Non-Direct Contact ECG Signal Classification Using a Hybrid Deep Learning Framework With Validation in Bedside Heart...

Area of Science:

• Biomedical Engineering
• Wearable Health Technology
• Signal Processing

Background:

• Growing demand for smart healthcare necessitates accurate, comfortable bedside ECG monitoring.
• Conventional wet electrodes cause skin irritation, limiting long-term use.
• Non-direct contact ECG methods offer potential for improved patient comfort and compliance.

Purpose of the Study:

• To present and validate a non-direct contact ECG recording system using capacitive coupling (cECG).
• To assess the system's accuracy in capturing Heart Rate Variability (HRV) during overnight sleep.
• To develop and evaluate a deep learning framework for cECG signal quality assessment (SQA) and noise filtering.

Main Methods:

• Developed a bedside cECG system that records ECG signals through clothing.
• Implemented a deep learning model for cECG signal quality assessment, noise reduction, and on/off-bed detection.
• Collected overnight sleep data from 6 subjects, comparing cECG-derived HRV features with traditional wet electrode ECG.

Main Results:

• The deep learning model achieved high accuracy (94.7%) for SQA and excellent on/off-bed monitoring accuracy (99.79%).
• cECG-derived HRV features demonstrated strong correlations with reference wet electrode ECG data.
• Mean Absolute Percentage Errors (MAPE) for HRV features were generally low, with PNN50 (8.15%), HF (13.25%), and SD1 (5.18%) showing the largest values.

Conclusions:

• The proposed cECG system provides a reliable, comfortable alternative for bedside ECG recording.
• The integrated deep learning framework effectively enhances signal quality and enables accurate HRV analysis.
• This technology offers a promising solution for continuous, non-invasive sleep monitoring and HRV assessment.