A Virtual Reality-Based Multimodal Approach to Diagnosing Panic Disorder and Agoraphobia Using Physiological...

Area of Science:

• Psychiatry and Psychology
• Biomedical Engineering
• Computational Neuroscience

Background:

• Virtual reality (VR) offers immersive environments for anxiety disorder assessment.
• Physiological signals (HRV, SCR) and self-reported anxiety are key indicators of distress.
• Machine learning can integrate multimodal data for diagnostic classification.

Purpose of the Study:

• To evaluate the efficacy of machine learning models in classifying panic disorder and agoraphobia using VR-based multimodal data.
• To assess the diagnostic potential of combining heart rate variability (HRV), skin conductance response (SCR), and self-reported anxiety during VR exposure.

Main Methods:

• Seventy-six participants (38 patients, 38 controls) underwent VR exposure in road and supermarket scenarios.
• Multimodal data collection included self-reported anxiety, HRV, and SCR.
• Six machine learning classifiers (GNB, k-NN, LRR, SVC, RF, SGB) were employed for data analysis.

Main Results:

• The optimal machine learning model achieved an accuracy of 0.83 for classification.
• Models demonstrated high specificity and precision (≥0.80), with varying sensitivity (≥0.82).
• Self-reported anxiety showed higher feature importance than physiological measures; patients exhibited blunted SCR responses.

Conclusions:

• VR exposure combined with self-reported anxiety and physiological data shows feasibility as a diagnostic aid for panic disorder and agoraphobia.
• The current approach requires further refinement to enhance sensitivity and clinical utility.
• Multimodal data integration via machine learning holds potential for objective assessment of anxiety-related disorders.