Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Survey on Computational Pathology Foundation Models: Datasets, Adaptation Strategies, and Evaluation Tasks

Dong Li, Guihong Wan, Xintao Wu, Xinyu Wu, Ajit J. Nirmal, Christine G. Lian, Peter K. Sorger, Yevgeniy R. Semenov, Chen Zhao

TL;DR

The paper addresses the problem of building robust computational pathology foundation models (CPathFMs) under limited labeled data and heterogeneous real-world datasets. It surveys pre-training datasets, adaptation strategies, and downstream evaluation tasks, distinguishing uni-modal image-based approaches from multi-modal image-text methods and detailing representative models such as DINO, DINOv2, BEiT, CLIP, and CoCa in pathology. The authors provide a taxonomy of datasets (public and private), adaptation techniques, and a six-category evaluation framework, then discuss gaps and practical directions for achieving clinically relevant performance. This synthesis clarifies methodological gaps, highlights the importance of standardized benchmarks, and outlines actionable pathways to improve generalization, reliability, and regulatory readiness of CPathFMs in clinical workflows.

Abstract

Computational pathology foundation models (CPathFMs) have emerged as a powerful approach for analyzing histopathological data, leveraging self-supervised learning to extract robust feature representations from unlabeled whole-slide images. These models, categorized into uni-modal and multi-modal frameworks, have demonstrated promise in automating complex pathology tasks such as segmentation, classification, and biomarker discovery. However, the development of CPathFMs presents significant challenges, such as limited data accessibility, high variability across datasets, the necessity for domain-specific adaptation, and the lack of standardized evaluation benchmarks. This survey provides a comprehensive review of CPathFMs in computational pathology, focusing on datasets, adaptation strategies, and evaluation tasks. We analyze key techniques, such as contrastive learning and multi-modal integration, and highlight existing gaps in current research. Finally, we explore future directions from four perspectives for advancing CPathFMs. This survey serves as a valuable resource for researchers, clinicians, and AI practitioners, guiding the advancement of CPathFMs toward robust and clinically applicable AI-driven pathology solutions.

A Survey on Computational Pathology Foundation Models: Datasets, Adaptation Strategies, and Evaluation Tasks

TL;DR

The paper addresses the problem of building robust computational pathology foundation models (CPathFMs) under limited labeled data and heterogeneous real-world datasets. It surveys pre-training datasets, adaptation strategies, and downstream evaluation tasks, distinguishing uni-modal image-based approaches from multi-modal image-text methods and detailing representative models such as DINO, DINOv2, BEiT, CLIP, and CoCa in pathology. The authors provide a taxonomy of datasets (public and private), adaptation techniques, and a six-category evaluation framework, then discuss gaps and practical directions for achieving clinically relevant performance. This synthesis clarifies methodological gaps, highlights the importance of standardized benchmarks, and outlines actionable pathways to improve generalization, reliability, and regulatory readiness of CPathFMs in clinical workflows.

Abstract

Computational pathology foundation models (CPathFMs) have emerged as a powerful approach for analyzing histopathological data, leveraging self-supervised learning to extract robust feature representations from unlabeled whole-slide images. These models, categorized into uni-modal and multi-modal frameworks, have demonstrated promise in automating complex pathology tasks such as segmentation, classification, and biomarker discovery. However, the development of CPathFMs presents significant challenges, such as limited data accessibility, high variability across datasets, the necessity for domain-specific adaptation, and the lack of standardized evaluation benchmarks. This survey provides a comprehensive review of CPathFMs in computational pathology, focusing on datasets, adaptation strategies, and evaluation tasks. We analyze key techniques, such as contrastive learning and multi-modal integration, and highlight existing gaps in current research. Finally, we explore future directions from four perspectives for advancing CPathFMs. This survey serves as a valuable resource for researchers, clinicians, and AI practitioners, guiding the advancement of CPathFMs toward robust and clinically applicable AI-driven pathology solutions.
Paper Structure (12 sections, 3 figures, 2 tables)

This paper contains 12 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Computational Pathology (CPath)
  4. ViT-Based SSL Frameworks for FMs
  5. Challenges in Pre-training CPathFMs
  6. Pre-training Datasets in CPathFMs
  7. Adaptation Strategies in CPathFMs
  8. Uni-Modal CPathFMs
  9. Multi-Modal CPathFMs
  10. Evaluation Tasks
  11. Future Directions
  12. Conclusion

Figures (3)

  • Figure 1: An illustrative example of data modalities and challenges in CPath. The figure illustrates different histopathology data types, including WSIs, tile images at multiple magnifications (Field of View, FoV), and imaging types (H&E, IHC, MxIF). These elements are critical for developing CPathFMs, highlighting the complexity of multi-scale image representation and domain-specific challenges.
  • Figure 2: Overview of the pre-training pipeline for CPathFMs. The process involves data curation, including image curation, text curation, and dataset filtering, followed by uni-modal and multi-modal pre-training. The final CPathFMs are evaluated across multiple downstream tasks categorized into six main perspectives.
  • Figure 3: Taxonomy of evaluation tasks for pre-trained CPathFMs. Uni-modal and multi-modal CPathFMs are highlighted in purple and red, respectively.