An enhanced hypergraph CNN with adaptive focal loss for automated ECG heartbeat classification

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Automated cardiovascular disease (CVD) diagnosis using electrocardiogram (ECG) analysis shows promise.
• Current deep learning methods face challenges like class imbalance, temporal dependencies, and modeling inter-beat relationships, limiting clinical use.

Purpose of the Study:

• To develop a hybrid deep learning framework for robust automated CVD diagnosis.
• To address limitations of existing models in handling class imbalance and complex ECG patterns.

Main Methods:

• A hybrid framework combining Convolutional Neural Network (CNN) for feature extraction, k-nearest neighbor hypergraph construction for inter-beat dependencies, and an enhanced hypergraph neural network (EHGNN).
• Implementation of focal loss with adaptive class weighting to manage class imbalance.
• Utilized MIT-BIH Arrhythmia and St. Petersburg INCART 12-lead Arrhythmia databases for evaluation.

Main Results:

• Achieved 98.66% accuracy for five-class classification on the MIT-BIH dataset (AAMI system).
• Attained 95.46% accuracy for three-class arrhythmia detection on the INCART database.
• Demonstrated consistent performance across datasets, indicating strong generalization capabilities.

Conclusions:

• The proposed framework effectively handles class imbalance, temporal dependencies, and inter-beat relationships in ECG data.
• The model shows significant potential for clinical diagnostic systems by capturing intricate heartbeat patterns.
• This approach offers a robust solution for automated CVD diagnosis, overcoming limitations of standard models.