Deep Learning-based 12-Lead Electrocardiogram for Low Left Ventricular Ejection Fraction Detection in Patients

Background

Reduced left ventricular ejection fraction (LVEF) initiates heart failure, and promptly identifying low ejection fraction is crucial for managing progression and averting mortality. In this study we developed an artificial intelligence-enabled electrocardiogram (AI-ECG) algorithm to identify patients with low ejection fraction and predict LVEF values.

Methods

The electrocardiogram data were used as input, and the algorithm generated the probability of the patient suffering a low ejection fraction and estimated the LVEF value. A 5-year follow-up study on a group of individuals who initially had normal LVEF values was also performed. Furthermore, external validation of the algorithm performance was conducted using the Medical Information Mart for Intensive Care-IV database.

Results

The algorithm’s performance on the test set yielded an area under the curve value of 0.965 for detecting LVEF ≤ 50%. The algorithm had an accuracy of 92.8%, sensitivity of 88.8%, and specificity of 92.9%. For LVEF regression, the method showed a mean absolute error of 5.28 (95% confidence interval, 5.23-5.33) for the testing set. Additionally, the algorithm obtained an area under the curve value of 0.848 and a mean absolute error value of 9.56 during external validation. Patients with false positive results had a significantly greater likelihood of developing a low ejection fraction compared with patients who received true negative results (26.2% vs 2.0%; P < 0.0001).

Conclusions

The AI-ECG algorithm is capable of identifying low ejection fraction in patients with high accuracy. The AI-ECG algorithm is an efficient, prompt, and cost-effective screening tool for early heart failure.