Automatic Right Ventricular Hypertrophic Detection Integrating Electrocardiography-based QRS Biomarkers with Machine Learning

Despite its widespread use, the electrocardiogram (ECG) exhibits limited sensitivity in detecting right ventricular hypertrophy (RVH), primarily due to the overshadowing effects of the left ventricular activation. This study addresses this diagnostic challenge by extracting morphological QRS biomarkers from 12-lead recordings to classify RVH patients versus healthy controls. Leveraging a publicly available database comprising 9,001 patients (101 RVH, 8,900 control), we extracted features including width, amplitudes, slopes between fiducial points, and Hermite transform coefficients. Utilizing logistic regression, random forest, and support vector machine algorithms following sequential feature selection, our classifiers achieved a minimum accuracy of 88% on an independent validation dataset of 1,456 individuals (69 RVH, 1,387 control). Notably, logistic regression and random forest demonstrated valuable sensitivity, reaching 85% and 87%, respectively. The three or four selected features align with clinical recommendations, underscoring their potential utility in enhancing RVH detection via ECG biomarkers, driven by machine learning algorithms.