Automated and Interpretable Patient ECG Profiles for Disease Detection, Tracking, and Discovery

The electrocardiogram or ECG has been in use for over 100 years and remains\nthe most widely performed diagnostic test to characterize cardiac structure and\nelectrical activity. We hypothesized that parallel advances in computing power,\ninnovations in machine learning algorithms, and availability of large-scale\ndigitized ECG data would enable extending the utility of the ECG beyond its\ncurrent limitations, while at the same time preserving interpretability, which\nis fundamental to medical decision-making. We identified 36,186 ECGs from the\nUCSF database that were 1) in normal sinus rhythm and 2) would enable training\nof specific models for estimation of cardiac structure or function or detection\nof disease. We derived a novel model for ECG segmentation using convolutional\nneural networks (CNN) and Hidden Markov Models (HMM) and evaluated its output\nby comparing electrical interval estimates to 141,864 measurements from the\nclinical workflow. We built a 725-element patient-level ECG profile using\ndownsampled segmentation data and trained machine learning models to estimate\nleft ventricular mass, left atrial volume, mitral annulus e' and to detect and\ntrack four diseases: pulmonary arterial hypertension (PAH), hypertrophic\ncardiomyopathy (HCM), cardiac amyloid (CA), and mitral valve prolapse (MVP).\nCNN-HMM derived ECG segmentation agreed with clinical estimates, with median\nabsolute deviations (MAD) as a fraction of observed value of 0.6% for heart\nrate and 4% for QT interval. Patient-level ECG profiles enabled quantitative\nestimates of left ventricular and mitral annulus e' velocity with good\ndiscrimination in binary classification models of left ventricular hypertrophy\nand diastolic function. Models for disease detection ranged from AUROC of 0.94\nto 0.77 for MVP. Top-ranked variables for all models included known ECG\ncharacteristics along with novel predictors of these traits/diseases.\n