An ECG Stitching Scheme for Driver Arrhythmia Classification Based on Deep Learning.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Electrocardiogram (ECG) monitoring in vehicles is challenging due to noise from vibrations and driver interaction.
• Detecting driver arrhythmias is crucial for road safety.
• Existing methods struggle with noisy, fragmented ECG data acquired during driving.

Purpose of the Study:

• To develop an ECG signal stitching scheme for reliable arrhythmia detection in drivers.
• To enhance the classification accuracy of arrhythmias from noisy, in-vehicle ECG signals.
• To investigate the effectiveness of deep learning models for analyzing stitched ECG data.

Main Methods:

• ECG signal preprocessing, R-peak detection, and TP interval segmentation.
• A novel P-peak estimation method for abnormal beat identification.
• Transformation of ECG segments using Continuous Wavelet Transform (CWT) and Short-Time Fourier Transform (STFT).
• Classification using Convolutional Neural Networks (CNNs) with transfer learning (GoogleNet).

Main Results:

• The proposed ECG stitching scheme successfully processed noisy driving data.
• GoogleNet achieved the highest classification accuracy (82.39%) using CWT-transformed stitched ECG data.
• Original, non-stitched ECG data yielded a higher classification accuracy (88.99%).

Conclusions:

• The ECG stitching scheme offers a viable method for arrhythmia detection in noisy driving environments.
• Deep learning models, particularly GoogleNet with CWT, show promise for analyzing processed ECG signals.
• Further research may improve stitching accuracy to match original data performance for enhanced driver safety.