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

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 routine. Two questions then matter beyond the size of the accuracy loss: which modality was responsible, and whether the model fails loudly or silently once that modality is dropped. The distinction is per-example and modality-level, and is separate from post-hoc feature attribution (e.g. SHAP). Models are replaced often; the evaluation that answers these questions is reused. We present a model-agnostic modality-failure framework: given N modality embeddings, any mask-aware probe, and labels, it returns a per-example failure taxonomy, a per-modality complementarity matrix that attributes error to modalities, and a loud-vs-silent dropout profile separating monitorable failures from those that pass unflagged far from the decision boundary, using only deployment-observable signals. We release it as a small, unit-tested harness and validate it against planted ground truth. Across seeds it recovers that planted modality dominance and complementary subset, reports per-modality loud-vs-silent rates, and scales to a three-modality complementarity matrix; because the planted structure is known by construction, this validates recovery of per-example attribution rather than clinical performance. We then instantiate the framework on frozen EchoJEPA and HuBERT-ECG embeddings for LVEF and the EF <= 40% HFrEF gate over a paired MIMIC-IV cohort, where on the held-out test split (n = 245) dropping echo nearly doubles error. The narrow echo-to-ECG overlap that bounds cohort size is itself a deployment finding for cardiac foundation models. All of our work can be found at https://github.com/criticaldata/PRIMED-AI.

Uncovering Overconfident Failures in CXR Models via Augmentation-Sensitivity Risk Scoring

Deep learning models achieve strong performance in chest radiograph (CXR) interpretation, yet fairness and reliability concerns persist. Models often show uneven accuracy across patient subgroups, leading to hidden failures not reflected in aggregate metrics. Existing error detection approaches -- based on confidence calibration or out-of-distribution (OOD) detection -- struggle with subtle within-distribution errors, while image- and representation-level consistency-based methods remain underexplored in medical imaging. We propose an augmentation-sensitivity risk scoring (ASRS) framework to identify error-prone CXR cases. ASRS applies clinically plausible rotations ($\pm 15^\circ$/$\pm 30^\circ$) and measures embedding shifts with the RAD-DINO encoder. Sensitivity scores stratify samples into stability quartiles, where highly sensitive cases show substantially lower recall ($-0.2$ to $-0.3$) despite high AUROC and confidence. ASRS provides a label-free means for selective prediction and clinician review, improving fairness and safety in medical AI.