Loud or Silent? A Reusable Framework for Per-Modality Failure Analysis in Multimodal Clinical AI

Quang Bui, Shlok Jaiswal, Samuel Paik-Heintz +14

Multimodal clinical models are usually judged on accuracy with every modality present, but deployment removes modalities; an echocardiogram is often unavailable where an ECG is rou…