Digital PATH Project
Digital and Computational Pathology Tool Harmonization Project
Quick Links
Share on Social Media
How do biomarker results differ across AI pathology platforms and how can their performance be effectively compared?
What is the unmet need and why does it matter?
Traditionally, pathologists examine tissue obtained through a biopsy to diagnose cancer, determine the type and stage, and identify biomarkers that not only indicate a patient’s likely response to certain treatments but also provide other clinical insights. Digital pathology enables innovative approaches to these assessments through the scanning and digitization of slides for storage, viewing, and analysis. Analyses can include the use of computational pathology platforms with artificial intelligence (AI)/machine learning (ML) algorithms to support the pathologist in tissue analysis.
Similar to other projects in Friends Diagnostics Harmonization Portfolio, we assessed sources of variability in biomarker outputs using HER2 as a use case to identify opportunities to harmonize methodology and support more consistent measurement and use (i.e., the TMB Harmonization Project and the HRD Harmonization Project).To start the project, Friends convened a group of algorithm developers, patient advocates, government officials, pathologists, and drug developers to assess the comparability of biomarker measurements across computational pathology platforms, identify factors that may contribute to any observed variability, and propose areas where greater alignment may be needed.
How are we helping to find solutions?
To initiate the project, a multi-stakeholder working group evaluated existing regulatory frameworks and developed proposals to support robust development of these emerging technologies including a risk-based approach to assessing evidentiary needs for validation. We published a white paper of findings and presented results during our public meeting in September 2023.
We used findings from the white paper to inform an approach to compare outputs from digital pathology platforms. Our working group aligned on an analysis plan and shared a common dataset of digital images of breast cancer from over 1000 patients. Platform developers evaluated HER2 (a growth-promoting protein and established biomarker) in breast cancer samples using their independently developed approach, then we compared outputs across multiple computational pathology platforms. Our objective was to characterize the level of variability of assessments across digital and computational pathology platforms, as this had only been assessed across human readers, not AI tools. The results from this pilot may help inform future regulatory thinking around efficient approaches for performance demonstration that could be applied for validation of future AI-enabled test platforms.
How does this impact patients?
Digital and computational pathology platforms have the potential to support greater accuracy, reproducibility, and standardization of pathology features, expedite diagnosis or pathological scoring, establish new biomarkers, and identify and select the appropriate patients for treatments – all of which can contribute to improving patient outcomes. Supporting the robust development of these platforms and identifying potential sources of variability will help to inform future use and advancements in technology that deliver more precise patient care. Without Friends coordination and support from collaborative partners, groups may never align on a solution to improve consistency across assays and interpreting results would be more challenging for patients and providers.
Project Outcomes
Friends prioritizes sharing findings from our projects with the public to inform policy:
Friends Led Peer Reviewed Literature
McKelvey BA, Torres-Saavedrab PA, Li J et al. Agreement Across 10 Artificial Intelligence Models in Assessing Human Epidermal Growth Factor Receptor 2 (HER2) Expression in Breast Cancer Whole-Slide Images. Modern Pathology 2026;
https://doi.org/10.1016/j.modpat.2025.100944.
Friends Public Meetings
In February of 2025, Friends held a public meeting to discuss key considerations for establishing a reference dataset to validate digital pathology tools.
In February of 2024, a public meeting was hosted by Friends to convene experts to discuss the objectives of the Digital PATH demonstration project to support the identify factors that may contribute to variability and propose strategies for alignment in the field.
In 2023, Friends hosted a virtual public meeting to explore future opportunities to leverage existing data to support the proposals identified in our initial research.
Project Partners
Digital PATH Project: 4D Path, Amgen Inc., AstraZeneca, BostonGene Corporation, Bristol Myers Squibb, Caris Life Sciences, Inc., Daiichi Sankyo, Inc., EMD Serono, Inc., Emory University, Friends Advisory Advocates, GA Green Consulting LLC, GSK, Indica Labs, Johnson & Johnson Innovative Medicine, Karolinska Institutet, Kulig Consulting, LLC, Loxo@Lilly, Lunit, Molecular Characterization Laboratory (MoCha) at Frederick National Laboratory, MD Anderson Cancer Center, Merck & Co., Inc, National Cancer Institute (NCI), Nucleai, Panakeia Technologies Ltd, PathAI, Roche Diagnostics, Sanofi, Tempus AI, Inc., the U.S. Food and Drug Administration (FDA), Ziekenhuis Aan de Stroom (ZAS) Hospital, University of North Carolina, and Verily Life Sciences LLC.
Landscape Assessment:4D Path, Amgen Inc., AstraZeneca, Bristol Myers Squibb, EMD Serono, Inc., GSK, Kulig Consulting, LLC, Loxo@Lilly, Massachusetts General Hospital, MD Anderson Cancer Center, Merck & Co., Inc., Neomorph, Inc., Paige.AI, Inc., PathAI, Sanofi, Tempus AI, Inc., the U.S. Food and Drug Administration (FDA), University Hospital of Antwerp, and University of North Carolina at Chapel Hill.