Stress Estimation in VRET Using PPG

Detecting Stress Levels in Virtual Reality Therapy with Photoplethysmography Signals

Overview

Stress Estimation in VRET Using PPG Signals is a project designed to detect and estimate stress levels in subjects during Virtual Reality Exposure Therapy (VRET) by analyzing Photoplethysmography (PPG) signals. This platform uses machine learning and deep learning to classify mental states—relaxed or stressed—in real-time, providing a foundation for adaptive, personalized therapy.

In this study, 19 subjects were exposed to relaxing and stressful VR environments. The system automatically processes PPG data and demographic information to estimate stress levels, making it a non-invasive solution for real-time stress monitoring and therapy adjustment. The recent update includes a deep learning method and subject information embedding for improved accuracy and adaptability.

Key Features

  • PPG Signal Stress Detection: The platform uses PPG signals to estimate the subject’s mental state, which can be used to provide real-time feedback to personalize therapy sessions based on the subject’s stress levels.

  • Machine Learning and Deep Learning Models: The platform uses both traditional machine learning methods (e.g., SVM, Random Forest, LDA) and deep learning models to classify stress states with up to 65% accuracy. The deep learning method enables better detection of stress levels by utilizing more complex patterns in the data.

  • Human Representation: Demographic data, including age, gender, VR and gaming experience, and time since last stimulant intake, is embedded to enhance the model’s performance and provide more personalized predictions.

  • Adaptive Therapy Potential: This stress estimation system can be integrated into VRET platforms to dynamically adapt the therapy environment based on the patient’s real-time stress levels, enhancing its effectiveness.

Study Details

  • Participants: 19 healthy subjects were exposed to a relaxing and a stressful VR environment, and their PPG signals were collected.
  • Signal Processing: The PPG data was segmented into windows with overlapping intervals. Machine learning models were trained using Leave-One-Subject-Out (LOSO) cross-validation to ensure the system’s generalization to unseen subjects.
  • Results: The Deep Learning model achieved a 62% accuracy, which increased to 65% when demographic information was incorporated into the model through embedding.

GitHub Repository

Explore the full implementation and extend this project by visiting the GitHub repository. The repository includes code and instructions for running the signal classification system, including both machine learning and deep learning models.

For more detailed information, refer to the paper: Stress Estimation in VRET Using PPG Signals.

Future Impact

By enabling real-time stress detection using non-invasive PPG sensors, this platform offers new possibilities for adaptive therapy systems. These systems can automatically adjust the VR environment based on the patient’s mental state, providing a more tailored and effective therapeutic experience.