Convolutional squeeze-and-excitation network for ECG arrhythmia detection

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Automatic arrhythmia detection from electrocardiograms (ECG) is crucial for cardiovascular disease management.
• Convolutional neural networks (CNNs) process ECG signals into feature channels, but can include redundant information impacting diagnosis.
• Feature channels require adaptive weighting to prioritize arrhythmia-specific information.

Purpose of the Study:

• To introduce the Squeeze-and-Excitation (SE) block for the first time in ECG-based automatic arrhythmia detection.
• To develop a model combining residual convolutional modules and SE blocks for enhanced ECG feature extraction.
• To improve the accuracy and efficiency of multi-type arrhythmia diagnosis.

Main Methods:

• Integration of the SE block with a residual convolutional module for ECG signal analysis.
• Utilizing 1D convolution for temporal information extraction and shortcut connections for network optimization.
• Explicitly modeling channel interdependencies to adaptively enhance discriminative features and suppress noise.

Main Results:

• The proposed model effectively extracts discriminative arrhythmia features from multiple feature channels.
• No additional data preprocessing, such as denoising, is required, preserving biological information.
• Achieved high performance on the CPSC 2018 and PhysioNet/CinC 2017 ECG datasets.

Conclusions:

• The SE-Residual convolutional model demonstrates significant potential for clinical application in automatic arrhythmia detection.
• The method enhances diagnostic efficiency by adaptively integrating ECG feature channel information.
• The model's robust feature extraction capabilities minimize the need for preprocessing, improving workflow.