Real-time anxiety classification in Virtual Reality Exposure Therapy
Loading...
Files
Full Text E-thesis
Date
2025
Authors
Mevlevioğlu, Deniz
Journal Title
Journal ISSN
Volume Title
Publisher
University College Cork
Published Version
Abstract
Virtual Reality Exposure Therapy (VRET) extends the reach of anxiety disorder treatment options by making them accessible and cost-efficient. Past research has supported the effectiveness of VRET in treating various disorders, such as generalised anxiety disorder and specific phobias. Based on the results of a scoping review, feedback options during therapy are currently inaccessible, and the limitations of biased self-report systems can hinder immersion into the virtual reality environment. Therefore, this thesis proposes a decision-support framework integrating continuous, seamless feedback on patient state into VRET. On-body physiological sensors are used to collect information on the patient's heart rate, skin conductivity and electrical brain activity. Ergonomic challenges arise while using such sensors with movement in Virtual Reality, and noise is introduced into signals from the electrical interference of the head-mounted display. Another challenge is establishing the ground truth for anxiety levels, which varies widely among past research. The proposed solution uses multi-modal signal acquisition combined with state of the art noise removal techniques to account for these limitations. The ground truth of anxiety is established by developing a Virtual Reality adaptation of the emotional Stroop task. The proposed emotional Virtual Reality Stroop Task (eVRST) is a cognitive task that requires users to log the colours of the words they see. High negative affect and high arousal words are chosen to evoke feelings of anxiety in users. The task has been split into three conditions to represent three distinct anxiety levels. A convolutional neural network is trained and deployed to classify the current anxiety state of the patient between no anxiety, mild anxiety and severe anxiety conditions with 75% accuracy. These calculations are carried out every five seconds and displayed to the therapist through a user interface that is not visible to the patient. Feedback from machine learning and the visualisation of raw signals help therapists make meaningful decisions and personalise the therapy to the patient's ever-changing needs. The feedback framework is transferrable, easily deployed, and scalable. In this thesis, recommendations for Virtual Reality bio-feedback systems are also highlighted to aid future research.
Description
Keywords
Virtual reality , Artificial Intelligence , Human-computer interaction , Biosensors , Machine learning , Virtual Reality Exposure Therapy
Citation
Mevlevioğlu, D. 2025. Real-time anxiety classification in Virtual Reality Exposure Therapy. PhD Thesis, University College Cork.