Extraction of morphological QRS-based biomarkers in hypertrophic cardiomyopathy for risk stratification using L1 regularized logistic regression

Hypertrophic cardiomyopathy (HCM) is an inherited cardiac disease characterized by an unexplained thickening of the heart ventricles. It is the first cause of sudden cardiac death in young adults. No reliable biomarkers for risk assessment have been presented so far, but the electrocardiograms of HCM patients are often abnormal due to structural and electrical abnormalities. The goal of our study was to extract morphological QRS biomarkers in order to discriminate between HCM patients and control patients by analyzing fifty 12-lead Holter recordings (29 HCM – 21 control). Morphological features such as QRS width or slopes from the QRS complex directly and the coefficients of the first four Hermite transform basis were extracted. Classification was then performed using those features in an L1 regularized logistic regression algorithm. Classification between control and HCM patients reached 95.7% of accuracy (sensitivity of 94.96% for HCM and specificity of 96.90%) using only two main features: the percentage of negative regions of the QRS complex with respect to the isoelectric level and the 3rd coefficient of its Hermite fitting showing interesting connections to cardiac electrophysiology.