Deep learning of ECG waveforms for diagnosis of heart failure with a reduced left ventricular ejection fraction

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Heart failure with reduced ejection fraction (HFrEF) significantly impacts patient survival.
• Accurate identification of HFrEF is crucial for timely intervention and management.
• Current diagnostic methods may have limitations, especially in resource-limited settings.

Purpose of the Study:

• To evaluate the performance of a deep learning algorithm, DeepECG-HFrEF, in identifying HFrEF from electrocardiograms (ECGs).
• To assess the clinical implications of DeepECG-HFrEF, including its association with patient survival.
• To determine the utility of DeepECG-HFrEF in diagnosing left ventricular systolic dysfunction (LVSD).

Main Methods:

• The DeepECG-HFrEF algorithm was trained to detect LVSD (ejection fraction < 40%).
• Performance was assessed using the area under the receiver operating characteristic curve (AUC) in a cohort of acute heart failure patients.
• Five-year mortality was analyzed using the Kaplan-Meier method based on algorithm classification.

Main Results:

• The study included 1291 ECGs from 690 patients (mean age 67.8 years, 56% men).
• DeepECG-HFrEF achieved an AUC of 0.844 for identifying HFrEF in acute symptomatic heart failure patients.
• Patients classified as HFrEF (+) by the algorithm demonstrated significantly lower 5-year survival rates (log-rank p < 0.001).

Conclusions:

• The DeepECG-HFrEF algorithm demonstrates acceptable performance in discriminating HFrEF within a real-world clinical setting.
• The algorithm's classification of HFrEF is associated with increased mortality, highlighting its prognostic value.
• DeepECG-HFrEF shows potential for aiding in LVSD identification and risk stratification, particularly in resource-limited environments.