HARDC : A novel ECG-based heartbeat classification method to detect arrhythmia using hierarchical attention based...

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Deep learning models analyze ECG signals for cardiac arrhythmia detection.
• Traditional models face challenges with contextual correlation and gradient dispersion.

Purpose of the Study:

• To develop a novel hybrid model for improved arrhythmia classification.
• To address limitations of existing dilated CNN models.

Main Methods:

• Developed a hybrid hierarchical attention-based bidirectional recurrent neural network with dilated CNN (HARDC).
• Utilized dilated CNN and BiGRU-BiLSTM for feature fusion.
• Incorporated a hierarchical attention mechanism for enhanced prediction.
• Applied data preprocessing including Z-Score normalization, filtering, denoising, segmentation, and CGAN for synthetic data generation.

Main Results:

• HARDC model achieved 99.60% accuracy, 98.21% F1 score, 97.66% precision, and 99.60% recall on MIT-BIH generated ECG data.
• Demonstrated significantly improved classification results and feature extraction interpretability.
• Reduced runtime compared to normal convolution using dilated CNN.
• Outperformed existing models in arrhythmia classification.

Conclusions:

• The HARDC model offers an innovative and cost-effective strategy for ECG analysis.
• This hybrid approach shows significant promise for automated, high-performance arrhythmia classification.
• The method provides enhanced interpretability for feature extraction in ECG signals.