A multiview feature fusion model for heartbeat classification.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Electrocardiograms (ECG) are crucial for diagnosing cardiac arrhythmias based on waveform analysis.
• Existing deep learning methods often struggle with classification accuracy for new individuals due to single-view feature limitations.
• Accurate arrhythmia classification is essential for timely diagnosis and patient management.

Purpose of the Study:

• To develop a multiview fusion classification model for enhanced ECG analysis.
• To improve the accuracy of classifying five types of heartbeats: normal (N), left bundle branch block (LB), right bundle branch block (RB), atrial premature contraction (APC), and premature ventricular contraction (PVC).
• To provide a robust algorithmic model for single-lead wearable ECG devices.

Main Methods:

• Extraction of handcrafted and deep learning-based view features from heartbeats.
• Fusion of multi-perspective features in a fully connected layer.
• Implementation of a Random Forest classifier with Bayesian optimization for hyper-parameter tuning.
• Validation using the MIT-BIH database for inter-patient classification.

Main Results:

• Achieved high average accuracy (98.93%), precision (96.92%), sensitivity (96.46%), and specificity (99.33%) for classifying five heartbeat types.
• Demonstrated superior performance in inter-patient classification, crucial for real-world applications.
• The multiview fusion approach significantly boosted classification performance compared to single-view methods.

Conclusions:

• The proposed multiview fusion framework effectively enhances ECG detection performance for new individuals.
• This approach offers a practical and accurate algorithmic solution for arrhythmia detection in single-lead wearable devices.
• The study highlights the potential of feature fusion techniques in advancing automated cardiac diagnostics.