Virtual Reality Exposure Therapy (VRET) provides a controlled and immersive approach for the treatment of specific phobias, including fear of heights. However, conventional exposure protocols generally rely on predefined stimulus hierarchies and do not account for the patient’s current emotional state. Passive Brain-Computer Interfaces (pBCIs) offer the possibility of overcoming this limitation by estimating affective states from physiological signals and dynamically adapting the virtual environment. This thesis investigates the feasibility of a neuroadaptive Virtual Reality Environment (VRE) for height exposure based on the real-time estimation of fear from multimodal physiological signals. A complete closed loop system was developed integrating a virtual urban environment, electroen- cephalography (EEG), electrocardiography (ECG), real-time signal processing, and machine learning. Nine participants were exposed to four virtual height levels. EEG and ECG signals were synchronously acquired and processed to extract spectral and asymmetry-based EEG features, together with heart rate and heart rate variability measures. Several machine learning models were evaluated during offline validation, and the best performing approach was subsequently used to train personalised classifiers for real-time neuroadaptive experiments. In the online phase, the estimated fear state was used to dynamically increase or decrease the virtual height. An additional post-hoc cross-subject analysis was performed to investigate the generalisability of the extracted physiological features across individuals. Offline classification demonstrated that the acquired physiological signals contained discriminative in- formation, although performance varied substantially between individuals. Frontal alpha and beta asym- metry features were among the most consistently selected EEG measures, while heart rate-related features provided complementary information. The online experiments demonstrated the technical feasibility of the closed loop system, with classification accuracies ranging from 40% to 90% across participants and an average accuracy of 68.0%. The post-hoc cross-subject analysis achieved a best balanced accuracy of 65.6%, suggesting that part of the fear-related physiological information can be generalised across individuals, although with reduced performance. Overall, the findings provide an initial proof of concept for the integration of multimodal physiological monitoring and adaptive Virtual Reality exposure. The developed framework demonstrates the technical feasibility of real-time fear estimation and environment adaptation, while also highlighting the strong inter-subject variability that remains a major challenge. Further studies involving larger and clinically relevant populations, improved physiological feature characterisation, and more robust machine learning approaches are required to establish a reliable and generalisable neuroadaptive system for exposure therapy.

Toward a Neuroadaptive Virtual Environment for Acrophobia Exposure Therapy

SIMONCELLI, MICHELE
2025/2026

Abstract

Virtual Reality Exposure Therapy (VRET) provides a controlled and immersive approach for the treatment of specific phobias, including fear of heights. However, conventional exposure protocols generally rely on predefined stimulus hierarchies and do not account for the patient’s current emotional state. Passive Brain-Computer Interfaces (pBCIs) offer the possibility of overcoming this limitation by estimating affective states from physiological signals and dynamically adapting the virtual environment. This thesis investigates the feasibility of a neuroadaptive Virtual Reality Environment (VRE) for height exposure based on the real-time estimation of fear from multimodal physiological signals. A complete closed loop system was developed integrating a virtual urban environment, electroen- cephalography (EEG), electrocardiography (ECG), real-time signal processing, and machine learning. Nine participants were exposed to four virtual height levels. EEG and ECG signals were synchronously acquired and processed to extract spectral and asymmetry-based EEG features, together with heart rate and heart rate variability measures. Several machine learning models were evaluated during offline validation, and the best performing approach was subsequently used to train personalised classifiers for real-time neuroadaptive experiments. In the online phase, the estimated fear state was used to dynamically increase or decrease the virtual height. An additional post-hoc cross-subject analysis was performed to investigate the generalisability of the extracted physiological features across individuals. Offline classification demonstrated that the acquired physiological signals contained discriminative in- formation, although performance varied substantially between individuals. Frontal alpha and beta asym- metry features were among the most consistently selected EEG measures, while heart rate-related features provided complementary information. The online experiments demonstrated the technical feasibility of the closed loop system, with classification accuracies ranging from 40% to 90% across participants and an average accuracy of 68.0%. The post-hoc cross-subject analysis achieved a best balanced accuracy of 65.6%, suggesting that part of the fear-related physiological information can be generalised across individuals, although with reduced performance. Overall, the findings provide an initial proof of concept for the integration of multimodal physiological monitoring and adaptive Virtual Reality exposure. The developed framework demonstrates the technical feasibility of real-time fear estimation and environment adaptation, while also highlighting the strong inter-subject variability that remains a major challenge. Further studies involving larger and clinically relevant populations, improved physiological feature characterisation, and more robust machine learning approaches are required to establish a reliable and generalisable neuroadaptive system for exposure therapy.
2025
Toward a Neuroadaptive Virtual Environment for Acrophobia Exposure Therapy
passive BCI
Exposure Therapy
Neuroadaptive System
Virtual Reality
Online EEG analysis
File in questo prodotto:
File Dimensione Formato  
Simoncelli_Michele.pdf

accesso aperto

Dimensione 21.52 MB
Formato Adobe PDF
21.52 MB Adobe PDF Visualizza/Apri

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/118576