A screening method for aortic valve disease based on spectral analysis of a single-channel electrocardiogram using machine learning algorithms

Background Given the high prevalence of aortic heart disease and the importance of early diagnosis, there is a need to develop a simple screening method to identify significant aortic heart disease based on available biosignals, one of which is a single-channel electrocardiogram (ECG). Aim We aimed to develop a machine learning algorithm for analyzing single-channel electrocardiograms to identify patients with moderate to severe aortic stenosis and regurgitation, as well as significant aortic valve disease. Methods Patients ( n = 2,503) underwent echocardiographic examination and single-channel I-lead ECG recording using a portable electrocardiograph. Moderate to severe aortic stenosis on echocardiography was present in 62 (2.4%) patients, and moderate to severe aortic regurgitation was present in 157 (6.3%). Single-channel ECG analysis was performed using advanced signal processing and machine learning techniques. Results Both models were evaluated under an identical 10-fold group cross-validation framework. For detecting moderate or severe aortic stenosis, diagnostic accuracy was 0.650 (Se 57.9%, Sp 65.0%) for Balanced Random Forest and 0.662 (Se 46.4%, Sp 66.5%) for XGBoost. For moderate or severe aortic regurgitation, accuracy was 0.726 (Se 60.5%, Sp 73.3%) for Balanced Random Forest and 0.712 (Se 61.7%, Sp 71.9%) for XGBoost. For significant aortic valve disease, accuracy was 0.724 (Se 61.1%, Sp 73.4%) for Balanced Random Forest and 0.691 (Se 67.8%, Sp 69.2%) for XGBoost. Conclusions The results indicate good diagnostic accuracy and potential applicability for screening for significant aortic heart disease using single-channel ECG parameters analyzed with modern signal processing and machine learning techniques. Clinical Trial Registration : https://clinicaltrial.gov/ , identifier NCT07099417.