Deep-Learning Model Predicts Arrhythmia 30 Minutes before Onset

Atrial fibrillation, the most common type of cardiac arrhythmia worldwide, affected approximately 59 million people in 2019. Characterized by an irregular and often rapid heart rate, atrial fibrillation occurs when the heart's upper chambers (atria) beat out of sync with the lower chambers (ventricles). Addressing arrhythmia can require aggressive interventions such as electrically shocking the heart back to a normal rhythm or surgically removing areas that generate faulty signals. Associated with increased risks of heart failure, dementia, and stroke, atrial fibrillation presents significant challenges to healthcare systems, emphasizing the importance of early detection and treatment. Traditional detection methods, relying on heart rate and electrocardiogram (ECG) data, typically identify atrial fibrillation just before its onset, offering no advanced warning.

Now, researchers from the Luxembourg Centre for Systems Biomedicine (LCSB) of the University of Luxembourg (Esch-sur-Alzette, Luxembourg) have achieved a breakthrough with the development of an advanced deep-learning model that can predict the onset of atrial fibrillation. Their model, named WARN (Warning of Atrial fibRillatioN), successfully provides early warnings about 30 minutes before atrial fibrillation begins, with approximately 80% accuracy.

Image: The deep-learning model can predict arrhythmia 30 minutes before it happens (Photo courtesy of 123RF)

This innovative model was trained and tested using 24-hour recordings from 350 patients, marking a significant improvement over previous prediction methods by offering a much earlier warning. The potential to integrate this technology into wearable devices could transform patient management, allowing for preemptive interventions that enhance outcomes. Notably, WARN stands out as the first method to offer a substantial lead time before the onset of atrial fibrillation, setting a new standard in arrhythmia prediction.

“Our work departs from this approach to a more prospective prediction model,” said Prof. Jorge Goncalves, head of the Systems Control group at the LCSB. “We used heart rate data to train a deep learning model that can recognize different phases – sinus rhythm, pre-atrial fibrillation and atrial fibrillation – and calculate a “probability of danger” that the patient will have an imminent episode.”

Related Links:
University of Luxembourg