A CNN Model for Cardiac Arrhythmias Classification Based on Individual ECG Signals.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Wearable electrocardiogram (ECG) devices offer an alternative to conventional medical devices in connected home healthcare.
• Challenges remain, particularly noise interference from user mobility, impacting the reliability of wearable ECG monitoring.
• Robust classification models are needed for accurate cardiac arrhythmia detection in real-world wearable applications.

Purpose of the Study:

• To develop an insensitive and robust classification model for cardiac arrhythmia detection using wearable ECG data.
• To address the limitations of existing systems, specifically noise sensitivity caused by user mobility.

Main Methods:

• A one-dimensional seven-layer convolutional neural network (CNN) was designed for automatic feature extraction and classification.
• The model was trained and evaluated on a large volume of original, noisy ECG signals.
• A record-based ten-fold cross-validation scheme was employed to ensure model robustness and independence of training and testing datasets.

Main Results:

• The CNN model achieved high diagnostic accuracy (0.9874 for original, 0.9876 for de-noised signals).
• Excellent sensitivity (0.9811/0.9813) and specificity (0.9905/0.9907) were recorded for both original and de-noised signals.
• The model demonstrated superior performance compared to recent literature in cardiac arrhythmia classification.

Conclusions:

• The proposed CNN approach is effective for detecting cardiac arrhythmias from wearable ECG signals, performing well on both noisy and de-noised data.
• The model's robustness and accuracy make it suitable for integration into wearable ECG monitoring systems and connected home healthcare applications.
• This AI-driven solution reduces computational workload while maintaining high diagnostic performance.