked Questions (FAQs)

General

• Q1: What are the primary benefits of implementing a cloud-based digital pathology system like Medmain's PidPort? A1: Medmain's PidPort offers several advantages, including secure cloud-based management and rapid search of pathology image data and case information. It facilitates seamless collaboration and consultations between departments and across multiple facilities, eliminating the need to transport physical slides. The platform is accessible from a PC or tablet without the need for application installation, reducing initial setup costs and providing flexibility to access data from anywhere with an internet connection.[1]

• Q2: Is Medmain's PidPort compliant with regulatory standards? A2: When implementing any digital pathology solution, it is crucial to ensure compliance with relevant regulatory requirements such as HIPAA, HITECH, CLIA, and CAP guidelines.[2] For specific details on Medmain's compliance, please refer to our official documentation or contact a support representative.

Image Acquisition & Uploading

• Q3: What are the common causes of whole slide image (WSI) scan failures? A3: Scan failures can arise from several factors, including issues with the glass slide preparation, such as "champagne" bubbling from mounting media, dirt, dust, or markings on the slide.[3] Other causes can be machine errors, incompatible slide dimensions, or broken slides.[3]

• Q4: What image file formats are compatible with the PidPort platform? A4: While many whole-slide scanners use proprietary formats, some viewers are vendor-neutral and can support multiple image formats.[4][5] For a comprehensive list of supported file formats for PidPort, please consult our technical specification sheet.

• Q5: How large are whole slide image files and how does this impact upload times? A5: The file size of a whole slide image can be substantial, often exceeding 1 GB per slide, depending on the scanning resolution (e.g., 20x or 40x).[5][6] Upload speed is therefore highly dependent on your institution's network bandwidth.

Image Viewing & Analysis

• Q6: I am experiencing lag when viewing or panning whole slide images. What can I do? A6: Latency in viewing digital slides is a common issue and can disrupt workflow. This is often related to network and IT infrastructure. A stable, high-speed internet connection is crucial for the smooth operation of a cloud-based digital pathology system.[7] Ensure your network meets the recommended specifications for optimal performance.

• Q7: Can I perform annotations and share them with collaborators? A7: Yes, PidPort provides a range of annotation tools that allow you to draw and write on the images. These annotations can be shared with other users to facilitate collaboration and discussion.[8]

• Q8: Does the platform support Z-stacking for viewing images at different focal planes? A8: Many modern whole-slide scanners support the capture of multiple focal planes (Z-stacks). However, this can significantly increase file size, and the image viewer must support this functionality.[4][5] Please check the PidPort user manual for details on Z-stack viewing capabilities.

Integration & Workflow

• Q9: Can PidPort be integrated with our existing Laboratory Information System (LIS)? A9: Integration with an LIS is a critical component of a seamless digital pathology workflow.[9] PidPort can be integrated with LIS by including a URL link of the cases within the electronic medical records.[1] For deeper, system-level integration, please contact our technical support team to discuss your specific requirements.

• Q10: What are the key challenges when integrating a digital pathology system into an existing workflow? A10: The main challenges include ensuring data compatibility between the digital pathology platform and the LIS for smooth data exchange of patient information, images, and reports.[10] It is also important that the integrated system has an intuitive user interface to avoid disrupting daily operations.[10]

Troubleshooting Guides

Issue: Image Upload Failure

This guide will help you troubleshoot common issues that may arise during the upload of whole slide images to the Medmain PidPort platform.

Step	Action	Expected Outcome	If Unsuccessful
1	Verify File Format	The file format of your whole slide image is on the list of supported formats for PidPort.	Convert the image to a supported format using a file conversion utility or re-scan the slide in a compatible format.
2	Check File Size and Integrity	The file size is within the acceptable limits and the file is not corrupted.	If the file is excessively large, consider re-scanning at a lower resolution if appropriate for the analysis. If corruption is suspected, try re-scanning the slide.
3	Assess Network Connection	Your internet connection is stable and has sufficient upload bandwidth.	A typical 50Mbps fiber-optic connection should be able to transfer a 1GB slide in about 3 minutes.[11] If your connection is significantly slower, contact your IT department to discuss potential upgrades.
4	Review Slide Preparation	The physical glass slide is clean, free of debris, and properly coverslipped.	Poor slide preparation can lead to scanning errors.[3] Re-prepare the slide if necessary and re-scan.

Issue: Poor Image Quality in Viewer

If you are experiencing blurry, out-of-focus, or artifact-ridden images in the PidPort viewer, follow these steps.

Step	Action	Expected Outcome	If Unsuccessful
1	Check Original Scan Quality	The original scanned image file is of high quality when viewed with a local image viewer.	If the original scan is of poor quality, the issue lies with the scanning process. Common scanning errors include out-of-focus areas and stitching artifacts.[3] Re-scan the slide, ensuring the scanner is properly calibrated.
2	Adjust Viewer Settings	The image quality is satisfactory after adjusting brightness, contrast, and color balance within the PidPort viewer.	If adjustments do not resolve the issue, proceed to the next step.
3	Clear Browser Cache	Clearing your web browser's cache and reloading the image resolves any temporary display issues.	If the problem persists, try accessing the image from a different computer or network to rule out local hardware or network issues.
4	Report the Issue	If the above steps do not resolve the problem, report the issue to Medmain technical support, providing the case ID and a description of the image quality problem.	Our support team will investigate the issue with the specific image file on our servers.

Quantitative Data Summary

The following tables provide illustrative data on common digital pathology implementation challenges. Please note that this data is hypothetical and intended for demonstrative purposes.

Table 1: Reported Image Upload Errors and Their Frequencies

Error Type	Frequency (%)	Common Cause
Unsupported File Format	35	User attempts to upload a file type not supported by the platform.
File Corrupt	25	Incomplete file transfer or issue during the scanning process.
Network Timeout	20	Unstable or slow internet connection.
Exceeds File Size Limit	15	Image scanned at an unnecessarily high resolution.
Other	5	Miscellaneous errors.

Table 2: Average Time to Resolve Common Technical Support Tickets

Issue Category	Average Resolution Time (Hours)
User Account & Login	1.5
Image Upload & Processing	4.8
Image Viewing & Annotation	3.2
LIS Integration	24.5
General Inquiry	2.0

Experimental Protocols

Protocol: Standard Operating Procedure for Whole Slide Image Upload to PidPort

• Objective: To ensure consistent and successful uploading of whole slide images to the Medmain PidPort platform.

• Scope: This protocol applies to all laboratory personnel responsible for digitizing and uploading pathology slides.

• Materials:

  • Calibrated whole slide scanner

  • Computer with stable internet connection

  • PidPort user credentials

• Procedure:

  • Slide Preparation: Ensure the glass slide is clean, properly labeled with a barcode, and free from artifacts such as dust, mounting media residue, and air bubbles under the coverslip.

  • Scanning:

    • Select the appropriate scanning protocol (e.g., 20x or 40x magnification) based on the diagnostic or research requirements.

    • Initiate the scan and monitor for any hardware or software errors during the process.

    • Upon completion, perform a preliminary quality check of the scanned image thumbnail for completeness and clarity.

  • File Naming and Location:

    • Save the scanned image file using a standardized naming convention that includes the case ID and slide number.

    • Ensure the file is saved in a designated network location for upload.

  • Uploading to PidPort:

    • Log in to your Medmain PidPort account.

    • Navigate to the appropriate case or create a new case.

    • Use the bulk upload function to select and upload the whole slide image file(s).[8]

    • Monitor the upload progress and verify that the file successfully uploads without errors.

  • Post-Upload Verification:

    • Once the upload is complete, open the image in the PidPort viewer to confirm that it displays correctly and that the image quality is acceptable.

    • Attach any relevant case documents or reports to the case in PidPort.[8]

Visualizations

Caption: Troubleshooting workflow for image upload failures.

Caption: A typical workflow for remote consultation using a cloud-based platform.

References

• 1. Object moved [documents.philips.com]
• 2. Digital Pathology: Considerations for a Successful Integration [gestaltdiagnostics.com]
• 3. Types and frequency of whole slide imaging scan failures in a clinical high throughput digital pathology scanning laboratory - PMC [pmc.ncbi.nlm.nih.gov]
• 4. FAQs [digitalpathologyassociation.org]
• 5. midwestmicroscopes.com [midwestmicroscopes.com]
• 6. Digital Pathology FAQs | Tissuepathology.com [tissuepathology.com]
• 7. tiger-technology.com [tiger-technology.com]
• 8. 9 Key Questions About Whole Slide Imaging [grundium.com]
• 9. Rethinking the Role of the LIS in Digital Pathology Adoption – Pathology News [pathologynews.com]
• 10. labos.co [labos.co]
• 11. proscia.com [proscia.com]

PidPort Technical Support Center: Best Practices & Troubleshooting

Welcome to the PidPort Technical Support Center. This guide provides best practices for data management and troubleshooting for common issues encountered during your research and diagnostic workflows.

Frequently Asked Questions (FAQs)

Q1: What is PidPort and what are its primary functions?

Q2: What are the best practices for organizing my data in PidPort?

Effective data organization is crucial for efficient workflow and collaboration. Here are some best practices:

• Establish a Clear Naming Convention: Use a consistent and descriptive naming convention for your files and folders.[5] Incorporating details like project name, date, and sample ID can significantly improve data retrieval.

• Use a Logical Folder Structure: Create a hierarchical folder structure that reflects your projects or studies. This helps in keeping your data organized and easily accessible.[6]

• Leverage Metadata and Tags: Utilize PidPort's tagging and metadata features to add descriptive information to your cases.[3] This will make your data more discoverable and understandable for yourself and collaborators.[6]

• Regular Data Audits: Periodically review your data to ensure it is up-to-date and relevant.[7] This practice helps in maintaining a clean and efficient database.

Q3: What security measures does PidPort have in place to protect my data?

PidPort employs robust security measures to safeguard your data. Data is stored on Amazon Web Services (AWS), and Auth0 is used for authentication.[3] All communication is encrypted, and measures are in place to prevent unauthorized logins, with continuous monitoring for detection and tracking.[3]

Troubleshooting Guides

Data Upload and Import

Q4: I'm having trouble uploading my WSI files. What are the common causes and solutions?

Upload failures can be frustrating. Here are some common causes and how to resolve them:

• File Size Limitations: Large file sizes can sometimes lead to upload failures.[8] Check if your file exceeds the recommended size limits in the table below. If so, consider breaking the file into smaller segments if possible.

• Incorrect File Format: Ensure your file is in a supported format. PidPort is optimized for WSI, but compatibility with specific scanner outputs can vary.

• Network Instability: A poor or unstable internet connection can interrupt the upload process. Ensure you have a stable connection before initiating large uploads.

• Mismatched Headers or Metadata: If your upload includes a metadata file (e.g., CSV), ensure the headers match the expected format in PidPort.[9]

Table 1: Recommended WSI Upload Specifications

Parameter	Recommendation	Notes
File Formats	.svs, .tiff, .vsi	Check PidPort documentation for a full list of supported formats.
Max File Size	2 GB	For larger files, a high-speed, stable internet connection is crucial.
Metadata File	.csv, .xlsx	Ensure column headers match PidPort's template for seamless import.

Image Viewing and Analysis

Q5: The WSI viewer is slow or unresponsive. How can I improve its performance?

Viewer performance can be affected by several factors:

• Browser and Hardware: Ensure you are using a recommended web browser (e.g., latest version of Chrome or Firefox) and that your computer has sufficient RAM and a dedicated graphics card.

• Clear Browser Cache: An overloaded browser cache can slow down performance. Regularly clear your cache and cookies.

• Network Bandwidth: High-resolution WSI streaming requires significant bandwidth. A slow internet connection will impact viewer performance.

Q6: The AI analysis module returned an error or unexpected results. What should I do?

If you encounter issues with the AI analysis:

• Verify Image Quality: Poor image quality, artifacts, or out-of-focus regions can affect AI performance. Ensure the uploaded WSI is of high quality.

• Check Analysis Parameters: Double-check the parameters you've selected for the analysis. Incorrect settings can lead to erroneous results.

• Consult the Documentation: Refer to the specific documentation for the AI module you are using for any known limitations or specific requirements.

• Contact Support: If the issue persists, contact PidPort support with the case ID and a description of the problem.

Experimental Protocol: Digital Pathology Workflow for Tumor Tissue Analysis

This protocol outlines a standard workflow for preparing and analyzing a tumor tissue sample using digital pathology and PidPort.

1. Sample Preparation and Staining:

• Excise tumor tissue and fix in 10% neutral buffered formalin for 24-48 hours.
• Process the tissue through graded alcohols and xylene, and embed in paraffin (B1166041) wax.
• Cut 4-5 µm sections using a microtome and mount on glass slides.
• Perform Hematoxylin and Eosin (H&E) staining to visualize cellular morphology.

2. Whole Slide Imaging (WSI):

• Scan the stained slides at 40x magnification using a high-resolution WSI scanner.
• Export the digital image in a compatible format (e.g., .svs).

3. Data Upload to PidPort:

• Log in to your PidPort account and create a new project.
• Upload the WSI file, along with a metadata file containing relevant case information (e.g., patient ID, tumor type, date of collection).

4. Image Analysis in PidPort:

• Use the PidPort viewer to perform a quality check of the digital slide.
• Utilize the built-in annotation tools to mark regions of interest (ROIs), such as the tumor margin or areas of high cellularity.
• Run the appropriate AI analysis module (e.g., tumor detection, cell counting) on the selected ROIs.

5. Data Review and Collaboration:

• Review the analysis results generated by the AI.
• Use PidPort's sharing features to securely share the case with colleagues for a second opinion or collaborative review.

Visualizations

Caption: Recommended data management workflow in PidPort.

Caption: Troubleshooting decision tree for image upload errors.

Caption: Simplified EGFR signaling pathway, often analyzed in cancer pathology.

References

• 1. PidPort AI Diagnostics | AlfaDiag [alfadiag.com]
• 2. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 3. pidport.medmain.com [pidport.medmain.com]
• 4. google.com [google.com]
• 5. scinote.net [scinote.net]
• 6. mylaminin.net [mylaminin.net]
• 7. wiley.com [wiley.com]
• 8. flatfile.com [flatfile.com]
• 9. 5 CSV File Import Errors (and How to Fix Them Quickly) [ingestro.com]

How to resolve discrepancies in Medmain's AI analysis

Frequently Asked Questions (FAQs)

Q1: What are the most common reasons for discrepancies between the AI analysis and our internal pathologist's assessment?

A: Discrepancies can arise from several factors. The most common include:

• Image Quality: Poor image resolution, artifacts from slide preparation, or improper scanning can affect the AI's ability to accurately identify features.

• Data Bias: The AI model may have been trained on a dataset that differs significantly from your specific sample population, leading to variations in interpretation.[1][2][3][4]

• Low-Prevalence Features: The AI may not be optimized to detect rare or novel features that a trained pathologist might identify.

• Subjectivity in Pathological Interpretation: Pathology can have subjective elements, and different experts may have slightly different interpretations of the same slide.[5]

Q2: How does Medmain's AI handle novel or rare cell morphologies?

A: Medmain's AI models are trained on vast datasets to recognize a wide range of cellular features.[6][7] However, for truly novel or exceptionally rare morphologies not represented in the training data, the AI may classify them as "unknown" or assign a low confidence score. In such cases, manual review by a qualified pathologist is crucial.

Q3: What is the confidence score, and how should it be interpreted?

A: The confidence score represents the AI's level of certainty in its analysis. A high confidence score indicates that the features detected closely match the patterns the AI was trained on. A low confidence score may suggest an ambiguous or unusual feature, warranting closer inspection.

Q4: Can we retrain or fine-tune the Medmain AI model with our own data?

A: Medmain is exploring options for collaborative model training and validation with partner institutions. Please contact our research collaborations team to discuss potential projects for creating custom-trained models to enhance performance on your specific datasets.

Troubleshooting Guides

Issue 1: AI Fails to Identify Clearly Visible Pathological Features

If the Medmain AI analysis fails to identify features that are clearly visible to a trained professional, follow these steps to troubleshoot the issue.

Experimental Protocol:

• Image Quality Assessment:

  • Open the whole-slide image (WSI) in the PidPort™ viewer.

  • Zoom in to the region of interest (ROI) at maximum magnification.

  • Check for common image quality issues such as blurriness, artifacts (e.g., dust, air bubbles), and inconsistent staining.

• Metadata Verification:

  • Ensure that all case and sample metadata (e.g., tissue type, staining protocol) are correctly entered in PidPort™. Incorrect metadata can lead to the application of an inappropriate AI model.

• AI Model Selection:

  • Verify that the correct AI analysis module was selected for the specific tissue and suspected pathology. Medmain offers different AI models for various tissues and diseases.[8]

• Confidence Score Review:

  • Examine the confidence score for the AI's classification of the region. A very low score may indicate the AI had difficulty interpreting the features.

• Comparison with Control Slides:

  • Analyze a set of control slides with known positive and negative features to ensure the AI is performing as expected on validated samples.

Data Presentation:

Troubleshooting Step	Parameter to Check	Acceptable Range/Value	Unacceptable Range/Value
1. Image Quality	Resolution	> 0.5 µm/pixel	< 0.5 µm/pixel
Artifacts	None visible at 40x	Presence of dust, bubbles, folds
2. Metadata	Tissue Type	Matches sample	Mismatch
Staining Protocol	Matches lab protocol	Mismatch
3. AI Model	Selected Model	Appropriate for tissue	Inappropriate for tissue
4. Confidence Score	AI Certainty	> 0.90	< 0.90

Troubleshooting Workflow:

Caption: Workflow for troubleshooting missed pathological features.

Issue 2: AI Flags False Positives in Healthy Tissue

When the AI incorrectly identifies pathological features in what is considered healthy tissue, use the following guide to investigate.

Experimental Protocol:

• Define Region of Interest (ROI):

  • Clearly annotate the region flagged by the AI as a false positive.

• Expert Pathologist Review:

  • Have a senior pathologist review the annotated ROI to confirm it is indeed non-pathological.

• Staining and Artifact Analysis:

  • Examine the ROI for any staining irregularities or preparation artifacts that might mimic pathological features.

• Sub-population Analysis:

  • If your samples come from a specific sub-population (e.g., a particular genetic background or treatment group), consider if this group exhibits unique cellular morphologies that the AI may not have been trained on.

• Report to Medmain:

  • Utilize the "Report Discrepancy" feature in PidPort™, providing the case ID, annotated ROI, and your pathologist's findings. This feedback is invaluable for future AI model improvements.

Data Presentation:

Discrepancy Factor	Potential Cause	Recommended Action
Staining	Inconsistent stain uptake	Review and optimize staining protocol
Presence of precipitate	Filter stains before use
Preparation Artifacts	Tissue folding, knife marks	Improve tissue sectioning technique
Air bubbles under coverslip	Refine coverslipping procedure
Sample Characteristics	Unique cellular morphology	Consult with Medmain for model validation
Edge effects from biopsy	Exclude biopsy edges from analysis area

Logical Relationship Diagram:

References

• 1. Reducing misdiagnosis in AI-driven medical diagnostics: a multidimensional framework for technical, ethical, and policy solutions - PMC [pmc.ncbi.nlm.nih.gov]
• 2. escholarship.org [escholarship.org]
• 3. researchgate.net [researchgate.net]
• 4. AI in Pathology: What could possibly go wrong? - PubMed [pubmed.ncbi.nlm.nih.gov]
• 5. youtube.com [youtube.com]
• 6. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 7. 製品情報 | メドメイン株式会社 [medmain.com]
• 8. Medmain inc. [en.medmain.com]

Medmain AI Technical Support Center: Enhancing Accuracy for Specific Tissue Types

Frequently Asked Questions (FAQs)

Q1: My AI model's accuracy is lower than expected for a specific tissue type. What are the initial troubleshooting steps?

A1: Low accuracy can often be traced back to the training data. Start by assessing the quality and quantity of your dataset. Key considerations include:

• Image Quality: Artifacts such as tissue folds, bubbles, and out-of-focus regions can mislead the AI.[5][6] Implement quality control steps to identify and exclude poor-quality images.

Q2: I have a limited number of annotated slides for a rare tissue type. How can I improve my model's performance?

A2: Data scarcity is a common challenge, especially for rare diseases.[7] Several techniques can be employed to mitigate this:

• Transfer Learning: This is a powerful technique where a model pre-trained on a large dataset (like ImageNet) is fine-tuned using your smaller, specific dataset.[8][9][10] This leverages the features learned from the large dataset, often leading to better performance with less data.

• Data Augmentation: Artificially increase the size and diversity of your training set by applying various transformations to your existing images.[11][12][13][14][15] This can include geometric transformations (rotation, flipping) and color augmentations (adjusting brightness, contrast, and stain colors).

• Generative Adversarial Networks (GANs): GANs can generate synthetic histopathology images that are realistic enough to be used for training, effectively expanding your dataset.[11][13][14]

Q3: My model performs well on data from my institution but fails on external datasets. What causes this and how can I fix it?

• Stain Normalization: Apply stain normalization techniques to standardize the color appearance of images from different sources.[15]

• Color Augmentation: During training, randomly alter the color properties of your images. This makes the model more robust to color variations it might encounter in unseen data.[15]

• Multi-center Training: If possible, incorporate data from multiple institutions into your training set. This exposes the model to a wider range of variability.[5][17][18]

Troubleshooting Guides

Guide 1: Addressing Poor Segmentation of Tissue Structures

Problem: The AI model is inaccurately segmenting specific regions of interest within the tissue.

Possible Causes & Solutions:

Cause	Solution
Insufficient Annotation Detail	Re-annotate a subset of images with highly precise, pixel-wise delineations of the target structures.[19]
Poor Contrast in Tissue	Enhance image contrast using pre-processing techniques like histogram equalization.[12]
Complex Morphological Features	Employ more advanced deep learning architectures, such as U-Net, which are specifically designed for image segmentation tasks.

Guide 2: Mitigating Bias and Improving Generalizability

Problem: The AI model shows bias towards certain patient demographics or sample acquisition sites.

Possible Causes & Solutions:

Cause	Solution
Batch Effects	Technical variations in slide preparation can introduce biases.[16] Use data from diverse batches during training and validation.
Unrepresentative Training Data	Ensure the training dataset is balanced and represents the diversity of the patient population.[1]
Overfitting to a Single Domain	Utilize domain adaptation techniques, such as adversarial training, to encourage the model to learn domain-invariant features.[17][18]

Experimental Protocols

Protocol 1: Data Augmentation for Histopathology Images

This protocol outlines a standard procedure for augmenting a dataset of whole slide image (WSI) patches.

• Image Patch Extraction: Tile the WSIs into smaller, manageable patches (e.g., 256x256 or 512x512 pixels).

• Geometric Augmentation: For each patch, apply a series of random transformations:

  • Rotation: Randomly rotate the image by 90, 180, or 270 degrees.

  • Flipping: Randomly flip the image horizontally and/or vertically.

  • Cropping: Apply random resized crops.[12]

• Color Augmentation:

  • Color Jitter: Randomly change the brightness, contrast, saturation, and hue of the image.[12][15]

  • Stain Augmentation: Simulate variations in Hematoxylin and Eosin (H&E) staining by altering the color channels.

• Dataset Splitting: After augmentation, divide the dataset into training, validation, and testing sets, ensuring that patches from the same patient are not split across different sets to avoid data leakage.[20]

Protocol 2: Transfer Learning Workflow

This protocol describes the steps for implementing a transfer learning approach.

• Model Selection: Choose a pre-trained convolutional neural network (CNN) architecture, such as ResNet, VGG, or Inception. Densely connected and residual networks often yield good performance.[8]

• Model Freezing: "Freeze" the weights of the initial layers of the pre-trained model. These layers have learned to recognize general features like edges and textures.

• Classifier Head Replacement: Remove the original final classification layers of the model and replace them with new, randomly initialized layers that are suited to your specific classification task (e.g., number of tissue types).

• Fine-Tuning:

  • Initial Training: Train only the new classification layers on your custom dataset for a few epochs.

  • Full Model Training: Unfreeze some or all of the earlier layers and continue training the entire network with a lower learning rate. This allows the model to adapt the pre-trained features to the specifics of your histopathology images.[10]

Visualizations

Caption: Troubleshooting workflow for low AI model accuracy.

Caption: Experimental workflow for transfer learning.

References

• 1. The Importance of Data in Building an AI Diagnostic – Pathology News [pathologynews.com]
• 2. Opportunities and Obstacles for AI in Pathology – Pathology News [pathologynews.com]
• 3. aiforia.com [aiforia.com]
• 4. 6407092.fs1.hubspotusercontent-na1.net [6407092.fs1.hubspotusercontent-na1.net]
• 5. Preparing Data for Artificial Intelligence in Pathology with Clinical-Grade Performance [mdpi.com]
• 6. Whole Slide Images in Artificial Intelligence Applications in Digital Pathology: Challenges and Pitfalls - Turkish Journal of Pathology [turkjpath.org]
• 7. Artificial Intelligence and Digital Pathology: Challenges and Opportunities - PMC [pmc.ncbi.nlm.nih.gov]
• 8. openaccess.thecvf.com [openaccess.thecvf.com]
• 9. researchgate.net [researchgate.net]
• 10. AI-augmented pathology: the experience of transfer learning and intra-domain data diversity in breast cancer metastasis detection - PMC [pmc.ncbi.nlm.nih.gov]
• 11. openreview.net [openreview.net]
• 12. researchgate.net [researchgate.net]
• 13. mdpi.com [mdpi.com]
• 14. Selective Synthetic Augmentation with HistoGAN for Improved Histopathology Image Classification - PMC [pmc.ncbi.nlm.nih.gov]
• 15. towardsdatascience.com [towardsdatascience.com]
• 16. pixelscientia.com [pixelscientia.com]
• 17. Learning generalizable AI models for multi-center histopathology image classification - PMC [pmc.ncbi.nlm.nih.gov]
• 18. researchgate.net [researchgate.net]
• 19. Scientists train deep-learning models to scrutinize biopsies like a human pathologist | EurekAlert! [eurekalert.org]
• 20. Preparing Data for Artificial Intelligence in Pathology with Clinical-Grade Performance - PMC [pmc.ncbi.nlm.nih.gov]
• 21. What is Validation & How to Validate an AI Image Analysis Solution [digitalpathologyplace.com]
• 22. How to Validate AI Algorithms in… | College of American Pathologists [cap.org]

User tips for navigating the PidPort interface efficiently

Welcome to the PidPort Technical Support Center. Here you will find troubleshooting guides and frequently asked questions (FAQs) designed to help researchers, scientists, and drug development professionals navigate the PidPort interface efficiently and resolve common issues encountered during their experiments.

Frequently Asked Questions (FAQs)

Q1: How can I optimize my data import process for large datasets?

A1: For large datasets, it is recommended to use the "Batch Import" feature. This allows for the asynchronous upload of multiple files, preventing interface freezes and timeouts. Additionally, ensure your data is in one of the recommended formats to expedite parsing. Pre-processing your data to remove irrelevant columns or rows can also significantly improve import speed.[1]

Q2: What is the best way to manage and organize my experimental protocols within PidPort?

A2: PidPort offers a "Protocol Library" where you can store, version, and share your experimental methodologies. To effectively manage your protocols, we recommend the following:

• Use a consistent naming convention: Include the date, experiment type, and a brief descriptor.

• Utilize version control: When you modify a protocol, save it as a new version to maintain a clear audit trail.

• Leverage templates: Create templates for frequently used protocols to ensure consistency across your team.[2]

Q3: How can I ensure the reproducibility of my analysis workflows?

A3: PidPort is designed to facilitate reproducible research.[3] To ensure the reproducibility of your workflows, you should:

• Document every step: Use the built-in annotation feature to add notes and explanations for each analysis stage.

• Save your workflows: Once an analysis pipeline is established, save it as a workflow that can be re-run with new data.

• Track software versions: PidPort automatically logs the versions of all tools and libraries used in your analysis. Be sure to include this information when publishing your results.[1]

Troubleshooting Guides

Issue: My analysis pipeline is running slower than expected.

Potential Causes and Solutions:

Cause	Solution
Inefficient Data Queries	Optimize your data queries by being more specific in your filtering criteria. Avoid querying entire large tables when only a subset of data is needed.
Computational Bottlenecks	Identify which step in your pipeline is the most time-consuming. Consider using a smaller, subsampled dataset to test and optimize this step before running the full analysis.[4]
Outdated Tool Versions	Check if you are using the latest versions of the analysis tools within PidPort. Newer versions often include performance improvements and bug fixes.[1]

Issue: I am seeing unexpected or inconsistent results from my analysis.

Potential Causes and Solutions:

Cause	Solution
Data Quality Issues	Use PidPort's "Data Quality Control" module to check for missing values, outliers, and batch effects. Low-quality input data can lead to erroneous results.[5]
Incorrect Parameter Settings	Review the parameters for each tool in your analysis pipeline. A common mistake is using default parameters that are not suitable for your specific dataset or biological question.[6]
Software or Pipeline Bugs	If you suspect a bug, first try running a built-in test dataset to see if the issue persists. If it does, please report the issue to our support team, including your workflow file and a description of the expected versus actual results.[7]

Experimental Protocols & Workflows

Protocol: Phospho-Proteomics Data Analysis Workflow

This protocol outlines the standard workflow for analyzing phospho-proteomics data within the PidPort interface. The goal is to identify and quantify phosphorylation events and perform downstream analysis to understand signaling pathway activation.

Methodology:

• Data Import: Raw mass spectrometry data files (e.g., .raw, .mzML) are imported into a new project.

• Peptide Identification and Quantification: The "Peptide Pro" tool is used to identify peptides and quantify their abundance. Key parameters include specifying the correct enzyme (e.g., Trypsin), fixed and variable modifications (e.g., Carbamidomethyl (C), Phospho (STY)), and the appropriate protein database.

• Phospho-site Localization: The "Phospho-Site Locator" tool is run to determine the specific location of phosphorylation on the identified peptides with a high degree of confidence (e.g., PTM score > 0.75).

• Normalization: Data is normalized to account for variations in sample loading and instrument performance. A common method is median normalization.

• Statistical Analysis: A t-test or ANOVA is performed to identify statistically significant changes in phosphorylation levels between different experimental conditions.

• Downstream Analysis: Significantly regulated phospho-sites are used for kinase enrichment analysis and pathway analysis to infer the activity of upstream kinases and signaling pathways.

Phospho-Proteomics Data Analysis Workflow in PidPort.

Logical Relationship: Troubleshooting a Failed Analysis Job

This diagram illustrates the logical steps to take when an analysis job fails within the PidPort interface.

Methodology:

• Check Job Status: The first step is to check the status of the job in the "Job Manager." The status will indicate if the job failed, was terminated, or is still running.

• Review Logs: If the job has failed, the most crucial step is to review the error logs. The logs will typically contain specific error messages that can pinpoint the cause of the failure.

• Common Issues: Based on the error message, common issues to investigate include input file errors (e.g., incorrect format), invalid parameters, or insufficient computational resources.

• Resolution:

  • If it's an input file error , validate and re-upload the file.

  • If parameters are invalid , consult the tool's documentation and adjust the settings.

  • If resources are insufficient , try re-running the job on a higher-memory compute node if available.

• Contact Support: If the issue cannot be resolved after following these steps, contact the PidPort support team and provide the job ID and the error logs.

Logical workflow for troubleshooting a failed analysis job.

Signaling Pathway: Simplified MAPK/ERK Pathway

This diagram represents a simplified version of the Mitogen-Activated Protein Kinase (MAPK) signaling pathway, a crucial pathway in cell proliferation, differentiation, and survival, which is often studied using data analyzed in PidPort.

Methodology:

The diagram illustrates the flow of signal transduction from a growth factor receptor to the activation of the transcription factor ERK.

• Ligand Binding: A growth factor binds to a Receptor Tyrosine Kinase (RTK).

• Receptor Dimerization & Autophosphorylation: This binding causes the receptor to dimerize and autophosphorylate.

• Adaptor Protein Recruitment: The phosphorylated receptor recruits adaptor proteins like GRB2.

• GEF Activation: GRB2 recruits and activates a Guanine nucleotide Exchange Factor (GEF) called SOS.

• Ras Activation: SOS activates the small G-protein Ras by promoting the exchange of GDP for GTP.

• Kinase Cascade: Activated Ras initiates a kinase cascade, sequentially activating RAF, MEK, and finally ERK.

• Nuclear Translocation & Gene Expression: Activated ERK translocates to the nucleus to phosphorylate transcription factors, leading to changes in gene expression that promote cell growth and proliferation.

Simplified representation of the MAPK/ERK signaling pathway.

References

• 1. Bioinformatics Pipeline Troubleshooting [meegle.com]
• 2. Rising Scholars - Best Practices for Managing Research Data [risingscholars.net]
• 3. scribd.com [scribd.com]
• 4. 10xgenomics.com [10xgenomics.com]
• 5. blog.omni-inc.com [blog.omni-inc.com]
• 6. Step-by-Step Guide: 52 Common Mistakes in Bioinformatics and How to Avoid Them - Omics tutorials [omicstutorials.com]
• 7. ssg.lancs.ac.uk [ssg.lancs.ac.uk]

Technical support and resources for Medmain users

Welcome to the . This resource is designed to provide researchers, scientists, and drug development professionals with comprehensive troubleshooting guides and frequently asked questions to ensure a seamless experience with our platform.

Frequently Asked Questions (FAQs)

Here are answers to some common questions about using the Medmain platform.

Q1: What are the recommended system requirements for accessing the Medmain platform?

A1: For optimal performance, we recommend using a computer with at least 4GB of RAM and a modern web browser such as the latest versions of Google Chrome or Mozilla Firefox. Internet Explorer and Microsoft Edge are not recommended. The platform is compatible with both Windows and Mac operating systems.[1] A stable, high-speed internet connection is crucial for uploading and viewing large whole-slide images.[2]

Q2: Which whole-slide image file formats are compatible with the Medmain platform?

A2: Our platform supports a variety of common whole-slide image formats, including SVS (Leica), NDPI (Hamamatsu Photonics), JPEG, and TIFF.[1] If you are using a different format, please contact our support team to discuss potential conversion options.

Q3: How is data security and patient privacy managed on the Medmain cloud platform?

A3: We prioritize the security and privacy of your data. Our platform is built on Amazon Web Services (AWS) for robust storage and utilizes Auth0 for secure authentication. All data transmission is encrypted, and we have measures in place to prevent unauthorized access and monitor for any potential threats.[1]

Q4: Can I integrate the Medmain platform with my existing Laboratory Information System (LIS) or Electronic Health Record (EHR)?

A4: Currently, you can easily link to cases on the Medmain platform from your electronic medical records by including the URL.[1] We are actively working on deeper, system-level integrations with major LIS and EHR providers. Please contact our sales team to discuss your specific integration needs.

Q5: What are the key features of the Medmain AI analysis modules?

Troubleshooting Guides

This section provides step-by-step solutions to specific technical issues you may encounter during your experiments.

Issue 1: Whole-Slide Image Upload Failure

Symptoms:

• The image upload process fails to complete.

• You receive an error message such as "Upload Failed" or "Invalid File Format."

Possible Causes:

• The image file format is not supported.

• The file size is excessively large, leading to a timeout.

• An unstable internet connection is interrupting the upload.

• The file may be corrupted.

Resolution Steps:

• Verify File Format: Ensure your whole-slide image is in a supported format (SVS, NDPI, JPEG, or TIFF).[1]

• Check Internet Connection: A stable, high-speed connection is necessary for large file uploads.[2] If your connection is unstable, try uploading at a different time or from a location with a more reliable network.

• Compress File (if applicable): For very large TIFF or JPEG files, consider using a lossless compression method before uploading.

• Check for File Corruption: Try opening the image file on your local machine with a compatible viewer to ensure it is not corrupted.

• Contact Support: If the issue persists, please contact our technical support team and provide the filename, format, and a description of the error message.

Issue 2: Slow Image Viewer Performance

Symptoms:

• Lag when panning or zooming in the image viewer.

• Images take a long time to load or appear pixelated.

Possible Causes:

• Insufficient local computer memory (RAM).

• An outdated web browser.

• Slow internet connection.

• High server load (rare).

Resolution Steps:

• Check System Requirements: Ensure your computer meets the recommended 4GB of RAM.[1]

• Update Your Browser: Make sure you are using the latest version of Google Chrome or Mozilla Firefox.

• Clear Browser Cache: Clearing your browser's cache and cookies can sometimes resolve performance issues.

• Test Internet Speed: Run an internet speed test to confirm your connection meets the requirements for handling large image data.

• Reduce Concurrent Tabs: Close unnecessary browser tabs and applications to free up system resources.

Issue 3: AI Analysis Results Not Displaying

Symptoms:

• The AI analysis completes, but no results or overlays are shown on the image.

• An error message related to the AI analysis is displayed.

Possible Causes:

• The tissue type is not yet supported by our current AI modules.

• Poor image quality (e.g., out-of-focus regions, significant artifacts) may hinder the AI's ability to process the image.[7]

• The specific analysis module was not correctly selected.

Resolution Steps:

• Confirm Supported Tissue Type: Check our documentation to ensure the tissue type in your slide is supported by the selected AI analysis module. Medmain's AI can currently determine tissues of the stomach, colon, breast, lung, and pancreas, as well as cells of the uterine cervix.[8]

• Assess Image Quality: Review the whole-slide image for any quality issues such as blurriness or significant staining artifacts that might interfere with the analysis.

• Re-run Analysis: Double-check that you have selected the correct AI module for your intended analysis and try running the process again.

• Contact Support: If the problem persists, please provide the case ID and a screenshot of the issue to our support team for further investigation.

Experimental Protocols

Below is a detailed methodology for a common experiment utilizing the Medmain platform.

Experiment: Quantification of Immune Cell Infiltration in Tumor Microenvironment

Methodology:

• Slide Preparation: Formalin-fixed, paraffin-embedded (FFPE) breast cancer tissue sections (4 µm thick) are stained with an anti-CD8 antibody using standard immunohistochemistry (IHC) protocols.

• Whole-Slide Imaging: The stained slides are digitized at 40x magnification using a compatible whole-slide scanner to generate SVS files.

• Data Upload: The SVS files are uploaded to the Medmain cloud platform. Associated case information, such as patient de-identified ID and diagnosis, is also entered.

• Tumor Annotation: A pathologist manually annotates the tumor regions on the whole-slide images using the platform's drawing tools.

• Data Export: The quantitative results, including the total number of CD8+ cells and the area of the annotated region, are exported from the platform as a CSV file.

• Data Analysis: The density of CD8+ cells (cells/mm²) is calculated for each sample.

Quantitative Data Summary

The following table presents sample data from the experiment described above.

Sample ID	Tumor Area (mm²)	CD8+ Cell Count	CD8+ Cell Density (cells/mm²)
BC-001	12.5	6,250	500
BC-002	15.2	9,880	650
BC-003	9.8	3,430	350
BC-004	18.1	14,480	800

Visualizations

Signaling Pathway Diagram

References

• 1. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 2. Key Issues and Answers Regarding Whole Slide Imaging - Wuhan Landing Intelligence Medical Co., Ltd. [landingmed.com]
• 3. 製品情報 | メドメイン株式会社 [medmain.com]
• 4. Medmain inc. [en.medmain.com]
• 5. Medmain inc. [en.medmain.com]
• 6. Behind the Scenes of Building an AI Algorithm in Digital Pathology – Pathology News [pathologynews.com]
• 7. precipoint.com [precipoint.com]
• 8. pidport.medmain.com [pidport.medmain.com]

Medmain Cloud Platform: Data Security & Privacy Technical Support Center

This technical support center provides researchers, scientists, and drug development professionals with essential information regarding the data security and privacy measures implemented on Medmain's cloud platform, PidPort.

Frequently Asked Questions (FAQs)

General Security

Q1: What are the foundational security measures Medmain has in place for its cloud platform?

A1: Medmain's PidPort platform is built on a secure infrastructure, utilizing Amazon Web Services (AWS) for storage and Auth0 for authentication. All communications with the platform are encrypted to prevent unauthorized access and the system is continuously monitored to detect and track any suspicious activities.[1][2]

Data Encryption & Storage

Q2: How is my research data protected while it is being uploaded to, stored on, and downloaded from the Medmain platform?

A2: Data is protected through encryption both when it is being transferred (in transit) and when it is stored on the platform (at rest).[3][4] While specific encryption standards for PidPort are not publicly detailed, industry best practices for platforms handling sensitive health information recommend strong encryption protocols like AES-256 for data at rest and TLS for data in transit.[5]

Q3: Where is my data physically stored?

A3: Medmain utilizes Amazon Web Services (AWS) for its cloud storage infrastructure.[2] AWS provides a secure and compliant environment for sensitive data, although the specific geographic location of the servers may vary. For details on data residency, it is recommended to consult your service agreement with Medmain.

Access Control & Authentication

Q4: How do you ensure that only authorized individuals can access our research data?

A4: The platform uses Auth0 for its authentication function, which is a robust identity and access management (IAM) solution.[2] This allows for secure user authentication and helps prevent unauthorized logins. It is crucial for users to maintain the confidentiality of their login credentials.

Q5: Can I implement Multi-Factor Authentication (MFA) for my account?

A5: While the available documentation does not explicitly state the availability of MFA, it is a standard feature in modern authentication services like Auth0 and a highly recommended security practice for protecting sensitive data.[4][6] Users should check their account settings or contact Medmain support for instructions on enabling MFA.

Compliance & Privacy

Q6: Is the Medmain platform compliant with regulations like HIPAA and GDPR?

A6: While Medmain's website emphasizes its security measures, it does not make explicit claims of compliance with specific regulations like the Health Insurance Portability and Accountability Act (HIPAA) or the General Data Protection Regulation (GDPR).[1] However, the use of secure infrastructure like AWS, which can be configured to be HIPAA compliant, suggests a commitment to data protection.[3][7] For specific compliance requirements related to your research, it is essential to have a Business Associate Agreement (BAA) in place with Medmain if you are handling Protected Health Information (PHI).[3]

Q7: What are my responsibilities as a researcher to ensure data privacy on the platform?

A7: As a user, you play a critical role in maintaining data security. This includes using strong, unique passwords, enabling MFA if available, being cautious about sharing your login credentials, and adhering to your institution's data governance policies. It is also your responsibility to ensure that you have the necessary ethical approvals and patient consent for the data you upload and share.

Troubleshooting Guides

Issue: I am concerned about the security of a shared link to a dataset.

• Step 1: Review Sharing Settings: When sharing data, review the available settings. If possible, set an expiration date for the link and protect it with a password.

• Step 2: Share with Specific Collaborators: Whenever possible, share data directly with specific, authenticated users on the platform rather than using a general link.

• Step 3: Revoke Access: If you suspect a link has been compromised, revoke access immediately through your account's data sharing management interface.

• Step 4: Contact Support: If you cannot resolve the issue or have further concerns, contact Medmain's technical support.

Issue: I suspect unauthorized access to my account.

• Step 1: Change Your Password Immediately: If you can still log in, change your password to a new, strong, and unique one.

• Step 2: Enable Multi-Factor Authentication (MFA): If you haven't already, enable MFA for an added layer of security.

• Step 3: Review Account Activity: Check your account's login history and activity logs for any unrecognized actions.

• Step 4: Report the Incident: Immediately report your suspicions to Medmain's security or support team. Provide as much detail as possible, including the date and time of the suspected access.

Data Security and Privacy Overview

Feature	Description
Storage Infrastructure	Utilizes Amazon Web Services (AWS) for secure and scalable cloud storage.[2]
Authentication	Employs Auth0 for robust user authentication and access management.[2]
Data Encryption	Data is encrypted during transmission (in transit) and while stored (at rest).[2][3]
Monitoring	The platform is continuously monitored to detect and prevent unauthorized access.[2]

Visualizing Data Security on Medmain's Platform

Caption: Logical data flow and security layers within the Medmain cloud platform.

Caption: Troubleshooting workflow for reporting a data security or privacy concern.

References

• 1. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 2. pidport.medmain.com [pidport.medmain.com]
• 3. cloudsecurityalliance.org [cloudsecurityalliance.org]
• 4. cymulate.com [cymulate.com]
• 5. jetbase.io [jetbase.io]
• 6. cspm.cleardata.com [cspm.cleardata.com]
• 7. Healthcare Compliance | Healthcare & Life Sciences | AWS [aws.amazon.com]

PidPort Technical Support Center: Managing and Organizing Large Datasets

Welcome to the PidPort Technical Support Center. This guide provides researchers, scientists, and drug development professionals with detailed answers to frequently asked questions and troubleshooting steps for managing and organizing large datasets within the PidPort platform.

Frequently Asked Questions (FAQs)

Getting Started with Large Datasets

Q: What is the best way to structure my project when I'm about to upload a large dataset (e.g., >500GB of Whole Slide Images)?

A: To ensure efficient data management and collaboration from the outset, we recommend establishing a clear project structure. Before uploading, create a Data Management Plan (DMP) that outlines your organizational strategy.[1][2][3] A well-structured project in PidPort should include:

• A consistent folder hierarchy: Group related files logically. For example, you could structure your folders by experiment, date, and data type (e.g., /Project_Alpha/Experiment_01_2025-11-20/Raw_Images/).[4]

• A standardized naming convention for files and folders: A consistent naming scheme helps in quickly finding and sorting files.[4][5][6] Avoid special characters and spaces; use underscores instead.[5] A good practice is to include key information such as project name, experiment type, date, and version number in the filename (e.g., ProjectA_LungCancer_Case001_Scan1_v01.svs).

• A README.md file in the root directory: This file should describe the project, the data contained within, and any important context for future users (including yourself).[5]

Q: Are there any file size limitations I should be aware of before uploading to PidPort?

A: While PidPort is designed to handle large files such as Whole Slide Images (WSI), individual file uploads may be subject to browser or network limitations. For datasets exceeding 100GB, we recommend using the PidPort Command Line Interface (CLI) Uploader for a more robust and resumable upload process. If an upload fails, the CLI tool can resume from where it left off, saving significant time.

Q: What are the FAIR data principles and how can I apply them in PidPort?

A: The FAIR principles are a set of guidelines to make data more reusable and discoverable. They stand for:

• Findable: Data should be easily discoverable by both humans and machines.

• Accessible: Data should be retrievable by authorized users.[2]

• Interoperable: Data should be able to be combined with other datasets and used by different applications.[2]

• Reusable: Data should be well-described so that it can be used in future studies.[2]

Within PidPort, you can apply these principles by:

• Adding rich metadata to your datasets: Describe the experimental conditions, sample preparation, and imaging parameters.

• Using standardized vocabularies and ontologies when describing your data.

• Setting appropriate access permissions to share your data with collaborators or the wider research community.

Data Organization and Management

Q: How can I keep track of different versions of my datasets within a project?

A: PidPort has built-in version control capabilities. When you upload a file with the same name as an existing file, you will be prompted to either overwrite the existing file or save it as a new version. We strongly recommend saving new versions rather than overwriting, as this allows you to track changes and revert to previous versions if needed.[5] Your file naming convention should also include version numbers (e.g., _v01, _v02).[4]

Q: What is metadata and why is it important for my datasets in PidPort?

A: Metadata is "data about data."[5] It provides context and descriptive information about your files. In PidPort, you can add metadata tags to your datasets, including information about the patient (anonymized), sample type, staining protocol, and imaging system used. This is crucial for:

• Data discovery: Rich metadata makes it easier to search for and find specific datasets.

• Reproducibility: It provides the necessary information for other researchers to understand and potentially reproduce your experiments.[1]

• Long-term value: Well-documented data is more likely to be reusable in the future.

Q: Can I create custom metadata fields for my project?

A: Yes, PidPort allows project administrators to define custom metadata schemas tailored to their specific research needs. This ensures that all data uploaded to the project is consistently annotated with the required information.

Troubleshooting Guides

Data Upload and Integrity Issues

Q: My large dataset upload failed. What are the common causes and how can I fix it?

A: Large dataset upload failures can be frustrating. Here are some common causes and their solutions:

Potential Cause	Troubleshooting Steps
Network Interruption	For large uploads, a stable internet connection is crucial. If your connection is unstable, we recommend using the PidPort CLI Uploader, which can resume interrupted uploads.
File Size Limits	While PidPort's storage is scalable, your browser may have upload size limitations. For files larger than 50GB, the PidPort CLI Uploader is the recommended method.[7]
Incorrect File Path or Name	Double-check that the file paths and names in your upload script or command are correct and do not contain any unsupported special characters.
Insufficient Project Storage	Check your project's storage quota to ensure you have enough space for the new dataset. Contact your administrator if you need to increase the quota.

Q: I've uploaded my data, but some files appear to be corrupted. What should I do?

A: Data corruption can occur during transfer. To mitigate this, PidPort automatically performs a checksum verification on all uploaded files. If a file is flagged as corrupted, you will be notified. You can also manually verify file integrity by comparing the checksum of the uploaded file with your local copy. If corruption is confirmed, please re-upload the affected files.

Data Organization and Accessibility

Q: I'm having trouble finding a specific dataset within my project. What are some best practices for data discovery?

A: To improve data findability in PidPort, leverage the platform's search and filtering capabilities. Ensure that you and your team are consistently applying metadata tags to all datasets. You can search by filename, metadata tags, and even annotations within images. A well-defined folder structure and naming convention will also significantly aid in browsing and locating your data.[4][5][6]

A: Access issues are typically related to permissions settings. In PidPort, you can manage permissions at the project, folder, and individual file level. To resolve this:

• Navigate to the dataset .

• Click on the "Share" or "Permissions" button.

• Verify that your collaborator's email address is listed and that they have the appropriate access level (e.g., "Viewer" or "Editor").

• If they are not listed, add them and set the desired permission level.

Experimental Protocols

Protocol: Whole Slide Image (WSI) Staining for PD-L1 Immunohistochemistry

This protocol outlines the methodology for staining tissue samples for the PD-L1 protein, a common biomarker in cancer research.

• Tissue Preparation: Formalin-fixed, paraffin-embedded (FFPE) tissue sections (4-5 µm) are mounted on positively charged glass slides.

• Deparaffinization and Rehydration: Slides are deparaffinized in xylene and rehydrated through a series of graded ethanol (B145695) washes.

• Antigen Retrieval: Heat-induced epitope retrieval (HIER) is performed using a citrate (B86180) buffer (pH 6.0) in a pressure cooker for 20 minutes.

• Staining: Slides are stained with a primary antibody against PD-L1 (e.g., clone 22C3) for 60 minutes at room temperature.

• Detection: A polymer-based detection system with a horseradish peroxidase (HRP) enzyme is used, followed by a diaminobenzidine (DAB) chromogen, resulting in a brown precipitate at the site of the target antigen.

• Counterstaining: Slides are counterstained with hematoxylin (B73222) to visualize cell nuclei.

• Dehydration and Mounting: Slides are dehydrated through graded ethanol and xylene washes and mounted with a permanent mounting medium.

• Scanning: Slides are digitized using a high-resolution whole slide scanner at 40x magnification to create the WSI.

Visualizations

Experimental Workflow: From Sample to Analysis in PidPort

The following diagram illustrates the typical workflow for a research project utilizing PidPort for managing and analyzing WSI data.

Caption: A diagram illustrating the experimental workflow from sample preparation to collaborative analysis within PidPort.

Signaling Pathway: Simplified PD-1/PD-L1 Axis

This diagram shows a simplified representation of the PD-1/PD-L1 signaling pathway, which is a key target in cancer immunotherapy.

Caption: A simplified diagram of the PD-1/PD-L1 signaling pathway, a crucial interaction in immuno-oncology.

References

• 1. Rising Scholars - Best Practices for Managing Research Data [risingscholars.net]
• 2. rdm.unimi.it [rdm.unimi.it]
• 3. dawsoncollege.qc.ca [dawsoncollege.qc.ca]
• 4. File organisation, naming and version control | Research Data Management [data.cam.ac.uk]
• 5. Data Organization Best Practices – Research Data Service – U of I Library [researchdataservice.illinois.edu]
• 6. Making sure you're not a bot! [library.ucsd.edu]
• 7. communities.springernature.com [communities.springernature.com]

Validation & Comparative

A Comparative Analysis of Digital Pathology Platforms for Research and Drug Development

An in-depth evaluation of Medmain's PidPort alongside leading alternatives, providing researchers, scientists, and drug development professionals with a guide to navigating the evolving landscape of digital pathology.

Key Platform Features at a Glance

Digital pathology platforms offer a range of functionalities crucial for research and development. These typically include a viewer for digital slides, tools for image analysis, and features for collaboration and data management. The integration of AI for tasks like tumor detection, cell counting, and biomarker quantification is a key differentiator.

Feature	Medmain PidPort	Proscia Concentriq	PathAI AISight	Aiforia Platform	Indica Labs HALO	Visiopharm	Paige Platform
Deployment	Cloud-based[1]	Cloud-based, on-premise[2]	Cloud-based[3]	Cloud-based[4]	On-premise, cloud[5]	On-premise, cloud	On-premise, cloud
Primary Focus	AI-assisted diagnostics, telepathology, data management[6][7]	Enterprise-wide pathology workflows, AI integration[2]	AI-powered research and diagnostics[8][9]	User-friendly AI model creation and deployment[4][10]	Quantitative image analysis for research and clinical use[11]	AI-driven precision pathology software[12]	AI-powered cancer diagnostics and biomarker discovery[13][14][15][16][17]
AI Capabilities	Pre-developed AI for various organs (stomach, colon, breast, lung, pancreas, uterine cervix)[2]	Open platform for third-party and in-house AI, automated QC[2]	AI algorithms for tumor detection, cell quantification, biomarker analysis (e.g., PD-L1)[8][9]	Tools for users to create, validate, and deploy their own deep learning models[4][10]	Trainable AI for tissue classification, object detection, and segmentation[11][18]	Pre-built and customizable AI "apps" for specific analyses[12]	FDA-approved AI for prostate cancer detection, various cancer-specific AI modules[13][14][15][16][17]
Collaboration Tools	Secure case sharing, commenting, and annotations for remote consultation and conferences[1]	Real-time collaboration, case sharing, and integration with LIS/LIMS[2]	Secure data sharing, collaborative annotation, and project management tools[9]	Cloud-based platform for multi-user, real-time collaboration on AI model development and analysis[19]	HALO Link for collaborative image and data management[5]	Tools for sharing and reviewing analysis results	Secure, web-based platform for case review and collaboration
Regulatory Status	Research Use Only in some regions, clinical use in others (details vary)[6]	CE-IVD for Concentriq Dx[20]	AISight Dx is FDA-cleared for primary diagnosis[3][21]	CE-IVD marked AI models for specific cancers[9]	HALO AP is CE-IVD and FDA-cleared for primary diagnosis[22]	CE-IVD marked AI solutions[23]	FDA-approved AI applications (e.g., Paige Prostate)[13][14][15][16][17]
Openness	Primarily a closed system with proprietary AI[6]	Open platform with an API for integration of third-party AI[2]	Supports integration with third-party algorithms[9]	Allows users to build and validate their own AI models[10]	Open platform that integrates with third-party AI solutions[5]	Integrates with other platforms	Supports integration with other systems

Quantitative Performance Data

Direct, head-to-head comparative studies of all digital pathology platforms on standardized datasets are not yet widely available in the public domain. However, some vendors have published performance data for their AI algorithms, particularly those with regulatory approval.

One of the most well-documented examples is Paige Prostate , which has received FDA approval. A pivotal study demonstrated significant accuracy gains when pathologists used the AI software.

Performance Metric	Pathologists (Unassisted)	Pathologists (with Paige Prostate)	Improvement
Sensitivity for Cancer Detection	88.7%	96.6%	+7.9% [13]
Specificity for Cancer Detection	97.3%	98.0%	+0.7% [13]
False Negative Diagnoses Reduction	-	-	70% [13][16]
False Positive Diagnoses Reduction	-	-	24% [16]

While specific quantitative data for Medmain's PidPort's AI performance in a comparative setting is not publicly available, the company has published research on the development of its AI for specific use cases, such as the classification of gastric poorly differentiated adenocarcinoma[24]. The accuracy of these models is typically evaluated against a ground truth established by specialist pathologists.

Experimental Protocols

The validation of AI algorithms in digital pathology is a critical process to ensure their accuracy and reliability. While specific protocols vary between companies, a general framework is typically followed.

A General AI Model Validation Workflow:

References

• 1. communities.springernature.com [communities.springernature.com]
• 2. pidport.medmain.com [pidport.medmain.com]
• 3. hitconsultant.net [hitconsultant.net]
• 4. medicalexpo.com [medicalexpo.com]
• 5. indicalab.com [indicalab.com]
• 6. researchgate.net [researchgate.net]
• 7. Medmain inc. [en.medmain.com]
• 8. drugdiscoverytrends.com [drugdiscoverytrends.com]
• 9. clinicaltrialvanguard.com [clinicaltrialvanguard.com]
• 10. aiforia.com [aiforia.com]
• 11. indicalab.com [indicalab.com]
• 12. We are a leading provider of AI-driven precision pathology software for research and diagnostics | Visiopharm [visiopharm.com]
• 13. pathologyinpractice.com [pathologyinpractice.com]
• 14. An independent assessment of an artificial intelligence system for prostate cancer detection shows strong diagnostic accuracy — Paige.ai [paige.ai]
• 15. FDA clears AI-powered prostate cancer detection software | pharmaphorum [pharmaphorum.com]
• 16. FDA Approves Paige Prostate: AI-Driven Cancer Diagnosis Tool [trinitylifesciences.com]
• 17. Clinical Validation of Artificial Intelligence–Augmented Pathology Diagnosis Demonstrates Significant Gains in Diagnostic Accuracy in Prostate Cancer Detection  — Paige.ai [paige.ai]
• 18. indicalab.com [indicalab.com]
• 19. aiforia.com [aiforia.com]
• 20. Digital Pathology: Transforming Diagnosis in the Digital Age - PubMed [pubmed.ncbi.nlm.nih.gov]
• 21. pathai.com [pathai.com]
• 22. indicalab.com [indicalab.com]
• 23. Publications - Visiopharm · Pathology image analysis software [visiopharm.com]
• 24. Medmain inc. [en.medmain.com]

Navigating the Digital Frontier of Pathology: A Comparative Guide to Medmain's Diagnostic AI

For researchers, scientists, and drug development professionals, the accuracy and reproducibility of artificial intelligence (AI) in diagnostics are paramount. This guide provides an objective comparison of Medmain's "PidPort" diagnostic AI with other leading alternatives, supported by available experimental data. We delve into the performance metrics, experimental protocols, and the critical aspect of reproducibility to offer a comprehensive overview for informed decision-making in this rapidly evolving field.

Quantitative Performance Analysis

To facilitate a clear comparison, the following tables summarize the reported performance metrics of Medmain's PidPort AI and its competitors across various cancer types. It is important to note that these results are from different studies and may not be directly comparable due to variations in datasets and methodologies.

Table 1: Performance of Medmain's PidPort AI in Cancer Diagnosis

Cancer Type	Task	Key Performance Metric(s)	Source
Breast Cancer	Classification of Ductal Carcinoma in Situ (DCIS) and Invasive Ductal Carcinoma (IDC)	ROC AUC up to 0.960 for DCIS and 0.977 for IDC on independent test sets.	[1][2]
Prostate Cancer	Classification of prostate cancer in transurethral resection specimens	High accuracy in classification using deep learning and transfer learning.	[3]
Prostate Cancer	Classification for radical treatment on core needle biopsy images	High accuracy classification.	[4][5]
Multi-Organ	Classification of adenocarcinoma in stomach, colon, lung, breast, and lymph nodes	ROC-AUCs in the range of 0.91 - 0.98 on independent test sets.	[6][7][8][9]
Gastric Cancer	Classification of poorly differentiated adenocarcinoma	Successful classification in endoscopic submucosal dissection whole slide images.	[10]
Urothelial Carcinoma	Classification of suspected lesions in urine cytology	Successful classification of the presence of suspected urothelial carcinoma lesions.

Table 2: Performance of Competitor Diagnostic AI Platforms

Competitor	AI Product	Cancer Type	Task	Key Performance Metric(s)	Source
Deep Bio	DeepDx Prostate	Prostate Cancer	Gleason Grading	Strong concordance with pathologists. Outperformed manual assessments in predicting biochemical progression-free survival.	[1][11]
Deep Bio	DeepDx Prostate	Prostate Cancer	Cancer Detection (Radical Prostatectomy)	Sensitivity: 0.997, Specificity: 0.88.	[12][13]
Deep Bio	DeepDx Prostate	Prostate Cancer	Cancer Detection (Core Needle Biopsy)	Sensitivity: 99%, Specificity: 97%.	[12][13]
Paige	Paige Prostate	Prostate Cancer	Cancer Detection	Improved cancer detection by an average of 7.3% compared to unassisted pathologists. Reduced false negatives by 70% and false positives by 24%.	[2][14][15][16]
PathAI	AIM-HER2	Breast Cancer	HER2 Scoring	Automated digital HER2 scoring with interpretable heatmaps.	[17]
Visiopharm	Qualitopix	General	IHC Staining QA	Quantitative analysis for scoring cell lines-derived, stained and digitized control slides.	[18]

Experimental Protocols and Methodologies

Medmain's PidPort AI

• Multi-Organ Adenocarcinoma Classification:

  • Methodology: A deep learning model was trained using weakly-supervised learning on whole slide images (WSIs) from four medical institutions. The training datasets included biopsy specimens of stomach, colon, lung, and breast without detailed annotations.

  • Validation: The model was evaluated on five independent test sets, including stomach, colon, lung, breast, and lymph node specimens, from five different medical institutions to assess its generalization capabilities. The performance was measured using the receiver operating characteristic area under the curve (ROC-AUC).[6][7][8][9]

• Breast Cancer (DCIS and IDC) Classification:

  • Methodology: Deep learning models were trained to classify biopsy and surgical histopathological WSIs into benign, DCIS, and IDC.

The general workflow for developing and validating Medmain's diagnostic AI can be visualized as follows:

Medmain's AI Development and Validation Workflow.

Competitor AI Platforms

• Deep Bio's DeepDx Prostate:

  • Methodology: The AI algorithm was trained on prostate core needle biopsy (CNB) images from two hospitals in South Korea.

• Paige's Paige Prostate:

  • Methodology: The AI software is designed to identify areas of high concern for cancer on prostate biopsy images.

Reproducibility of Diagnostic AI

A critical aspect of any diagnostic tool is its ability to produce consistent and reliable results. In the context of AI, reproducibility means that an independent team can achieve the same results using the original authors' data and methods.[4][10][19] The field of digital pathology faces challenges in ensuring the reproducibility of deep learning models due to factors such as variations in data preprocessing, staining inconsistencies, and differences in scanner technologies.[4][10][19]

To address these challenges, a structured approach to validation is essential. This includes:

• Internal Validation: Assessing the model's performance on a subset of the data from the same source used for training, often using techniques like cross-validation.[20]

• External Validation: Testing the model on entirely independent datasets from different institutions, using different scanners and patient populations to evaluate its generalizability.[20]

• Transparent Reporting: Detailed documentation of the experimental setup, including data sources, preprocessing steps, model architecture, and training parameters, is crucial for enabling other researchers to reproduce the results.[3][19]

The following diagram illustrates the logical flow for ensuring the reproducibility of a diagnostic AI model:

Logical workflow for achieving AI reproducibility.

Conclusion

Medmain's PidPort AI demonstrates high accuracy across a range of cancer types in various validation studies. The performance metrics, particularly the high ROC-AUC values, indicate a strong potential for clinical utility. Competitors such as Deep Bio and Paige also showcase impressive performance, with Paige Prostate having the notable distinction of FDA authorization, which was based on a study demonstrating improved pathologist accuracy.

For researchers and drug development professionals, the key takeaway is the critical importance of scrutinizing the validation methodologies and the diversity of the datasets used to train and test these AI models. While direct comparative studies are lacking, the available data suggests that multiple powerful AI tools are emerging in the digital pathology landscape. The reproducibility of these models remains a key area for continued research and standardization to ensure their reliable and safe integration into clinical and research workflows. As the field matures, prospective, multi-center validation studies will be crucial in establishing the true real-world performance of these promising technologies.

References

• 1. Deep Bio validates AI's role in prostate cancer prognosis < Device/ICT < Article - KBR [koreabiomed.com]
• 2. FDA Approves Paige Prostate: AI-Driven Cancer Diagnosis Tool [trinitylifesciences.com]
• 3. Reproducibility of deep learning in digital pathology whole slide image analysis [ideas.repec.org]
• 4. Reproducibility of deep learning in digital pathology whole slide image analysis | PLOS Digital Health [journals.plos.org]
• 5. Medmain inc. [en.medmain.com]
• 6. Weakly supervised learning for multi-organ adenocarcinoma classification in whole slide images - PubMed [pubmed.ncbi.nlm.nih.gov]
• 7. Weakly supervised learning for multi-organ adenocarcinoma classification in whole slide images | PLOS One [journals.plos.org]
• 8. Weakly supervised learning for multi-organ adenocarcinoma classification in whole slide images | PLOS One [journals.plos.org]
• 9. Weakly supervised learning for multi-organ adenocarcinoma classification in whole slide images - PMC [pmc.ncbi.nlm.nih.gov]
• 10. Reproducibility of deep learning in digital pathology whole slide image analysis - PMC [pmc.ncbi.nlm.nih.gov]
• 11. Deep Bio’s AI Algorithm Demonstrates Significant Prognostic Advancements in Prostate Cancer Study – Deep Bio [deepbio.co.kr]
• 12. Deep Bio’s AI Model for Prostate Cancer Assessment Validated by Researchers – Deep Bio [deepbio.co.kr]
• 13. Deep Bio’s prostate cancer AI algorithm shows high accuracy in prostate cancer diagnosis: study < Hospital < Article - KBR [koreabiomed.com]
• 14. urologytimes.com [urologytimes.com]
• 15. targetedonc.com [targetedonc.com]
• 16. onclive.com [onclive.com]
• 17. pathai.com [pathai.com]
• 18. Validating an AI-based analytic tool for IHC staining QA: precision studies of the digital pathology pipeline - Visiopharm · Pathology image analysis software [visiopharm.com]
• 19. Reproducibility of deep learning in digital pathology whole slide image analysis - PubMed [pubmed.ncbi.nlm.nih.gov]
• 20. mdpi.com [mdpi.com]

AI-Assisted Pathology: A Comparative Analysis of Medmain's "PidPort" and Alternatives in Diagnostic Concordance

For Immediate Release

In the rapidly evolving landscape of digital pathology, the concordance between artificial intelligence (AI) diagnostic support systems and human pathologists is a critical measure of their efficacy and potential for clinical integration. This guide provides a comparative analysis of studies on the diagnostic concordance of Medmain's "PidPort" AI with pathologist diagnoses, alongside other leading AI pathology platforms, including Paige, PathAI, Ibex Medical Analytics, and Aiforia. The information is intended for researchers, scientists, and drug development professionals to objectively assess the performance of these technologies.

Medmain's PidPort: AI-Powered Pathological Analysis

Medmain's PidPort is a cloud-based digital pathology support system that utilizes AI for the analysis of pathological images. The platform is developed through extensive collaboration with medical institutions, leveraging a vast dataset of hundreds of thousands of pathological specimens. Medmain has developed AI models for various tissues, including the stomach, colon, breast, lung, pancreas, and uterine cervix.

A notable study published in Scientific Reports details the performance of a deep learning model developed by Medmain for detecting adenocarcinoma in endoscopic ultrasound-guided fine-needle aspiration biopsy (EUS-FNA) specimens of the pancreas. The model demonstrated a high degree of accuracy with a sensitivity of 93.02% and a specificity of 97.06%.

Further research published in Technology in Cancer Research & Treatment showcases a Medmain AI model for classifying gastric poorly differentiated adenocarcinoma, which achieved a high ROC-AUC of up to 0.975 in gastric ESD test sets. Another publication in Cancers details the successful development of a Medmain AI to differentiate malignant melanoma in digital dermatopathology histology specimens.

Comparative Performance of AI Pathology Platforms

The following tables summarize the available quantitative data from studies on the concordance of various AI platforms with pathologist diagnoses.

Table 1: Performance of Medmain's PidPort AI

Cancer Type	Publication	Key Performance Metrics
Pancreatic Adenocarcinoma	Scientific Reports	Sensitivity: 93.02%, Specificity: 97.06%
Gastric Poorly Differentiated Adenocarcinoma	Technology in Cancer Research & Treatment	ROC-AUC: up to 0.975
Gastric & Colonic Epithelial Tumors	Scientific Reports	Gastric Adenocarcinoma AUC: up to 0.97, Gastric Adenoma AUC: up to 0.99, Colonic Adenocarcinoma AUC: up to 0.96, Colonic Adenoma AUC: up to 0.99
Malignant Melanoma	Cancers	High accuracy in differentiating from other pigmented skin lesions.

Table 2: Performance of Competing AI Pathology Platforms

AI Platform	Cancer Type	Publication/Study	Key Performance Metrics
Paige	Prostate Cancer	Archives of Pathology & Laboratory Medicine	Increased pathologist sensitivity by 8%, Reduced cancer detection errors by 70%[1]
Prostate Cancer	Independent Validation	Standalone AI Sensitivity: 98.9%, Specificity: 93.3%
Ibex	Breast Cancer	Clinical Breast Cancer	Improved diagnostic accuracy from 97.1% to 100%[2]
Breast Cancer	Nature	Invasive Carcinoma & DCIS Specificity: 93.79%, Sensitivity: 93.20%[3]
Prostate Cancer	Validation Study (Japan)	Adenocarcinoma Detection AUC: 0.988, Gleason Score Differentiation AUC: 0.994[4]
PathAI	Non-Small Cell Lung Cancer (PD-L1)	Modern Pathology	AI identified more PD-L1-positive samples compared to manual scoring[5]
Aiforia	Prostate Cancer (Gleason Grading)	Case Study	AI-assisted vs. visual diagnosis agreement (Cohen's weighted kappa): 0.846, Recall: 96.8%, Precision: 89.8%[6]

Experimental Protocols and Methodologies

A critical aspect of evaluating these AI systems is understanding the design of their validation studies. The following provides an overview of the methodologies employed in some of the key studies cited.

Medmain Experimental Protocol (General Approach)

Medmain's studies typically involve the following workflow:

• Data Acquisition: Whole slide images (WSIs) are collected from multiple institutions.

• Annotation: Pathologists annotate the WSIs to create a ground truth dataset.

• Model Training: Deep learning models, often Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), are trained on a large subset of the annotated data.

• Validation: The trained models are then tested on independent datasets from different institutions to evaluate their generalization capabilities. Performance is measured using metrics like sensitivity, specificity, and Area Under the Curve (AUC).

General workflow for Medmain's AI validation studies.

Paige Prostate Detect Clinical Trial Methodology

The CONFIDENT P trial was a prospective clinical study evaluating the impact of Paige Prostate Detect on pathologists' workflow.

• Patient Allocation: Patients suspected of prostate cancer were allocated to either a control or an intervention arm.

• Control Arm: Pathologists assessed whole-slide images (WSIs) using standard hematoxylin (B73222) and eosin (B541160) (H&E) and immunohistochemistry (IHC) staining.[7]

• Intervention Arm: Pathologists used the Paige Prostate Detect AI algorithm on H&E slides and requested IHC only as needed.[7]

• Outcome Measurement: The primary outcome was the relative risk of IHC use per detected prostate cancer case.[8]

Workflow of the CONFIDENT P trial for Paige Prostate Detect.

Ibex Breast AI Validation Study Protocol

A multi-site, blinded clinical study was conducted to validate the Galen™ Breast AI solution.

• Study Design: A retrospective study using prospectively collected cases from multiple institutions.

• Ground Truth: The ground truth was established by a majority agreement of at least two pathologists who reviewed the cases independently and were blinded to the initial diagnosis.[9]

• AI Analysis: Whole slide images were processed by the Galen™ Breast algorithm.

Validation workflow for the Ibex Breast AI solution.

Conclusion

References

• 1. Medmain inc. [en.medmain.com]
• 2. businesswire.com [businesswire.com]
• 3. mdpi.com [mdpi.com]
• 4. ibex-ai.com [ibex-ai.com]
• 5. drugdiscoverytrends.com [drugdiscoverytrends.com]
• 6. aiforia.com [aiforia.com]
• 7. Prospective Clinical Implementation of Paige Prostate Detect Artificial Intelligence Assistance in the Detection of Prostate Cancer in Prostate Biopsies: CONFIDENT P Trial Implementation of Artificial Intelligence Assistance in Prostate Cancer Detection - PubMed [pubmed.ncbi.nlm.nih.gov]
• 8. ascopubs.org [ascopubs.org]
• 9. m.youtube.com [m.youtube.com]

Revolutionizing Cancer Diagnostics: A Comparative Analysis of Medmain's AI-Powered Pathology Platform

For Immediate Publication

In the rapidly evolving landscape of oncology, the integration of artificial intelligence (AI) is proving to be a watershed moment for digital pathology. Medmain Inc., with its "PidPort" platform, stands at the forefront of this transformation, offering a suite of deep learning models designed to enhance the accuracy and efficiency of cancer diagnosis across a variety of tissue types. This guide provides an in-depth comparison of Medmain's performance in several key cancer areas against other leading AI pathology solutions, offering valuable insights for researchers, scientists, and drug development professionals.

Executive Summary

Medmain Performance Benchmarks

Medmain has published several studies validating the performance of its AI models. The following tables summarize the reported performance metrics for different cancer types.

Gastric and Colorectal Cancer

Medmain's deep learning models for gastric and colonic epithelial tumors have demonstrated high accuracy in classifying adenocarcinoma, adenoma, and non-neoplastic tissues.

Cancer Type	Task	Performance Metric	Value	Citation
Gastric	Adenocarcinoma Classification	ROC AUC	Up to 0.97	[1]
Gastric	Adenoma Classification	ROC AUC	Up to 0.99	[1]
Colorectal	Adenocarcinoma Classification	ROC AUC	Up to 0.96	[1]
Colorectal	Adenoma Classification	ROC AUC	Up to 0.99	[1]

Breast Cancer

For breast cancer, Medmain has developed models to differentiate between invasive ductal carcinoma (IDC), ductal carcinoma in situ (DCIS), and benign tissue.

Task	Performance Metric	Value	Citation
DCIS Classification	ROC AUC	Up to 0.960
IDC Classification	ROC AUC	Up to 0.977

Lung Cancer

Medmain's AI models for lung cancer focus on the sub-classification of major histological subtypes from transbronchial lung biopsy (TBLB) specimens.

Task	Performance Metric	Value	Citation
Adenocarcinoma, Squamous Cell Carcinoma, Small-Cell Carcinoma, and Non-neoplastic Classification	ROC AUC	≥ 0.94	[2]
Classification on challenging indeterminate cases	ROC AUC	0.99	[3]

Pancreatic Cancer

In pancreatic cancer, Medmain's AI model has shown high accuracy in discriminating pancreatic adenocarcinoma from endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) biopsy specimens.

Task	Performance Metric	Value	Citation
Pancreatic Adenocarcinoma Discrimination	ROC AUC	0.98	[4]
Accuracy	94%	[4]
Sensitivity	0.93	[4]
Specificity	0.97	[4]

Competitive Landscape

Competitor Performance Overview

Company	Cancer Type(s)	Key Performance Highlights
PathAI	Breast Cancer	High concordance with manual pathologist scoring for HER2 (ICC 0.88-0.91).[5]
Paige	Prostate Cancer	FDA-approved Paige Prostate showed a 7.3% improvement in cancer detection on individual slide images.[6][7][8] Achieved 98.9% sensitivity and 93.3% specificity in an FDA study.[7]
Breast Cancer	Paige Breast Lymph Node demonstrated over 98% sensitivity in detecting metastases of any size.[9] Standalone performance showed 92.8% sensitivity and 94.9% specificity.[10]
Aiforia	Breast Cancer	High correlation with expert grading for ER, PR, and Ki67 scoring.[11][12] Ki-67 scoring showed high consistency with other image analysis software.[10]
Lung Cancer	CE-IVD marked AI model for PD-L1 scoring with high correlation to expert visual assessments and up to 46% faster reporting.[2][8]
Visiopharm	Breast Cancer	High concordance with manual HER2 scoring, particularly in identifying HER2-low cases.[7][13]
Lung Cancer	AI workflow for assessing PD-L1 Tumor Proportion Scores (TPS) in NSCLC.[14][15]

Experimental Protocols

To ensure a comprehensive understanding of the presented data, this section details the methodologies employed in the key studies cited.

Medmain's General Experimental Workflow

Medmain's approach typically involves the following steps:

• Data Acquisition: Histopathological slides are obtained from multiple medical institutions.

• Digitization: Glass slides are scanned to create whole-slide images (WSIs).

• Annotation: Pathologists annotate the WSIs to create training datasets.

• Model Training: Deep learning models, often Convolutional Neural Networks (CNNs), are trained on the annotated data.

• Validation: The trained models are validated on independent test sets from different institutions to assess their generalization capabilities.

Experimental workflow for Medmain's AI model development.

Paige's FDA Approval Study for Prostate Cancer

Paige's submission to the FDA for their Paige Prostate tool involved a robust clinical study:

• Study Design: 16 pathologists evaluated 527 digitized prostate biopsy slide images (171 cancerous, 356 benign).

• Methodology: Each pathologist assessed each slide twice: once without AI assistance and once with Paige Prostate.

• Data Diversity: The dataset included slides from over 150 institutions to ensure the model's generalizability.[13][16]

Key Signaling Pathways in Cancer

To provide a deeper biological context, this section visualizes critical signaling pathways implicated in some of the cancers discussed.

EGFR Signaling Pathway in Lung Cancer

The Epidermal Growth Factor Receptor (EGFR) signaling pathway is a key regulator of cell growth and is frequently dysregulated in non-small cell lung cancer (NSCLC).[17][18]

Simplified EGFR signaling pathway in lung cancer.

HER2 Signaling Pathway in Breast Cancer

The Human Epidermal Growth Factor Receptor 2 (HER2) is a key driver in a significant portion of breast cancers. Its signaling promotes cell proliferation and survival.[19][20]

Simplified HER2 signaling pathway in breast cancer.

PI3K/Akt Signaling Pathway

The PI3K/Akt pathway is a crucial intracellular signaling cascade that regulates cell growth, proliferation, and survival, and is often downstream of receptor tyrosine kinases like EGFR and HER2.[14][21]

Simplified PI3K/Akt signaling pathway.

Conclusion

References

• 1. Artificial Intelligence Helps Pathologists Increase Diagnostic Accuracy and Efficiency in the Detection of Breast Cancer Lymph Node Metastases  — Paige.ai [paige.ai]
• 2. itnonline.com [itnonline.com]
• 3. cancernetwork.com [cancernetwork.com]
• 4. Artificial Intelligence Assisted Assessment of HER2 IHC Using Semi- and Fully-Automated Workflows: Rapid Validation and Implementation as a Quality Assurance and Risk Mitigation Check Point - Visiopharm · Pathology image analysis software [visiopharm.com]
• 5. PathAI to Present Machine Learning-based Quality Control Tool for HER2 Testing in Breast Cancer at the American Society of Clinical Oncology Virtual Scientific Program 2021 [prnewswire.com]
• 6. aiforia.com [aiforia.com]
• 7. visiopharm.com [visiopharm.com]
• 8. aiforia.com [aiforia.com]
• 9. Paige introduces new AI software for breast cancer metastases detection [medicaldevice-network.com]
• 10. Artificial Intelligence Helps Pathologists Increase Diagnostic Accuracy and Efficiency in the Detection of Breast Cancer Lymph Node Metastases - PMC [pmc.ncbi.nlm.nih.gov]
• 11. aiforia.com [aiforia.com]
• 12. visiopharm.com [visiopharm.com]
• 13. Inter-Rater Agreement of HER2 and HER2-low Scores in Breast Cancer Between the Visiopharm Digital Image Analysis HER2 Application and a Group of Expert Pathologists - Visiopharm · Pathology image analysis software [visiopharm.com]
• 14. researchgate.net [researchgate.net]
• 15. researchgate.net [researchgate.net]
• 16. aiforia.com [aiforia.com]
• 17. researchgate.net [researchgate.net]
• 18. researchgate.net [researchgate.net]
• 19. Akt/PKB signaling pathway - Wikipedia [en.wikipedia.org]
• 20. Clinical validation of an AI-based pathology tool for scoring of metabolic dysfunction-associated steatohepatitis - PMC [pmc.ncbi.nlm.nih.gov]
• 21. PI3K / Akt Signaling | Cell Signaling Technology [cellsignal.com]

A Comparative Guide to Digital Pathology Solutions: An Independent Review

Core AI Model Performance: A Comparative Analysis

The true measure of a digital pathology solution lies in the accuracy and reliability of its AI algorithms. This section provides a comparative look at the performance of Medmain's DeepDx algorithms against those of its key competitors in various cancer types.

Prostate Cancer Diagnostics

Prostate cancer has been a primary focus for many digital pathology AI developers. Medmain's DeepDx Prostate has been subject to independent validation, allowing for a direct comparison with other leading solutions like Paige Prostate.

Table 1: Performance Comparison of AI in Prostate Cancer Detection

Feature	Medmain DeepDx Prostate	Paige Prostate
Sensitivity	Not explicitly stated in the provided study.	98.9% (slide-level)
Specificity	Not explicitly stated in the provided study.	93.3% (slide-level)
Concordance (vs. Pathologists)	More concordant with expert uropathologists than original pathology reports (kappa/quadratic-weighted kappa = 0.713/0.922 for Grade Groups).[1]	Pathologists' sensitivity increased by 5.7% with AI assistance, with only a 0.8% decrease in specificity.
Key Findings	Outperformed original pathology reports in detecting high-grade Gleason patterns (4/5).[1] When used as an assistive tool, it increased the diagnostic concordance of a general pathologist and significantly reduced examination time.[1]	In an independent validation, the AI system helped to detect focal cancer in a case with no prior definitive diagnosis.

A standardized approach to validation is crucial for comparing the robustness of different AI models.

• Medmain DeepDx Prostate:

  • Dataset: An independent external dataset of 593 whole-slide images of prostate biopsies (130 normal and 463 adenocarcinomas) was used for validation.[1]

  • Reference Standard: The Gleason scores (GSs) and grade groups (GGs) were determined by three expert uropathologists.[1]

  • Methodology: A general pathologist scored the dataset with and without the assistance of the DeepDx AI to evaluate its impact on diagnostic performance and efficiency.[1]

• Paige Prostate:

  • Dataset: An independent study used 619 whole-slide images from 40 prostate biopsy cases.

  • Reference Standard: The original diagnoses from the pathology reports were used for comparison. Discrepancies were evaluated by pathologists.

  • Methodology: The AI output for each whole-slide image was compared with the original pathologist's diagnosis.

Other Malignancies: A Look at the Broader Landscape

Table 2: Performance of Competitor AI Solutions in Various Cancers

Cancer Type	Competitor Solution	Sensitivity	Specificity	Key Findings
Dermatology (Melanoma)	Proscia DermAI	93%	91%	Correctly identified invasive melanoma and melanoma in situ in a prospective study.
Colorectal Cancer	Owkin MSIntuit CRC	96% - 98%	46% - 47%	Accurately rules out nearly 50% of MSS patients while correctly classifying over 96% of MSI patients.
Gastric Cancer	Olympus AI Tool	100%	≥50%	Successfully identified adenocarcinoma in gastric biopsy pathology specimens.[3]

Digital Pathology Platforms: A Feature-Based Comparison

Beyond the individual algorithms, the platform that integrates these tools is critical for workflow efficiency and collaboration. Medmain's PidPort serves as their central platform, competing with established solutions like Proscia's Concentriq and the Paige Platform.

Workflow and Collaboration

The core function of a digital pathology platform is to streamline the pathologist's workflow. This includes image viewing, case management, and facilitating collaboration.

Caption: A comparison of traditional vs. digital pathology workflows.

Table 3: Feature Comparison of Digital Pathology Platforms

Feature	Medmain PidPort	Proscia Concentriq	Paige Platform
Core Functionality	Cloud-based storage, viewing, and sharing of pathology images.[4]	Enterprise pathology platform for clinical and research workflows.	End-to-end solution with LIS interoperability, multi-scanner integration, and case management.
AI Integration	Equipped with independently developed AI for image analysis.[4]	Open platform that integrates Proscia's own AI applications and third-party algorithms.	Integrates Paige's suite of AI applications, including Paige Prostate and Paige Breast.
Collaboration Tools	Telepathology function for remote diagnosis and consultation.[5]	Facilitates remote case review and collaboration.	Enables seamless sharing and consultation on digital slides.
Key Differentiators	Focus on providing a total digital pathology solution from specimen digitization to diagnosis.[2]	Emphasis on an open platform to allow for a broad range of AI applications.	The first company to receive FDA approval for an AI-based pathology product for cancer diagnosis.

Experimental Workflows and Methodologies

Understanding the workflow of how these AI tools are integrated into the diagnostic process is essential for evaluating their real-world applicability.

Conclusion

Medmain's digital pathology solutions, particularly DeepDx Prostate, demonstrate performance comparable to leading competitors in independent studies. The PidPort platform offers a comprehensive suite of tools for transitioning to a digital workflow. However, a notable gap exists in the availability of independent, peer-reviewed performance data for Medmain's AI models for other cancer types, such as gastric, colon, breast, and lung cancer.

For organizations considering the adoption of digital pathology, the choice of platform will depend on specific needs. Medmain's all-in-one approach may be attractive to institutions looking for a complete solution from a single vendor. In contrast, platforms like Proscia's Concentriq offer greater flexibility through an open ecosystem of third-party AI applications. As the field continues to mature, the availability of robust, independent validation data across a wider range of malignancies will be crucial for making informed decisions.

References

• 1. Product Details | Medmain Inc. [en.medmain.com]
• 2. Medmain inc. [en.medmain.com]
• 3. cms.neogenomics.com [cms.neogenomics.com]
• 4. Digital Pathology Cloud Systems｜PidPort [us.medmain.com]
• 5. Medmain inc. [en.medmain.com]

Revolutionizing Drug Development: A Comparative Analysis of Medmain's PidPort in Clinical Pathology

For researchers, scientists, and drug development professionals, the integration of artificial intelligence (AI) in digital pathology is rapidly transforming the landscape of clinical trials and therapeutic research. Medmain's PidPort, a cloud-based digital pathology support system, emerges as a significant contender in this space. This guide provides an objective comparison of PidPort's clinical utility with other leading platforms, supported by available performance data and detailed experimental methodologies.

Performance in AI-Assisted Diagnostics: A Comparative Look

Quantitative validation of AI algorithms is paramount for their adoption in clinical and research settings. While comprehensive head-to-head studies are still emerging, available data from publications and company reports allow for a preliminary comparison of Medmain's PidPort with other prominent digital pathology platforms like Paige and Proscia.

A notable application of PidPort's AI is in the detection of various cancers. Research published in Scientific Reports detailed a deep learning model for detecting pancreatic ductal adenocarcinoma in endoscopic ultrasound-guided fine-needle biopsy specimens. Another study in Technology in Cancer Research & Treatment showcased a model for the classification of gastric signet ring cell carcinoma.[3] The reported performance of these models has been high, with one study achieving a Receiver Operating Characteristic Area Under the Curve (ROC-AUC) of at least 0.99.[3]

In comparison, Paige has published robust clinical validation data for its AI applications, particularly in the field of prostate cancer. Studies have demonstrated that the use of Paige's AI can significantly improve the diagnostic accuracy of pathologists.

The following table summarizes the available quantitative performance data for the AI modules of these platforms.

Platform	AI Application	Cancer Type	Key Performance Metrics	Source
Medmain PidPort	AI Model	Gastric Signet Ring Cell Carcinoma	ROC-AUC: ≥ 0.99	Technology in Cancer Research & Treatment[3]
Proscia Concentriq	DermAI™	Skin Biopsies	Accuracy: 98%	Scientific Reports[4][5]
Proscia Concentriq	AI Model	Melanoma Concordance Prediction	R² Correlation: 0.51	European Conference on Computer Vision 2022[6]

Experimental Protocols: A Closer Look at Validation Methodologies

Understanding the rigor of the experimental protocols used to validate these AI systems is crucial for assessing their clinical utility.

Medmain PidPort: Gastric Signet Ring Cell Carcinoma Detection

The study on gastric signet ring cell carcinoma (SRCC) employed a robust methodology to train and validate its deep learning models.

• Dataset: The training set consisted of 1,765 Whole Slide Images (WSIs), and the models were evaluated on two separate test sets of 999 and 455 WSIs.[3]

• Training Methods: The researchers utilized a combination of transfer learning, fully-supervised learning, and weakly-supervised learning to train the convolutional neural network (CNN) classifier.[3] In the fully-supervised approach, tiles were extracted from annotated regions of interest on the WSIs to train the model.[3] The weakly-supervised method involved an iterative process of inference and training, where the model identified the most probable tumor regions to refine its learning.[3]

• Evaluation: The performance of the best model was assessed using the ROC-AUC metric, which measures the model's ability to distinguish between cancerous and non-cancerous images.[3]

Visualizing the Digital Pathology Workflow

To better understand the practical implementation of these platforms, the following diagrams, generated using Graphviz, illustrate key workflows in digital pathology.

AI-Assisted Digital Pathology Workflow in Clinical Trials

This workflow demonstrates how a platform like PidPort can be integrated into a clinical trial, from patient sample collection to data analysis and patient stratification. The use of AI can potentially accelerate biomarker analysis and lead to more efficient and accurate patient selection.

Teleconsultation Workflow with PidPort

The teleconsultation workflow highlights how PidPort facilitates collaboration between pathologists in different geographical locations. This is particularly beneficial for obtaining second opinions on challenging cases, thereby improving diagnostic accuracy and patient care.

Conclusion

References

• 1. drugtargetreview.com [drugtargetreview.com]
• 2. mdpi.com [mdpi.com]
• 3. researchgate.net [researchgate.net]
• 4. clpmag.com [clpmag.com]
• 5. pathologyinpractice.com [pathologyinpractice.com]
• 6. pathologyinpractice.com [pathologyinpractice.com]

The Digital Microscope: A Comparative Analysis of AI-Powered Pathology Platforms

A Guide for Researchers and Drug Development Professionals

Overview of a Typical AI-Powered Digital Pathology Workflow

A generalized workflow for AI-assisted digital pathology.

Performance Comparison of AI Diagnostic Tools

The following tables summarize key performance metrics from validation studies of AI models for breast and prostate cancer detection. It is crucial to note that these results are not from direct, head-to-head comparative trials and were conducted using different datasets and protocols. Therefore, direct comparison of metrics should be approached with caution.

Table 1: AI Performance in Breast Cancer Diagnosis

Company	AI Model/Application	Key Performance Metric	Value	Test Set Details
Medmain	Invasive Ductal Carcinoma (IDC) Classification	ROC AUC	0.95–0.98	1 Core Needle Biopsy set (n=522) and 3 surgical test sets (n=1129)[1]
Ibex	Galen™ Breast	Accuracy (Microinvasive Carcinoma)	High Accuracy*	Challenging patient cohort from Brigham and Women's Hospital and Champalimaud Foundation

*Specific quantitative data for Ibex's Galen™ Breast microinvasive carcinoma study was not provided in the available resources, but results were reported as "highly accurate".

Table 2: AI Performance in Prostate Cancer Diagnosis

Company	AI Model/Application	Key Performance Metric	Value	Test Set Details
Paige	Paige Prostate Detect	Pathologist Sensitivity (with AI)	96.6% - 96.8%	527-610 prostate needle biopsy WSIs from >150 institutions[2][3][4]
Pathologist Sensitivity (without AI)	88.7% - 89.5%	As above[2][3]
Reduction in False Negatives	70%	As above[2]
Ibex	Galen™ Prostate	Standalone Sensitivity	96.6% - 98.46%	Puerto Rican male cohort & UPMC cohort; prostate core needle biopsies[5][6]
Standalone Specificity	96.7% - 97.33%	As above[5][6]
Standalone AUC	0.991 - 0.994	As above[5][6]

Detailed Experimental Protocols

Medmain: Invasive Ductal Carcinoma (IDC) Classification

Medmain published a study on their deep learning models for classifying invasive ductal carcinoma in whole-slide images (WSIs) of breast tissue.[1]

• Objective: To develop and validate AI models to classify WSIs of breast biopsies as containing IDC or not.

• Methodology: The models were trained using transfer learning and weakly-supervised learning techniques. Weakly-supervised learning allows the model to train on slide-level labels (e.g., "IDC is present in this slide") without requiring pathologists to manually delineate every tumor region, which is a highly time-consuming task.

• Dataset: The trained models were evaluated on four independent test sets:

  • One set of core needle biopsies (n=522).

  • Three sets of surgical specimens (n=1129).

• Ground Truth: The definitive diagnosis for each slide, used as the ground truth for evaluating the AI's performance, was presumably established by expert pathologists.

• Evaluation Metric: The performance was measured by the Area Under the Receiver Operating Characteristic Curve (ROC AUC), a common metric for evaluating the performance of binary classifiers. The models achieved ROC AUCs ranging from 0.95 to 0.98.[1]

Weakly-supervised learning trains on slide-level labels.

Paige: Paige Prostate Detect FDA Authorization Study

• Objective: To evaluate whether assistance from Paige Prostate Detect improves pathologists' accuracy in diagnosing prostate cancer on digital prostate biopsy images.

• • Unassisted Read: The pathologist reviewed the WSI using a standard digital pathology viewer.

  • Assisted Read: After a washout period, the pathologist reviewed the same WSI with the assistance of Paige Prostate, which highlights areas suspicious for cancer.[2]

• Ground Truth: The reference standard was based on the consensus diagnosis of a panel of expert pathologists.

Ibex Medical Analytics: Galen™ Prostate Validation Study

• Objective: To blindly validate the performance of the Galen™ Prostate algorithm for detecting, grading, and sizing prostate cancer in core needle biopsies.

• Dataset: The UPMC study used a set of prostate core needle biopsies. The CorePlus study in Puerto Rico also used prostate core needle biopsies, and its subsequent real-world integration involved over 122,000 slides from 9,200 cases over three years.[6]

• Ground Truth: The ground truth was established by the final, resolved diagnosis from expert pathologists at the respective institutions.

Conclusion

For researchers, scientists, and drug development professionals, the choice of an AI pathology platform will depend on the specific application, the need for regulatory-cleared tools versus research-use-only algorithms, and the desired integration into existing laboratory workflows. While the absence of direct comparative trials makes it difficult to declare a single "best" platform, the data indicates that the field is advancing rapidly, providing powerful new tools to improve the precision and efficiency of pathological analysis. As these technologies continue to evolve, further studies, including direct comparisons and real-world evidence generation, will be critical to fully understand their comparative strengths and optimal use cases.

References

• 1. Medmain inc. [en.medmain.com]
• 2. Clinical Validation of Artificial Intelligence–Augmented Pathology Diagnosis Demonstrates Significant Gains in Diagnostic Accuracy in Prostate Cancer Detection  — Paige.ai [paige.ai]
• 3. Paige receives FDA de novo authorization for AI product for prostate cancer detection | BioWorld [bioworld.com]
• 4. onclive.com [onclive.com]
• 5. medcitynews.com [medcitynews.com]
• 6. Validation and three years of clinical experience in using an artificial intelligence algorithm as a second read system for prostate cancer diagnosis—real-world experience - PMC [pmc.ncbi.nlm.nih.gov]
• 7. The Paige Prostate Suite: Assistive Artificial Intelligence for Prostate Cancer Diagnosis - NCBI Bookshelf [ncbi.nlm.nih.gov]
• 8. ibex-ai.com [ibex-ai.com]

Unveiling the Precision of AI in Pathology: A Comparative Look at Medmain's Technology

Medmain's proprietary AI technology, integrated into its "PidPort" digital pathology platform, has been the subject of scientific validation in several studies. This guide will delve into the quantitative performance and methodologies described in two key publications, while also drawing comparisons with established competitors in the AI pathology space, Paige and PathAI.

Performance Benchmarks: A Quantitative Comparison

To provide a clear overview of the performance of Medmain's AI models, the following table summarizes the key metrics reported in their peer-reviewed publications. For comparative purposes, data from a relevant study by a competitor is also included.

Technology	Indication	Key Performance Metric	Value	Publication
Medmain	Gastric Adenocarcinoma	AUC	0.97	Scientific Reports
Medmain	Gastric Adenoma	AUC	0.99	Scientific Reports
Medmain	Colonic Adenocarcinoma	AUC	0.96	Scientific Reports
Medmain	Colonic Adenoma	AUC	0.99	Scientific Reports
Paige	Colorectal Cancer (Adenocarcinoma vs. other tissues)	Accuracy	98.6%	(Details in cited publication)

Note: A direct comparison of performance metrics should be made with caution due to potential variations in datasets and experimental protocols. AUC (Area Under the Curve) is a measure of a classifier's ability to distinguish between classes.

In-Depth Look: Medmain's Validated AI Models

Medmain has published research demonstrating the capabilities of its deep learning models in identifying cancerous and precancerous lesions in histopathological images.

Gastric and Colonic Epithelial Tumors

A study published in Scientific Reports details the development and validation of deep learning models for the classification of gastric and colonic epithelial tumors.[1] The models were trained to distinguish between adenocarcinoma, adenoma, and non-neoplastic tissue. The high AUC values reported (0.96-0.99) indicate a strong performance in accurately classifying these different tissue types.[1]

Malignant Melanoma Differentiation

In a more recent publication in the journal Cancers, Medmain announced the successful development of an AI model to differentiate malignant melanoma in digital dermatopathology specimens.[2][3] This demonstrates the expansion of their technology to different areas of oncology.

The Competitive Landscape: Paige and PathAI

Paige , for instance, has published extensively on the application of their AI tools in various cancers, including those of the gastrointestinal tract.[4] Their research highlights the potential of AI to not only improve diagnostic accuracy but also to streamline the pathology workflow.

PathAI also has a robust publication record, with studies covering a wide range of cancer types and applications, from diagnostic classification to biomarker quantification. Their work underscores the growing role of AI in personalized medicine.

Experimental Protocols and Methodologies

A detailed understanding of the experimental setup is crucial for evaluating the robustness of any new technology. While the full, in-depth protocols for Medmain's studies are contained within the original publications, a general workflow for the development and validation of such AI models can be outlined.

Below is a generalized workflow representing the typical stages involved in creating and validating a deep learning model for histopathological classification.

A generalized workflow for developing and validating a deep learning model in digital pathology.

This diagram illustrates the key steps, from the initial collection and annotation of whole-slide images by pathologists to the final evaluation of the trained model on an independent test set. This rigorous process is essential for ensuring the reliability and generalizability of the AI algorithm.

Signaling Pathways in AI-Driven Diagnostics

An illustrative diagram of an AI-assisted diagnostic workflow in digital pathology.

This visualization demonstrates how a whole-slide image is first analyzed by the AI model, which in turn provides outputs such as heatmaps of suspicious regions and tissue classifications. These outputs then serve as a valuable aid to the pathologist, who performs the final review and makes the definitive diagnosis. This collaborative approach has the potential to enhance both the efficiency and accuracy of cancer diagnosis.

References

• 1. Lung and Colon Cancer Detection Using a Deep AI Model - PMC [pmc.ncbi.nlm.nih.gov]
• 2. Paige.AI publishes study on clinical-grade pathology - The Cancer Letter [cancerletter.com]
• 3. Paige’s Research Propels AI Utilization in Pathology — Paige.ai [paige.ai]
• 4. Paige.ai [paige.ai]

Safety Operating Guide

Safeguarding Your Laboratory: A Guide to Proper Waste Disposal

The responsible disposal of laboratory waste is paramount to ensuring the safety of personnel, protecting the environment, and maintaining regulatory compliance. This guide provides essential, step-by-step procedures for the proper disposal of waste generated from research, scientific, and drug development activities. Adherence to these protocols is critical for minimizing risks associated with hazardous materials.

Waste Classification and Segregation

Proper waste disposal begins with accurate classification and segregation at the point of generation. Different types of waste require distinct disposal pathways. The following table summarizes the primary categories of laboratory waste and their appropriate containers.

Waste Category	Description	Recommended Container	Disposal Pathway
General Waste	Non-hazardous materials similar to household trash, such as office paper, packaging, and uncontaminated gloves.	Standard trash receptacle	Municipal Landfill
Sharps Waste	Any item that can puncture or cut skin, including needles, syringes, scalpels, and broken glass.[1]	Puncture-resistant, leak-proof sharps container with a biohazard symbol.[1]	Autoclave or Incineration
Infectious Waste	Materials contaminated with blood, bodily fluids, or other potentially infectious agents, such as used culture dishes, swabs, and gloves.[1]	Red biohazard bags placed within a secondary leak-proof container.[1][2]	Autoclave or Incineration
Pathological Waste	Human or animal tissues, organs, and body parts.[1]	Yellow biohazard bags within a secondary leak-proof container.[1]	Incineration
Chemical Waste	Discarded solid, liquid, and gaseous chemicals, including solvents, reagents, and cleaning agents.[1]	Designated, labeled, and chemically compatible waste containers.	Specialized hazardous waste vendor
Pharmaceutical Waste	Expired, unused, or contaminated medications and vaccines.[1]	Blue pharmaceutical waste containers.[1]	Incineration or chemical treatment

Step-by-Step Disposal Protocols

1. Sharps Waste Disposal:

• Immediately after use, place the sharp item into a designated, puncture-resistant sharps container.[1]

• Do not attempt to recap, bend, or break needles.

• Ensure the sharps container is not overfilled; close the container when it is approximately three-quarters full.[1]

• Securely lock the lid of the container.

• Store the sealed container in a designated secure area away from general traffic until it is collected for disposal.[1]

2. Infectious Waste Disposal:

• Segregate infectious waste into red biohazard bags at the point of generation.[1][2]

• Ensure the red bag is placed inside a secondary, rigid container that is leak-proof and has a lid.[2]

• When the bag is three-quarters full, securely tie it closed.[2]

• Close the lid of the secondary container.

• Label the container with the appropriate biohazard symbol and any other required information.

• Store in a designated area for collection by a licensed medical waste disposal service.[1]

3. Chemical Waste Disposal:

• Identify the chemical waste and consult the Safety Data Sheet (SDS) for specific disposal instructions.

• Pour liquid chemical waste into a designated, properly labeled, and chemically compatible waste container. Use a funnel to prevent spills.

• Ensure the container is kept closed when not in use.

• Do not mix incompatible chemicals in the same waste container.

• Store the chemical waste container in a well-ventilated, designated satellite accumulation area.

• Arrange for pickup by a certified hazardous waste disposal company.

Visualizing Disposal Workflows

To further clarify the disposal processes, the following diagrams illustrate the decision-making and procedural steps for waste segregation and sharps disposal.

References

• 1. actenviro.com [actenviro.com]
• 2. 6 Steps to Proper Medical Waste Disposal - AEG Environmental [aegenviro.com]

Understanding "Medmain": A Guide to Safe Practices in Digital Pathology

It is important to clarify that "Medmain" is not a chemical substance requiring a specific safety data sheet for handling and disposal. Rather, Medmain Inc. is a Japanese health-tech company that has developed "PidPort," a digital pathology platform that utilizes artificial intelligence (AI) for pathological image diagnosis. Therefore, this guide will provide essential safety and logistical information relevant to the handling of pathological specimens within a digital pathology workflow, the environment in which a platform like PidPort would be used.

This information is crucial for researchers, scientists, and drug development professionals who work with biological tissues and aim to integrate digital pathology into their workflow. Adherence to these safety protocols is paramount to ensure the well-being of laboratory personnel and the integrity of the specimens.

Personal Protective Equipment (PPE) for Handling Pathological Specimens

The handling of human or animal tissues, organs, and body parts in a laboratory setting necessitates the use of appropriate Personal Protective Equipment (PPE) to prevent exposure to infectious agents and other biological hazards. The selection of PPE is determined by a risk assessment of the procedures being performed.[1] Laboratories handling pathological specimens typically operate at Biosafety Level 2 (BSL-2) or higher, depending on the nature of the agents being studied.[2][3][4]

Below is a summary of recommended PPE for handling pathological specimens:

PPE Component	Description	Purpose
Lab Coat	A dedicated, long-sleeved lab coat. Fluid-resistant options are recommended when working with liquid specimens.[5]	Protects skin and personal clothing from contamination.
Gloves	Disposable, single-use gloves (e.g., nitrile). Double-gloving may be necessary for certain procedures.[1]	Prevents direct contact with potentially infectious materials.
Eye Protection	Safety glasses with side shields, goggles, or a face shield.[1]	Protects the eyes and face from splashes and aerosols.
Face Mask	A surgical mask or a fitted respirator (e.g., N95) may be required depending on the aerosol transmission risk of the pathogens being handled.	Protects the respiratory tract from airborne pathogens.
Closed-toe Shoes	Sturdy, non-slip shoes that fully cover the feet.	Protects the feet from spills and falling objects.

Operational Plan: A Typical Digital Pathology Workflow

The integration of digital pathology streamlines the traditional histopathology workflow.[6][7] This involves the digitization of tissue samples on glass slides, enabling remote analysis and collaboration.[8][9]

A generalized workflow for digital pathology, from specimen collection to data sharing.

Disposal Plan for Pathological Waste

Pathological waste, which includes recognizable human or animal tissues, organs, and body parts, requires special handling and disposal to prevent the spread of infectious diseases and to comply with regulatory standards.[10] This type of waste is categorized as regulated medical waste and must be segregated from other waste streams at the point of generation.[10][11]

Key Disposal Steps:

• Segregation: Pathological waste should be placed in designated, leak-proof containers lined with a red biohazard bag.[12][13] These containers must be clearly labeled as "Pathological Waste" or "Incineration Only".[11][12]

• Containment: To prevent leakage, it is recommended to double-bag waste that is saturated with fluids and to use absorbent liners in the containers.[10][13]

• Storage: Pathological waste should be stored in a secure, refrigerated area to prevent decomposition and odor until it is collected for disposal.[13]

• Treatment: The most common and accepted method for the final disposal of pathological waste is incineration.[10][11] This high-temperature combustion process effectively destroys all infectious agents and reduces the volume of the waste.[10] Autoclaving may be used for some types of non-anatomical biohazardous waste, but incineration is generally required for pathological waste.[13]

• Documentation: Maintain accurate records of the amount of pathological waste generated and the dates of collection and disposal.

It is imperative that all personnel involved in the handling and disposal of pathological waste receive comprehensive training on these procedures and are aware of the specific regulations in their jurisdiction.[10]

References

• 1. Personal Protective Equipment Requirements for Laboratories – Environmental Health and Safety [ehs.ncsu.edu]
• 2. consteril.com [consteril.com]
• 3. microbenotes.com [microbenotes.com]
• 4. Guide to Biosafety Levels (BSL) 1, 2, 3, & 4 | Lab Manager [labmanager.com]
• 5. clinicallab.com [clinicallab.com]
• 6. dcmsys.com [dcmsys.com]
• 7. novopath.com [novopath.com]
• 8. The New Era of Pathology: Introduction to Digital Pathology Workflow with Slide-free Technology [digitalpathologyplace.com]
• 9. The Step-by-Step Guide to Implementing Digital Pathology in Your Lab [grundium.com]
• 10. sdmedwaste.com [sdmedwaste.com]
• 11. What Is Pathological Waste and How Should It Be Disposed Of? - INGENIUM [pureingenium.com]
• 12. ehs.uci.edu [ehs.uci.edu]
• 13. ehs.weill.cornell.edu [ehs.weill.cornell.edu]