Why the study?
Does an attention-based deep learning framework improve the detection of paroxysmal atrial fibrillation from ECG recordings compared to baseline models?
Population
2850 patients with 24-hour Holter Electrocardiogram recordings from the University of Virginia heart station.
Comparison
Attention-based deep learning framework applied… vs Baseline models
Design
Other
Key result
An attention-based bidirectional recurrent neural network detected paroxysmal atrial fibrillation from 24-hour Holter ECG recordings with an AUC of 0.94, exceeding baseline models.
Authors
Loading...
May support AI integration in ambulatory AF monitoring; leaves open prospective outcome validation before practice change.
Does an attention-based deep learning framework improve the detection of paroxysmal atrial fibrillation from ECG recordings compared to baseline models?
An attention-based deep learning framework achieved high accuracy (AUC 0.94) in detecting paroxysmal atrial fibrillation from Holter ECGs, demonstrating potential for integration into wearable sensors for long-term monitoring.
Effect estimate: AUC 0.94
Shashikumar et al. (2018) studied Paroxysmal Atrial Fibrillation (n=2,850). Attention-based bidirectional recurrent neural network vs. Baseline models was evaluated on Detection of paroxysmal AF episodes (AUC 0.94). An attention-based bidirectional recurrent neural network detected paroxysmal atrial fibrillation from 24-hour Holter ECG recordings with an AUC of 0.94, exceeding baseline models.