A Reparameterization Multifeature Fusion CNN for Arrhythmia Heartbeats Classification.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Arrhythmia classification is crucial for diagnosing heart conditions.
• Existing methods often struggle with complex heartbeat patterns.
• Deep learning offers potential for improved accuracy.

Purpose of the Study:

• To propose a novel deep learning method for arrhythmia heartbeats classification.
• To enhance classification accuracy by fusing multiple feature types.
• To develop a lightweight yet powerful deep learning model.

Main Methods:

• Extracted stationary wavelet transforms (SWT) and RR interval features.
• Designed a parallel multibranch convolutional neural network (1D-CNN).
• Employed reparameterization to create a lightweight network architecture.

Main Results:

• The proposed multifeature fusion method significantly outperformed existing techniques.
• The network achieved high accuracy in classifying arrhythmic heartbeats.
• The reparameterization strategy resulted in an efficient, lightweight model.

Conclusions:

• The novel deep learning approach provides a powerful tool for arrhythmia heartbeats classification.
• Multifeature fusion and optimized network design are key to improved performance.
• This method holds promise for clinical applications in cardiac monitoring.