Benchmarking Pathology Foundation Models for Breast Cancer Survival Prediction
A new study benchmarks pathology foundation models (PFMs) for predicting breast cancer survival using histopathology images. The research evaluated several PFMs across three independent patient cohorts, finding that H-optimus-1 performed best. Second-generation PFMs generally outperformed earlier ones, though performance gains are becoming modest. Notably, a smaller distilled model, H0-mini, achieved better results than its larger counterpart, H-optimus-0, while being significantly more efficient. AI
IMPACT Provides guidance on selecting efficient pathology foundation models for clinical deployment in cancer survival prediction.