A systematic survey of artificial intelligence methods for ECG-based cardiovascular disease prediction

Abstract

Cardiovascular diseases continue to be the leading cause of death globally, creating an urgent need for accurate and scalable diagnostic approaches. The electrocardiogram remains essential for cardiac assessment, yet the growing demand for expert interpretation places increasing strain on healthcare systems and introduces potential diagnostic variability. This systematic review examines how artificial intelligence methods are being applied to ECG-based cardiovascular disease prediction, with particular attention to data handling practices, modeling approaches, interpretability techniques, and evaluation strategies. The review aims to provide researchers with a comprehensive understanding of the field’s current state while identifying persistent challenges and opportunities for meaningful progress. Following PRISMA guidelines, we systematically searched PubMed, IEEE Xplore, Scopus, Web of Science, and other major databases for peer-reviewed studies published between January 2020 and November 2025. After screening 1247 records, 83 high-quality studies met our inclusion criteria and underwent rigorous quality assessment using a customized evaluation tool addressing data sources, preprocessing, model development, and clinical applicability. The analysis reveals that while deep learning approaches—particularly convolutional neural networks and hybrid architectures—have demonstrated impressive performance in controlled settings, significant gaps remain. Only 38.6% of studies showed low risk of bias, with external validation lacking in 72% of cases. Dataset diversity remains problematic, with 78% of data originating from North America or Europe and minimal representation of pediatric populations. Explainable AI methods appear in only 52% of recent studies, and demographic fairness assessments remain rare. This review provides a structured roadmap for researchers navigating the entire AI-ECG pipeline, from data acquisition through model deployment. By systematically identifying methodological gaps and proposing concrete solutions, we aim to accelerate the development of reliable, fair, and clinically useful AI systems for cardiovascular care.