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

2026-08-02

AI Summary

Multimodal clinical AI systems often suffer performance degradation in deployment due to missing modalities, yet it remains challenging to identify which modality fails and whether its failure is detectable (“loud”) or undetectable (“silent”). This work proposes a model-agnostic, fine-grained failure analysis framework that relies solely on observable signals during deployment. By leveraging modality embeddings, mask-aware probes, and labels, the method enables per-sample, per-modality failure attribution and outputs both a modality complementarity matrix and loud/silent failure profiles. In simulated data, the framework accurately recovers predefined modality dominance relationships; in the real-world MIMIC-IV cohort, it reveals that missing echocardiography nearly doubles error rates and, for the first time, quantifies the proportions of monitorable versus unmonitorable failures across modalities.

Abstract

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.

Problem

• multimodal clinical AI
• modality failure
• loud vs silent failure
• failure attribution
• deployment robustness

Innovation

• modality failure analysis
• loud vs silent failure
• model-agnostic framework
• complementarity matrix
• multimodal clinical AI