A screening method for predicting left ventricular dysfunction based on spectral analysis of a single-channel electrocardiogram using machine learning algorithms

Abstract

BackgroundAnalysis of a single-channel electrocardiogram can potentially be used as a screening method to detect systolic and diastolic dysfunction of the left ventricle. The purpose of our study was to develop a new screening method for detecting a decrease in systolic and/or diastolic function of the left ventricle based on single-channel ECG and pulse wave recording using machine learning methods. Materials and methodsThe study prospectively included 375 patients aged 18 years and above. Every participant underwent a transthoracic echocardiographic study with the aim of determining the systolic (EF below 52% for male and 54% for female) and diastolic dysfunction of the left ventricle, as well as, registering the 3 min single channel electrocardiogram and pulse wave using a portative electrocardiogram device. Spectral analysis of the electrocardiogram was performed which was based on the Fourier transform. ResultsFor ejection fraction decrease: Lasso regression showed a sensitivity of 92,3%, specificity of 90,2% (AUC = 0.922); Random Forest Classifier sensitivity − 100%, specificity − 84,3% (AUC = 0.931). For diastolic dysfunction: Lasso regression sensitivity − 86,2%, specificity − 84,5% (AUC 0.878); random forest classifier sensitivity − 82,8%, specificity 89,3% (AUC 0.854). Algorithm approbation has shown diagnostic accuracy of 95,6% left ventricular diastolic dysfunction of 2–3 grade. ConclusionsMachine learning models, based on the single lead ECG and pulse wave parameters, as well as age and gender may simplify screening diagnostics of ejection fraction decrease and diastolic dysfunction prior to echocardiographic study for in time heart failure diagnostics with high accuracy.