12-Lead ECG arrhythmia classification using cascaded convolutional neural network and expert feature.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Automatic electrocardiogram (ECG) analysis is crucial for diagnosing cardiac arrhythmias.
• Deep learning models show promise but are often limited to single-lead ECG analysis, neglecting multi-lead correlations.
• Multi-lead ECG provides a more comprehensive view of cardiac electrical activity.

Purpose of the Study:

• To develop and validate an automated 12-lead ECG arrhythmia classification method.
• To leverage both deep learning and expert-derived features for enhanced diagnostic performance.
• To address the limitations of single-lead ECG analysis in deep learning models.

Main Methods:

• A cascaded convolutional neural network (CCNN) was employed, utilizing 1D CNNs for single-lead feature extraction.
• Features were concatenated to capture temporal and spatial correlations for input into 2D ResNet blocks.
• Expert knowledge-based features were extracted and classified using a random forest, with results combined via stacking.

Main Results:

• The proposed method achieved a final score of 86.5% in the China ECG Intelligence Challenge.
• Successfully classified multi-label 12-lead ECG data into 9 distinct arrhythmia categories.
• Demonstrated superior performance compared to methods relying solely on single-lead ECG data.

Conclusions:

• The combined CCNN and expert feature approach offers a robust solution for 12-lead ECG arrhythmia classification.
• Integrating multi-lead information and diverse feature extraction techniques improves diagnostic accuracy.
• This method holds significant potential for clinical application in automated cardiac arrhythmia detection.