A CNN and Transformer Hybrid Network for Multi-Class Arrhythmia Detection from Photoplethysmography.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Wearable technology enables photoplethysmography (PPG)-based arrhythmia detection for early diagnosis.
• Current methods struggle with multi-class arrhythmia detection due to feature extraction challenges.
• Developing robust algorithms for identifying diverse arrhythmias from PPG signals is crucial.

Purpose of the Study:

• To introduce a novel hybrid convolutional neural network (CNN)-transformer model for multi-class arrhythmia detection.
• To enhance feature extraction by integrating local and global dependencies from PPG signals.
• To improve the accuracy of detecting multiple types of arrhythmias from PPG data.

Main Methods:

• A hybrid CNN-transformer network was developed, combining convolutional operations and self-attention mechanisms.
• A feature fusion layer with channel attention was implemented to integrate local and global PPG signal features.
• The model was evaluated on classifying sinus rhythm and five common arrhythmias.

Main Results:

• The hybrid model achieved high performance metrics: 87.0% precision, 87.1% recall, and 86.8% F1-score.
• The proposed method demonstrated superior performance compared to existing state-of-the-art techniques.
• Accurate classification of sinus rhythm and five distinct arrhythmia types was achieved.

Conclusions:

• The hybrid CNN-transformer network shows significant promise for accurate multi-class arrhythmia detection.
• This approach effectively captures discriminative features for improved arrhythmia classification from PPG signals.
• The findings support the potential of advanced AI models in wearable-based cardiac monitoring.