Multi-information fusion neural networks for arrhythmia automatic detection

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Electrocardiograms (ECGs) are crucial for diagnosing arrhythmias, which often manifest as abnormal beat morphologies and irregular intervals.
• Analyzing single ECG beats is challenging for simultaneous morphological and temporal information due to the absence of RR intervals.
• Existing methods struggle to integrate beat morphology and temporal dynamics for accurate arrhythmia detection.

Purpose of the Study:

• To introduce a novel Multi-information Fusion Convolutional Bidirectional Recurrent Neural Network (MF-CBRNN) for automated arrhythmia detection.
• To address the limitation of integrating morphological and temporal information from single ECG beats.
• To enhance the accuracy and efficiency of computer-aided diagnostic systems for cardiac arrhythmias.

Main Methods:

• The MF-CBRNN employs two parallel hybrid branches to process beat-based morphological and segment-based temporal ECG information.
• Each branch integrates Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (BLSTM) networks to extract features.
• Features from both branches are fused for comprehensive information aggregation and analysis.

Main Results:

• The MF-CBRNN achieved 99.56% accuracy and 96.40% F1-score in the intra-patient paradigm using MIT-BIH ECG data.
• The model demonstrated strong performance in the inter-patient paradigm with 96.77% accuracy and 77.83% F1-score.
• These results highlight the model's effectiveness in diverse patient scenarios.

Conclusions:

• The proposed MF-CBRNN model represents a significant advancement in automated arrhythmia detection.
• It achieves state-of-the-art performance compared to existing studies.
• The MF-CBRNN shows great promise as a component of future computer-aided diagnostic systems for cardiac conditions.