A screening method for left ventricular systolic dysfunction by single-channel electrocardiogram using machine learning algorithms

Background. Analysis of a single-channel electrocardiogram can potentially be used as a screening method to detect systolic dysfunction of the left ventricle. The purpose of our study was to develop a new screening method for detecting a decrease in systolic function of the left ventricle based on single-channel ECG and pulse wave recording using machine learning methods. Materials and methods. The study prospectively included 1039 patients aged 18 years and above. A transthoracic echocardiographic study and 1-minute single channel electrocardiogram were performed for each patient. Spectral analysis of the electrocardiogram based on the Fourier transform. More than 200 parameters were included in machine learning algorithms. Results. For ejection fraction decrease: Lasso regression showed a sensitivity of 92,2%, specificity of 90,1% ( AUC=0.920); Random Forest Classifier sensitivity - 88.2%, specificity - 83,3% (AUC=0.834). Algorithm approbation has shown diagnostic accuracy of 90,1% in left ventricular systolic dysfunction. Conclusions. Machine learning models, based on the single lead ECG parameters, as well as age and gender may simplify screening diagnostics of ejection fraction decrease prior to echocardiographic study for in time heart failure diagnostics with high accuracy.