ECG signal classification in wearable devices based on compressed domain

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Wearable devices generate large ECG datasets, challenging arrhythmia detection speed and accuracy.
• Deep Compressed Sensing (DCS) optimizes ECG monitoring but involves complex and costly reconstruction.
• Existing methods struggle with efficient and accurate ECG signal processing for arrhythmia diagnosis.

Purpose of the Study:

• To propose an improved classification scheme for deep compressed sensing models in ECG monitoring.
• To enhance the efficiency and accuracy of arrhythmia detection using wearable devices.
• To address the complexity and cost associated with traditional DCS reconstruction.

Main Methods:

• Developed a four-module framework: pre-processing, adaptive compression, and classification.
• Implemented adaptive compression in three convolutional layers for normalized ECG signals.
• Utilized a direct classification network on compressed data for four types of ECG signals.

Main Results:

• Achieved 98.16% accuracy, 98.28% average accuracy, 98.09% sensitivity, and 98.06% F1-score at a 0.2 compression ratio.
• Demonstrated superior performance compared to existing models on MIT-BIH and Ali Cloud Tianchi ECG databases.
• Validated the model's robustness and effectiveness in ECG signal analysis.

Conclusions:

• The proposed improved DCS classification scheme significantly enhances ECG monitoring for arrhythmia diagnosis.
• The model offers a more efficient, accurate, and potentially cost-effective solution for wearable ECG analysis.
• This approach holds promise for advancing real-time cardiac health monitoring.