Interpatient ECG Heartbeat Classification with an Adversarial Convolutional Neural Network.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Deep neural networks (DNNs) excel at extracting features from electrocardiogram (ECG) data for classification.
• Generalizing DNN models to new patients is challenging due to intersubject variability in ECG signals.
• Current methods often require large datasets from numerous subjects to achieve robust generalization.

Purpose of the Study:

• To develop a method for learning subject-invariant ECG features from limited patient data.
• To improve the generalization of ECG classification models in the presence of intersubject variability.
• To enhance the accuracy of classifying different types of heartbeats, particularly ectopic beats.

Main Methods:

• An adversarial deep neural network framework was proposed, integrating adversarial learning with a convolutional neural network (CNN).
• The framework aims to learn features that are invariant to individual patient differences (intersubject variability) while remaining discriminative for heartbeat classification.
• The model was trained and evaluated on the publicly available MIT-BIH arrhythmia database.

Main Results:

• The proposed method achieved state-of-the-art performance in detecting supraventricular ectopic beats (SVEBs), with a sensitivity of 78.8% and precision of 90.8%.
• Comparable performance was observed for the detection of ventricular ectopic beats (VEBs), showing a sensitivity of 92.5% and precision of 94.3%.
• The approach demonstrated effectiveness in ECG classification tasks with a limited number of subjects.

Conclusions:

• The adversarial deep neural network framework successfully learns subject-invariant features for ECG classification.
• This approach offers a promising solution for accurate arrhythmia detection, especially when dealing with limited patient data.
• The method enhances the generalization capability of models for real-world clinical applications.