Deepaware: A hybrid deep learning and context-aware heuristics-based model for atrial fibrillation detection

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Current automatic atrial fibrillation (AF) detection models using RR-interval (RRI) features perform well on benchmark datasets but exhibit high false positive rates (FPRs) in real-world ambulatory electrocardiogram (ECG) monitoring.
• These models struggle with non-AF arrhythmias and data collected under free-living conditions.

Purpose of the Study:

• To introduce DeepAware, a hybrid deep learning (DL) and context-aware heuristics (CAH) model, designed to effectively reduce FPRs and enhance AF detection performance.
• To evaluate DeepAware's efficacy on ambulatory ECG data from free-living conditions.

Main Methods:

• DeepAware integrates RRI features, P-wave analysis, and contextual features from ambulatory ECG data.
• The model combines deep learning algorithms with context-aware heuristics to improve detection accuracy and reduce false positives.

Main Results:

• DeepAware demonstrates superior generalizability and performance compared to state-of-the-art models on unseen ECG AF datasets.
• On ambulatory ECG recordings, DeepAware achieved high sensitivity (97.94%), specificity (98.39%), and accuracy (98.06%).
• Analysis confirmed that incorporating P-wave detection and CAH significantly reduces FPRs compared to RRI-based models alone.

Conclusions:

• The DeepAware model offers a substantial reduction in manual review workload for physicians by minimizing false positives.
• This facilitates more efficient long-term ambulatory monitoring for the early detection of atrial fibrillation.