Classification of multi-lead ECG based on multiple scales and hierarchical feature convolutional neural networks

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Arrhythmia detection is crucial for diagnosing cardiovascular diseases.
• Current deep learning methods struggle to integrate morphological and temporal ECG features effectively.
• A novel approach is needed to enhance multi-lead ECG classification accuracy.

Purpose of the Study:

• To develop and validate a deep learning model for accurate multi-lead ECG classification.
• To address the limitations of existing methods in integrating ECG morphological and temporal features.
• To improve the detection and classification of arrhythmias for better cardiovascular disease diagnosis.

Main Methods:

• A Convolutional Neural Network (CNN) incorporating mixed scales and hierarchical features was proposed.
• The Lead Encoder Attention (LEA) mechanism was integrated for multi-lead ECG classification.
• The model was validated on the MIT-BIH Arrhythmia (MIT-BIH-AR) and Chinese Cardiovascular Disease Database (CCDD) using intrapatient and interpatient approaches.

Main Results:

• The model achieved 99.5% accuracy in classifying normal and abnormal heartbeats on the MIT-BIH-AR database.
• On the CCDD, the model achieved a TPR95 of 78.5% and an accuracy of 88.5% for normal/abnormal ECG record classification.
• Cross-dataset experiments demonstrated the model's strong generalization capability.

Conclusions:

• The proposed CNN with LEA mechanism effectively integrates ECG morphological and temporal features for improved arrhythmia classification.
• The model shows high accuracy and strong generalization, offering a promising tool for cardiovascular disease diagnosis.
• This approach advances deep learning applications in analyzing complex ECG data for clinical use.