New Machine Learning Framework Enhances LVEF Diagnosis

A novel multimodal machine-learning framework has been developed to assess left ventricular ejection fraction (LVEF) through 12-lead ECG features combined with electronic health record (EHR) variables. Utilizing data from Hartford HealthCare, the model classifies LVEF into four categories based on a vast dataset comprising 36,784 ECG-echocardiogram pairs. It demonstrates superior performance in accuracy over traditional models, achieving an AUROC of up to 0.95 for severe LVEF reduction and 0.91 for normal readings.

The implications of this development are significant for primary care and settings with limited resources. By enabling ECG-based LVEF stratification, this multimodal approach not only enhances accessibility to cardiovascular diagnostics but also allows for improved patient prioritization for confirmatory imaging. This aligns with broader goals of integrating AI into healthcare systems, potentially revolutionizing patient triage protocols and resource allocation at scale.