[An arrhythmia classification method based on deep learning parallel network model].

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Arrhythmia classification is crucial for cardiovascular health.
• Existing methods face limitations in accurately identifying diverse arrhythmia types.
• Advanced deep learning approaches are needed to improve diagnostic performance.

Purpose of the Study:

• To develop and evaluate a parallel neural network classification method for improved arrhythmia detection.
• To classify four types of heartbeats: normal, supraventricular ectopic, ventricular ectopic, and fused beats.

Main Methods:

• ECG signal denoising, heartbeat segmentation, and data enhancement were performed.
• A densely connected convolutional network (DCNN) was combined with bidirectional long short-term memory (BiLSTM) and efficient channel attention (ECA) networks.
• A parallel network structure processed both small-scale and large-scale heartbeat waveform features for classification using Softmax.

Main Results:

• The proposed model achieved 99.36% overall accuracy, 96.08% average sensitivity, and 99.41% average specificity on the MIT-BIH Arrhythmia Database.
• Comparative experiments validated the superior performance of the parallel network architecture.
• The model demonstrated efficient convergence with a training time of 41 seconds.

Conclusions:

• The parallel multi-network approach effectively improves arrhythmia classification accuracy, sensitivity, and specificity.
• This method offers a promising, efficient technical solution for clinical arrhythmia diagnosis.
• The study highlights the potential of integrated deep learning models in cardiovascular diagnostics.