Multimodal AI with Foundation Models Predicts Platinum Resistance in Ovarian Cancer

A foundation model-powered trimodal AI combining MRI, pathology, and clinical data predicted platinum resistance in high-grade serous ovarian cancer, achieving AUCs up to 0.790 in external validation—outperforming late fusion.

• The trimodal approach used frozen foundation models for MRI and pathology whole-slide images, with attention-based aggregation and cross-modal fusion that handles missing modalities.
• External validation on three independent cohorts showed consistent superiority over a late-fusion model (e.g., AUC 0.790 vs 0.730 in cohort C).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.