Leveraging Foundation Models for Histological Grading in Cutaneous Squamous Cell Carcinoma using PathFMTools
Abdul Rahman Diab1, Emily E Karn2, Renchin Wu1
1Dana-Farber Cancer Institute.
Proceedings of Machine Learning Research
|May 25, 2026
Summary
PathFMTools simplifies adapting computational pathology foundation models for clinical use. This Python package enables efficient analysis and validation of models for tasks like cutaneous squamous cell carcinoma grading.
Area of Science:
- Pathology
- Computer Science
- Artificial Intelligence
Background:
- Adapting computational pathology foundation models to clinical tasks is complex.
- Challenges include whole-slide image (WSI) processing, feature opacity, and adaptation strategies.
Purpose of the Study:
- Introduce PathFMTools, a Python package for efficient pathology foundation model analysis.
- Evaluate foundation models (CONCH, MUSK) for cutaneous squamous cell carcinoma (cSCC) histological grading.
Main Methods:
- Developed and utilized PathFMTools for model execution, analysis, and visualization.
- Benchmarked adaptation strategies using 440 cSCC H&E WSIs.
- Interfaced with CONCH and MUSK vision-language foundation models.
Main Results:
- Demonstrated tradeoffs across different prediction approaches for cSCC grading.
- Validated the potential of using foundation model embeddings to train smaller specialist models.
- PathFMTools facilitated efficient analysis and validation.
Conclusions:
- Pathology foundation models show promise for real-world clinical applications.
- PathFMTools enables efficient adaptation and validation of these models.
- The study highlights the utility of foundation models in histological grading.
