CohortFinder - An Open-Source Tool for Data-Driven Partitioning of Digital Pathology and Imaging Cohorts - To Yield Robust Machine-Learning Models

Aug 5, 2026

Upload by Ezra

CohortFinder is an open-source tool designed to mitigate batch effects in digital pathology and medical imaging datasets through data-driven cohort partitioning. Systematic technical differences introduced by physical, temporal, and digitization processes can create visual or acquisition-dependent variability that degrades machine-learning generalizability. By partitioning cohorts to reduce these non-biological artifacts, CohortFinder improves downstream model performance across digital pathology and medical image processing tasks, enabling more robust learning on unseen data.

LanguageEnglish

CategoryTechnology

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Overview

Motivation: batch effects in digital pathology and imaging

CohortFinder approach and availability

Impact on machine-learning model performance

Zurich Open Repository and Archive University of Zurich University Library Strickhofstrasse 39 CH-8057 Zurich [www.zora.uzh.ch](www.zora.uzh.ch) Year: 2024 CohortFinder : An Open-Source Tool for Data-Driven Partitioning of Digital Pathology and Imaging Cohorts to Yield Robust Machine-Learning Models Fan, Fan ; Martinez, Georgia ; DeSilvio, Thomas ; Shin, John ; Chen, Yijiang ; Jacobs, Jackson ; Wang, Bangchen ; Ozeki, Takaya ; Lafarge, Maxime W ; Koelzer, Viktor H ; Barisoni, Laura ; Madabhushi, Anant ; Viswanath, Satish E ; Janowczyk, Andrew DOI: [https://doi.org/10.1038/s44303-024-00018-2](https://doi.org/10.1038/s44303-024-00018-2) Posted at the Zurich Open Repository and Archive, University of Zurich ZORA URL: [https://doi.org/10.5167/uzh-267293](https://doi.org/10.5167/uzh-267293) Journal Article Published Version The following work is licensed under a Creative Commons: Attribution 4.0 International (CC BY 4.0) License. Originally published at: Fan, Fan; Martinez, Georgia; DeSilvio, Thomas; Shin, John; Chen, Yijiang; Jacobs, Jackson; Wang, Bangchen; Ozeki, Takaya; Lafarge, Maxime W; Koelzer, Viktor H; Barisoni, Laura; Madabhushi, Anant; Viswanath, Satish E; Janowczyk, Andrew (2024) . CohortFinder : An Open-Source Tool for Data-Driven Partitioning of Digital Pathology and Imaging Cohorts to Yield Robust Machine-Learning Models. npj Imaging, 2(1):15 . DOI: [https://doi.org/10.1038/s44303-024-00018-2](https://doi.org/10.1038/s44303-024-00018-2) [https://doi.org/10.1038/s44303-024-00018-2](https://doi.org/10.1038/s44303-024-00018-2) | CohortFinder: an open-source tool for data-driven partitioning of digital pathology and imaging cohorts to yield robust machine-learning models
Check for updates | | | --- | --- | | Fan Fan1, Georgia Martinez2, Thomas DeSilvio2, John Shin2, Yijiang Chen2, Jackson Jacobs1, Bangchen Wang3, Takaya Ozeki4, Maxime W. Lafarge5, Viktor H. Koelzer5, Laura Barisoni3,6, Anant Madabhushi1,7, Satish E. Viswanath2 & Andrew Janowczyk1,8,9 | | | Batch effects (BEs) refer to systematic technical differences in data collection unrelated to biological variations whose noise is shown to negatively impact machine learning (ML) model generalizability. Here we release CohortFinder (http://cohort􀀁nder.com), an open-source tool aimed at mitigating BEs via data-driven cohort partitioning. We demonstrate CohortFinder improves ML model performance in downstream digital pathology and medical image processing tasks. CohortFinderis freely available for download at cohort􀀁nder.com. | | | The increased availability of digital pathology (DP) whole slide images (WSI) and radiographic imaging datasets has propelled the development of both machine and deep learning algorithms to aid in disease diagnosis, patient prognosis, and predicting therapy response1. These algorithms work by identifying patterns in digital data that are associated with clinical outcomes of interest. While large-scale data analysis was previously limited by storage, processing, and computational constraints, modern-day development and testing ofthese models increasingly involves the collection oflarge cohorts over both physical (e.g., institutions) and temporal (e.g., time points) spaces1. However, differences in non-biological preanalytical processesat these various spatiotemporal points likely impart undesirable batch effects (BE) in the ﬁnal digital data. For example, BEs in DP images generated in the same manner from the same tissue type yield signiﬁcant visual differences which may impact data interpretation (see Fig. 1A).
In DP, these BEs tend to originate from, but are not limited to, differencesin physical processes for data generation(tissue processing, storage, glass slide preparation) as well as digitization processes (scanners, color proﬁle management, compression approaches)1–6. In MR imaging cohorts, | these BEs may result from MRI acquisition protocols, patient preparation differences, or imaging arti

FAQs

What problem does CohortFinder address?

CohortFinder addresses batch effects—systematic technical differences unrelated to biology that negatively impact machine-learning model generalizability.

How does CohortFinder mitigate batch effects?

It uses data-driven cohort partitioning to reduce the influence of non-biological variability in digital pathology and medical imaging data.

What benefits does CohortFinder provide for machine learning tasks?

CohortFinder improves machine-learning model performance in downstream digital pathology and medical image processing tasks by increasing robustness to batch-effect-induced differences.

Aug 5, 2026

Upload by Ezra