An attention-based multimodal deep learning framework integrating EEG and ECG for enhanced stress detection.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Chronic stress poses a significant public health risk, impacting physical and mental well-being.
• Current stress assessment methods are limited by subjectivity and isolated physiological measures.
• Objective, accurate stress detection is crucial for timely intervention and support.

Purpose of the Study:

• To develop a novel, objective method for assessing psychological stress levels.
• To integrate Electroencephalogram (EEG) and Electrocardiogram (ECG) data using deep transfer learning.
• To enhance the accuracy and reliability of stress detection beyond traditional approaches.

Main Methods:

• A deep transfer learning model combining VGG16, EfficientNetB0, and ResNeXt50 was employed.
• EEG data was transformed into time-frequency maps via wavelet transformations.
• ECG data utilized time-series variability and raw signal patterns.
• An attention-based fusion layer integrated EEG and ECG data for stress assessment.

Main Results:

• The unified EEG-ECG model achieved 95.7% accuracy in identifying neutral, tense, and positive stress states.
• This performance significantly outperformed EEG-only (82.3%) and ECG-only (85.6%) methods.
• The system demonstrated high flexibility and cross-condition operational capability.
• EEG and ECG data were shown to mutually enhance each other's stress assessment capabilities.

Conclusions:

• The proposed deep learning approach offers a highly reliable and objective method for stress assessment.
• This technology can significantly support medical diagnostics, employee wellness programs, and personalized psychological care.
• Combining multi-modal physiological data with advanced AI enhances the precision of psychological state monitoring.