Two-dimensional ECG-based cardiac arrhythmia classification using DSE-ResNet.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electrocardiograms (ECGs) are crucial for diagnosing cardiac arrhythmias.
• Deep neural networks show promise for automated ECG analysis.
• Existing models often neglect inter-lead feature extraction from 12-lead ECGs.

Purpose of the Study:

• To develop a general deep learning model for classifying normal sinus rhythm and 8 types of cardiac arrhythmias.
• To address the limitation of lacking inter-lead feature extraction in current models.
• To improve the accuracy and reliability of automated cardiac arrhythmia detection.

Main Methods:

• A novel model, DSE-ResNet, utilizing a two-dimensional ECG representation and ResNet with detached squeeze-and-excitation modules.
• Splicing the 12-lead ECG into a 2D plane to enable simultaneous internal and inter-lead feature extraction.
• Employing orthogonal experiment design for hyper-parameter optimization and a multi-model voting strategy for enhanced classification.

Main Results:

• The DSE-ResNet model achieved an average [Formula: see text] for classifying normal rhythm and 8 cardiac arrhythmias on the CPSC2018 dataset.
• The model demonstrated superior performance, achieving the best [Formula: see text] in 2 sub-abnormal types compared to state-of-the-art models.
• The 2D ECG representation combined with DSE-ResNet effectively captures both intra-lead and inter-lead information.

Conclusions:

• The proposed 2D ECG and DSE-ResNet model offers advantages in detecting specific cardiac arrhythmias.
• This approach has significant potential as an auxiliary tool to assist clinicians in cardiac arrhythmia analysis.
• The model's ability to extract inter-lead features enhances its diagnostic capabilities for complex cardiac conditions.