Artificial intelligence-powered advancements in atrial fibrillation diagnostics: a systematic review.

Abstract

Background: Cardiovascular diseases remain one of the leading causes of mortality worldwide, with atrial fibrillation emerging as a clinically significant arrhythmia. The increasing prevalence of atrial fibrillation calls for advanced diagnostic tools for accurate detection to reduce adverse consequences, such as stroke and heart failure. Cardiovascular advancements in artificial intelligence have improved the detection and management of atrial fibrillation.

Objective: This review examines recent advancements in atrial fibrillation detection using artificial intelligence-driven tools-such as wearables, neural networks, and machine learning-and highlights their clinical relevance, limitations, and potential to transform cardiovascular care.

Methodology: A systematic review was conducted using PubMed, IEEE Xplore, and ScienceDirect to identify peer-reviewed studies between 2020 and 2024. Original clinical studies using artificial intelligence were included for the diagnosis of atrial fibrillation. Studies on conditions other than atrial fibrillation or incomplete data were excluded. Factors analyzed across all studies included diagnostic application, key findings, clinical implications, and limitations of artificial intelligence approaches.

Results: This review evaluated 11 studies on artificial intelligence-enhanced tools for atrial fibrillation diagnostics. Neural networks showed the highest diagnostic accuracy, outperforming clinicians in retrospective electrocardiogram analyses (80% vs. 75%). Wearable artificial intelligence-integrated devices, such as electrocardiogram wristbands, offer the highest accessibility and real-time monitoring, with sensitivities exceeding 94%, although they are limited by single-lead input and patient compliance. Machine learning models, including random forest and XGBoost, showed moderate performance (AUROC 0.74-0.89) with strengths in risk prediction and stratification. Key challenges included limited generalizability, small-sample sizes, and varying model accuracy.

Conclusions: This review highlights the potential of artificial intelligence to improve atrial fibrillation diagnostics through wearable technologies, neural networks, and machine learning. While these tools often outperform traditional methods, real-world use is limited by small, retrospective studies and a lack of validation. Future work should focus on equity, transparency, and expanding artificial intelligence use beyond atrial fibrillation diagnosis, with collaboration needed to ensure safe, effective clinical integration.