PULSE–HF is a deep learning framework that forecasts whether a patient’s left ventricular ejection fraction (LVEF) will decline below 40% within one year based on a standard 12-lead ECG and prior LVEF measurements. It is designed specifically for patients with a history of heart failure.
📘 Preprint Available:
"Forecasting left ventricular systolic dysfunction in heart failure with artificial intelligence"
by Payal Chandak et al., 2025
Heart failure is a major public health burden, with five-year mortality rates exceeding 50%. In heart failure patients with preserved ejection fraction, the ability to anticipate worsening systolic function—before symptoms emerge—opens new doors for:
- 🕒 Early intervention
- 📉 Improved prognostication
- 💊 Timely therapy initiation
- 🏥 Optimized echocardiogram scheduling
⚠️ Existing EHR-based models achieve only 54–68% AUROC for this task. PULSE-HF hits ~92% AUROC across multiple institutions.
PULSE–HF forecasts whether a patient's LVEF will fall below 40% within 1 year after an ECG is taken. It does this by combining:
- 🖥️ Raw 12-lead ECG waveform data
- 📊 History of past LVEF values
It also includes a Lead I version that performs comparably—ideal for wearables or home-based monitoring.
If you find this work helpful, please reference:
@article{chandak2025pulsehf,
title={Forecasting left ventricular systolic dysfunction in heart failure with artificial intelligence},
author={Chandak, Payal and Kyereme-Tuah, Abena and Hung, Judy and Gaggin, Hanna and Kohane, Isaac S. and Stultz, Collin M.},
journal={medRxiv},
year={2025},
doi={10.1101/2025.04.13.25325744},
}