How Well Do Multimodal Models Reason on ECG Signals?

Researchers introduce a new framework for evaluating how well multimodal AI models reason about ECG signals by breaking down reasoning into perception (pattern identification) and deduction (logical application of medical knowledge). The framework uses automated code generation to verify temporal patterns and compares model logic against established clinical criteria databases.

• →Current evaluation methods for health AI reasoning are either unscalable or fail to capture semantic correctness of clinical logic.
• →The new framework decomposes AI reasoning into two components: perception of signal patterns and deduction using domain knowledge.
• →An agentic framework generates code to empirically verify temporal structures in ECG reasoning traces.
• →Deduction evaluation measures alignment against structured databases of established clinical criteria.
• →This approach enables scalable assessment of true reasoning capabilities in medical AI applications.