Study evaluates foundation models for

Short summary

Study evaluates foundation models for multimodal cancer analysis using whole-slide images and transcriptomic data. Found image and omics representations provide complementary signals, with multimodal fusion improving performance. Conformal prediction shows the correct diagnosis remains recoverable within uncertainty sets when point predictions fail.

• •Foundation models show complementary signals between imaging and transcriptomics in oncology
• •Multimodal fusion improves performance when neither modality dominates diagnosis
• •Uncertainty quantification via conformal prediction recovers correct diagnosis when point predictions fail

Generated with AI, which can make mistakes.