Detection of Atrial Fibrillation based on Feature Fusion using Attention-based BiLSTM.

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence

Background:

• Atrial fibrillation (AF) is a prevalent cardiac arrhythmia requiring early detection for effective treatment.
• Traditional methods like Electrocardiogram (ECG) are often invasive and equipment-dependent.
• Photoplethysmography (PPG) presents a non-invasive alternative for monitoring heart rhythm.

Purpose of the Study:

• To develop and evaluate an advanced feature fusion technique for accurate AF detection.
• To leverage attention-based Bidirectional Long Short-Term Memory (BiLSTM) networks with PPG signals.
• To improve upon existing methods for non-invasive AF diagnosis.

Main Methods:

• Extraction of time-domain (TD) and frequency-domain (FD) features from PPG signals.
• Generation of deep learning features using an attention-based BiLSTM network.
• Fusion of extracted TD, FD, and deep learning features, followed by classification using a softmax function.

Main Results:

• The proposed feature fusion method achieved a high accuracy of 96.5% for AF detection.
• Excellent performance metrics were recorded: 93.20% recall, 94.50% precision, and 93.09% F-score.
• The approach demonstrated significant improvements in AF prediction and diagnostic capabilities.

Conclusions:

• The attention-based BiLSTM and PPG feature fusion method offers a highly accurate and non-invasive approach for AF detection.
• This technique can significantly support clinicians in the early diagnosis and management of atrial fibrillation.
• The study highlights the potential of advanced AI in enhancing cardiovascular diagnostics.