A Smartphone-Based M-Health Monitoring System for Arrhythmia Diagnosis.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Deep learning models for arrhythmia detection face challenges with temporal information, numerous parameters, and noise sensitivity.
• Existing methods struggle with accurate diagnosis from noisy long-term electrocardiogram (ECG) signals.

Purpose of the Study:

• To develop a smartphone-based m-health system for accurate and efficient arrhythmia diagnosis.
• To overcome limitations of current deep learning models in ECG signal processing and analysis.

Main Methods:

• A cycle-Generative Adversarial Network (GAN) based model was designed for ECG denoising, utilizing unsupervised pre-training for faster convergence.
• A Time Convolutional Network (TCN) model was developed for arrhythmia classification, capable of identifying 34 common arrhythmia types from eight-lead ECG data.
• The TCN model was deployed on the Android platform for an at-home ECG monitoring application.

Main Results:

• The proposed ECG denoising model effectively removes real-world noise, outperforming existing noise reduction techniques.
• The TCN-based arrhythmia diagnosis model achieved high recognition accuracy for 34 common arrhythmia events.
• The system demonstrated superior performance in denoising, accuracy, model size, and operational speed compared to existing methods.

Conclusions:

• The developed smartphone-based m-health system provides a robust solution for arrhythmia diagnosis, suitable for mobile deployment.
• The combination of cycle-GAN for denoising and TCN for diagnosis offers significant improvements for at-home ECG monitoring services.