ECG Deep Learning Dissects Physiology of Hypertrophic Cardiomyopathy

Hypertrophic cardiomyopathy (HCM) is mechanistically complex, and unravelling this would guide treatment and prognosis. One serious complication is left ventricular outflow tract (LVOT) obstruction, which increases afterload. Since afterload also increases with high blood pressure (BP), we hypothesised that obstructive HCM may share ECG features with hypertension. This study aimed to identify hypertensive-like ECG changes to discriminate physiological factors within HCM. A deep convolutional neural network was trained to quantify ECG changes in hypertension in n=37,378 UK Biobank participants and n=2,258 HCM Registry patients, generating a novel ECG strain score (ECG-SS). ECG-SS was raised in HCM independently of BP, indicating they have an alternative reason for strain. Sarcomere-negative HCM patients (no disease-causing gene) had bimodally distributed scores, suggesting two phenotypes distinguished by afterload. In this group, high ECG-SS was associated with obstruction and wall stress, but not with hypertrophy or fibrosis. To give clinical interpretability, a variational autoencoder allowed visualisation of ECG features, and demonstrated marked abnormalities in the high ECG-SS subgroup. In conclusion, ECG deep learning demonstrated, for the first time, LVOT obstruction to be the major cause of ECG abnormalities in HCM.