Multi-level Stress Assessment Using Multi-domain Fusion of ECG Signal | JoVE Visualize

Area of Science:

Biomedical Signal Processing
Machine Learning in Healthcare
Affective Computing

Background:

Current stress analysis using electrocardiogram (ECG) signals is limited to binary assessments, hindering comprehensive healthcare applications.
Existing research often analyzes ECG signals in isolation within spatial or transform domains, failing to leverage the benefits of domain fusion.
There is a need for advanced methods to assess multiple stress levels for more nuanced healthcare interventions.

Purpose of the Study:

To develop a novel deep learning framework for multi-level stress assessment using ECG signals.
To introduce a new dataset incorporating multiple stress levels for research and development.
To explore the efficacy of fusing spatial, time-frequency, and frequency domains for enhanced stress detection.

Main Methods:

ECG signals were converted into signal images based on R-R peaks, eliminating the need for manual feature extraction.
Signal images were transformed into multimodal and multi-domain representations using Gabor Wavelet Transform (GWT) for time-frequency and Discrete Fourier Transform (DFT) for frequency domains.
Convolutional Neural Networks (CNNs) were employed for feature extraction from these diverse signal representations, followed by decision-level fusion to improve classification accuracy.

Main Results:

The proposed method achieved an average accuracy of 85.45% in classifying multiple stress levels.
The fusion of different domains (spatial, time-frequency, frequency) significantly improved stress classification performance.
The ECG signal-to-image conversion approach proved effective for deep learning-based stress analysis.

Conclusions:

The developed deep learning framework offers a promising approach for accurate, multi-level stress assessment using ECG signals.
Domain fusion and ECG signal-to-image conversion are effective strategies for enhancing stress detection capabilities.
This research paves the way for more sophisticated, data-driven stress monitoring in healthcare settings.