ECG-LLM: Foundation Model for ECG-Based Cardiac Reasoning

2026-07-15

AI Summary

This work addresses the limitations of existing ECG-based AI systems, which rely on fixed labels or auto-generated reports and struggle to support individualized clinical reasoning in primary care settings where immediate cardiac imaging is often unavailable. To overcome this, we propose the first multimodal large language model grounded in 12-lead electrocardiography that integrates ECG signals, clinical data, and cardiac imaging modalities—such as echocardiography and cardiac MRI—by constructing structured question-answer pairs to establish a question-driven, end-to-end training paradigm. By transforming multimodal clinical data into language-based supervision signals, our model, trained on 679,112 multicenter ECGs, not only accurately reproduces standard ECG metrics but also predicts imaging-derived phenotypes, including ventricular volumes and aortic stenosis severity. It achieves state-of-the-art or comparable performance on ECG-QA and diagnostic report generation tasks, thereby transcending the constraints of conventional fixed-label prediction frameworks.

Abstract

Electrocardiography (ECG) is an inexpensive, standard-of-care test for cardiac symptoms, but front-line triage often lacks immediate access to definitive imaging such as echocardiography (ECHO) or cardiac magnetic resonance (CMR). Furthermore, most existing ECGAI systems are limited to fixed diagnostic labels or automated reports, constraining their use for patient-specific clinical reasoning. To address this gap, we introduce ECG-LLM, an ECG-conditioned large language model trained across four cohorts comprising 679,112 ECG studies from 186,409 patients. Using a novel multimodal-to-language supervision strategy, ECG-LLM is trained on clinically structured question-answer pairs derived from ECG signals, clinical context, CMR, and ECHO. This unified approach enables the model to answer diverse cardiovascular questions from a 12-lead ECG alone, spanning both conventional interpretation and phenotypes not directly visible on standard ECGs. ECG-LLM successfully recovers conventional ECG measurements, such as heart rate, and strongly predicts complex CMR-derived phenotypes, including ventricular and atrial volumes and ventricular function. Crucially, it detects vital echocardiographic phenotypes, including increased LV wall thickness, aortic stenosis, and right-ventricular systolic dysfunction. On standard ECG understanding tasks, ECG-LLM matches or exceeds existing baselines for diagnostic report generation and the ECG-QA benchmark. By moving beyond fixed-label prediction, this multimodal framework provides clinically valuable, question-driven cardiovascular reasoning to support general practitioner and front-line triage decisions when specialist review is delayed.

Problem

• ECG
• clinical reasoning
• cardiac diagnosis
• multimodal learning
• front-line triage

Innovation

• ECG-LLM
• multimodal-to-language supervision
• cardiac reasoning
• question-driven diagnosis
• foundation model