Atrial fibrillation stands as the prevailing cardiac arrhythmia globally, affecting approximately 59 million individuals in 2019. This irregular heartbeat poses escalated risks of heart failure, dementia, and stroke, imposing a substantial burden on healthcare systems. Consequently, the early detection and treatment of this condition emerge as paramount objectives.
Researchers from the Luxembourg Centre for Systems Biomedicine (LCSB) at the University of Luxembourg have recently unveiled a groundbreaking deep-learning model adept at predicting the shift from a normal cardiac rhythm to atrial fibrillation. This model offers early warnings, typically 30 minutes prior to onset, with an accuracy rate hovering around 80%. These findings, disseminated in the scientific journal Patterns, herald a promising avenue for integration into wearable technologies, facilitating early interventions and optimizing patient outcomes.
“We utilized heart rate data to train a deep learning model capable of discerning various phases—sinus rhythm, pre-atrial fibrillation, and atrial fibrillation—and computing a ‘probability of danger’ indicating the likelihood of an imminent episode,” elucidated Prof. Jorge Goncalves, leading the Systems Control group at LCSB.
Atrial fibrillation manifests as irregular upper chamber heartbeats misaligned with the ventricles, often necessitating intensive interventions ranging from electrical shock treatment to surgical ablation. Foreseeing an impending episode of atrial fibrillation empowers patients to enact preventive measures to stabilize their cardiac rhythm. However, prevailing methods reliant on heart rate and electrocardiogram (ECG) analysis only detect atrial fibrillation moments before its onset, lacking preemptive capabilities.
“In contrast, our approach pivots towards a more proactive prediction model,” Prof. Goncalves explained. “By leveraging heart rate data, our deep learning model forecasts various phases of cardiac rhythm, culminating in an early warning once the probability surpasses a designated threshold.”
This artificial intelligence model, christened WARN (Warning of Atrial fibRillatioN), underwent rigorous training and testing on 24-hour recordings from 350 patients at Tongji Hospital (Wuhan, China). Impressively, WARN issued early warnings, on average, 30 minutes ahead of atrial fibrillation onset, boasting remarkable accuracy. Notably, WARN represents the first method to furnish warnings significantly prior to onset, distinguishing it from prior research on arrhythmia prediction.
“Moreover, our model demonstrates exceptional performance utilizing solely R-to-R intervals, essentially heart rate data obtainable from user-friendly, affordable pulse signal recorders like smartwatches,” emphasized Dr. Marino Gavidia, the publication’s lead author and former member of the Systems Control group.
Furthermore, the researchers envisage implementing their deep-learning model into smartphones to process data from smartwatches, capitalizing on its minimal computational requirements. This seamless integration into wearable technologies aligns with the overarching goal of enabling patients to perpetually monitor their cardiac rhythm and receive timely warnings, affording ample opportunity for preemptive measures such as antiarrhythmic medication or targeted treatments.
“As we forge ahead, our focus shifts towards developing personalized models,” Prof. Goncalves iterated. “The daily utilization of a simple smartwatch furnishes continual insights into individual heart dynamics, enabling us to iteratively refine and enhance our model’s performance, ultimately leading to earlier warnings and improved patient outcomes.”
In essence, this innovative approach holds promise for revolutionizing cardiac care, potentially catalyzing novel clinical trials and therapeutic interventions.
- : public
