Related Experiment Video
Updated: Aug 4, 2026
08:05
Detection and Isolation of Cancer in Prostate Biopsies Using Stimulated Raman Histology and Artificial Intelligence
Published on: June 10, 2025
1.1K
Pathologist-interpretable breast cancer subtyping and stratification from AI-inferred nuclear features
Ranjan Kumar Barman1, Saugato Rahman Dhruba1, Danh-Tai Hoang1
1Cancer Data Science Laboratory, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA.
Biorxiv : the Preprint Server for Biology
|September 29, 2025
Summary
EXPAND is a new AI tool for digital pathology that predicts breast cancer subtypes and survival using pathologist-interpretable features. This explainable AI model achieves performance comparable to black box models, enhancing diagnostic transparency.
Area of Science:
- Digital Pathology
- Artificial Intelligence
- Computational Biology
Background:
- Artificial intelligence (AI) offers advancements in digital pathology but often lacks human interpretability.
- Existing AI models in pathology can be 'black boxes', hindering clinical trust and adoption.
- Pathologist interpretability is crucial for integrating AI into routine cancer diagnostics.
Purpose of the Study:
- To introduce EXPAND (EXplainable Pathologist Aligned Nuclear Discriminator), the first pathologist-interpretable AI model for breast cancer diagnostics.
- To develop an automated workflow for predicting tumor subtypes and patient survival using interpretable features.
- To validate the performance and prognostic value of interpretable AI in breast cancer.
Main Methods:
- Development of EXPAND, an AI model focusing on 12 nuclear pathologist-interpretable features (NPIFs) from the Nottingham grading criteria.
- Implementation of a fully automated, end-to-end diagnostic workflow for NPIF extraction and prediction.
- Evaluation of EXPAND's performance in predicting HER2+, HR+, and TNBC tumor subtypes and patient survival.
Main Results:
- EXPAND achieved AUC values of 0.73 (HER2+), 0.79 (HR+), and 0.75 (TNBC), comparable to non-interpretable AI models.
- The 12 NPIFs demonstrated significant and independent prognostic value for patient survival.
- EXPAND's performance matched proprietary models using more complex feature sets.
Conclusions:
- EXPAND provides a pathologist-interpretable AI solution for breast cancer diagnostics, enhancing transparency.
- The 12 NPIFs serve as biologically grounded, interpretable biomarkers for survival stratification.
- This work paves the way for developing interpretable AI diagnostic models in other cancer types.
