Researchers at MIT, Mass General Brigham, and Harvard Medical School have developed PULSE-HF, a deep learning model that predicts heart failure progression up to a year in advance using ECG data. The model forecasts whether a patient's left ventricular ejection fraction will drop below 40% — the most severe heart failure threshold — achieving AUROCs of 0.87–0.91 across three patient cohorts. Unlike existing methods that detect current conditions, PULSE-HF performs forecasting, enabling clinicians to prioritize high-risk patients and reduce unnecessary hospital visits for low-risk ones. A single-lead ECG version performs comparably to the 12-lead version, making it viable for low-resource settings. Key challenges included cleaning messy ECG and echocardiogram datasets. The next step is prospective testing on real patients.
Sort: