Artificial Intelligence in Electrocardiography: From Automated Arrhythmia Detection to Predicting Hidden...

Cardiovascular diseases are among the most prevalent and deadly diseases affecting humans. The most widely used diagnostic tool to interrogate cardiovascular physiology and function is an electrocardiogram (ECG). Despite its widespread availability and use, the ECG is subject to interobserver variability and suboptimal sensitivity for asymptomatic or early-stage disease. Artificial intelligence (AI), particularly deep learning (DL) approaches, has provided a suite of methods to improve both the diagnostic and prognostic utility of the ECG in multiple cardiovascular domains. AI-enabled automated ECG interpretation (most commonly using convolutional neural networks (CNNs)) has reached and even surpassed expert-level performance for arrhythmia detection and classification. Additional data-driven approaches to ECG analysis have identified paroxysmal atrial fibrillation from a record of sinus rhythm ECGs, identified left ventricular systolic dysfunction, and predicted cardiac structure and ischemic burden (e.g., acute coronary syndromes). Pragmatic implementation has demonstrated higher diagnostic yield for asymptomatic left ventricular dysfunction in the primary care setting (EAGLE). Other emerging indications include expanded data-derived outputs, such as electrolyte disturbances, biological age, and cardiovascular risk prediction. Despite a growing list of promising applications, numerous translational hurdles remain before routine implementation. Generalizability is limited due to differences in training and target populations. Bias related to sex, race, and comorbidities is an important limiting factor to fair and equitable implementation. Other considerations include "black box" concerns with DL, clinical interpretability and adoption, medicolegal liability, and integration with clinical workflows and infrastructure. Related to these factors, data privacy, algorithmic fairness, accountability, and transparency are important to consider as AI-ECG continues to undergo regulatory scrutiny and outcomes-based validation. In conclusion, AI and ECG represent a major shift towards precision cardiology by improving prediction, screening, and early detection of cardiovascular disease. We anticipate continued improvements with prospective outcome studies, transparent and explainable approaches, and careful regulatory review to ensure safe and effective implementation in the clinic.