Dynamic electrocardiogram detection and diagnosis based on improved dilated convolutional network.

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence in Healthcare

Background:

• Electrocardiograms (ECGs) are vital for cardiac assessment but traditional systems face limitations in portability and diagnostic intelligence.
• Existing ECG monitoring systems often lack the integration of real-time, portable data acquisition with advanced, automated diagnostic capabilities.

Purpose of the Study:

• To develop and evaluate a dynamic ECG monitoring and diagnostic system that combines portability and intelligent arrhythmia classification.
• To enhance the accuracy and reliability of ECG analysis through innovative signal processing and deep learning techniques.

Main Methods:

• A compact, portable ECG acquisition terminal (80 mm × 100 mm, ~150 g) with Wi-Fi connectivity for wireless data transmission.
• A deep learning model integrating hybrid dilated convolution and hybrid attention mechanisms for arrhythmia classification.
• Hierarchical Kalman filtering for efficient ECG signal processing and noise suppression using the AD8232 chip.

Main Results:

• The system demonstrated high accuracy in heart rate measurements (resting ±2%, exercise ±4%) and noise suppression.
• The proposed deep learning model achieved 99.28% per-beat classification accuracy for five-class arrhythmia detection (AAMI EC57 standard).
• Macro-averaged F1-score of 96.45% and MCC of 0.9561 across four key arrhythmia classes (N, S, V, F) indicate robust diagnostic performance.

Conclusions:

• The developed dynamic ECG system effectively integrates portability and intelligent diagnosis, offering a significant advancement over traditional methods.
• The hybrid deep learning architecture provides a stable and reliable approach for accurate arrhythmia classification, enhancing clinical utility.
• This portable and intelligent ECG system holds promise for improved remote cardiac monitoring and early disease detection.