Embodiment is central to the therapeutic benefits of virtual reality (VR). Research shows that realistic avatars enhance embodiment, but scan-based models are costly and often inaccessible. Parameter-based avatars could be a practical alternative, yet their effectiveness relative to scans or abstract representations remains unclear. Cardio-visual illusions, which synchronize visual cues with the heartbeat, may also strengthen embodiment, though their role across realism levels is underexplored. Tackling this research gap, we conducted a 2× 4 within-subjects study examining cardio-visual synchrony (synchronous vs. asynchronous) and avatar realism, assessing embodiment through self-report and skin temperature. Results showed higher embodiment ratings for parameter-based and scan-based avatars compared to abstract ones. Under synchrony, parameter-based avatars did not differ from scans, while under asynchrony, scans retained higher scores. Although no physiological effects were observed, self-reported measures suggest that parameter-based avatars can approximate scan-level embodiment in VR psychotherapy.
Cancer-related pain exhibits marked temporal fluctuations that are poorly captured by conventional static assessment methods. While automated facial video analysis enables continuous and objective pain monitoring, existing approaches lack clinically interpretable temporal metrics and phenotypic characterisation. We developed and validated a two-stage artificial intelligence framework for pain assessment in cancer patients. A fine-tuned YOLOv8 model (mAP = 0.893) performed frame-level pain detection at 30 fps, after which we used the PainNet library to extract facial Action Units and computed Prkachin–Solomon Pain Intensity (PSPI) scores. To enhance patient stratification, we introduced novel temporal metrics, including Total Pain Burden (area under the PSPI–time curve), pain volatility, coefficient of variation, and inter-episode recovery time. Pain episodes were further clustered into six phenotypic categories based on duration and intensity. Five patients with cancer-related pain (mean video duration: 333.2± 134.2 s; 50, 011 frames) were analysed. The model flagged 44.7± 31.1 4.66± 1.32 , with substantial within-patient variability, while Total Pain Burden per minute varied seven-fold ( 44.65-310.52 ) despite similar mean PSPI values. This framework enables continuous pain monitoring and reveals clinically relevant temporal phenotypes and burden metrics not captured by intensity-based assessments, supporting a more personalised approach to pain management in oncology.
Early-onset Alzheimer disease (EOAD), affecting individuals under the age of 65, represents a particularly critical and impactful form of dementia, as it interferes with working age and active life stages. In this context, effective and engaging cognitive rehabilitation strategies are essential to support residual cognitive functions and maintain autonomy. This paper presents an XR-based cognitive training framework designed to stimulate memory, attention, and executive functions through interactive serious games. The proposed system leverages immersive and adaptive environments to simulate meaningful daily scenarios, provide immediate feedback, and personalize task difficulty according to the user’s abilities. It consists of three training games, targeting the development of object recognition, action sequencing, and narrative comprehension. The design emphasizes ecological validity, usability, and engagement, aiming to support patient-centered rehabilitation approaches. The proposed framework represents a promising step toward the development of clinically meaningful XR-based tools for cognitive rehabilitation in early-onset Alzheimer disease. The current work focuses on system design, with future studies planned for clinical validation.
This work presents the design, implementation and preliminary validation of an interactive AI-based system that anticipates social-crisis precursors in people with Autism Spectrum Disorder (ASD) by closing the loop between real-time Facial Emotion Recognition (FER), branching social narratives and a post-session Large Language Model (LLM) debriefing agent. A quantized MobileNetV2 model trained on the FER-2013 dataset was deployed on-device through TensorFlow Lite, the narrative was implemented in Ren’Py, and the LLM was served by a Llama-3.3-70B endpoint accessed through Supabase Edge Functions. The system was validated in an operational clinical setting at the “Caminito” Neuropsychological Center, with 15 patients with ASD aged 4–21 years, 5 therapists across 21 therapy sessions. Three dimensions were evaluated: technical stability, usability and preliminary clinical impact. The quantized FER model reached an overall accuracy of 60.85
Body image disturbance in Anorexia Nervosa (AN) is increasingly understood as a dysfunction of multisensory body representation rather than a purely cognitive distortion. From an engineering perspective, this reflects instability in the integration of visual, proprioceptive, and interoceptive signals underlying embodied self-perception. Virtual Reality (VR) offers a controllable multisensory environment to experimentally manipulate these processes and may support future therapeutic applications. We present the Enhanced Body Swap (EBS), a real-time immersive VR module developed within the ARCADIAVR project to manipulate body representation through controlled embodiment. The system employs a parametric virtual body representation based on a set of predefined avatar models with continuous BMI-driven geometric morphing and integrates real-time cardio-respiratory biofeedback to enhance interoceptive–exteroceptive congruence. Therefore, although the architecture is inspired by the Digital Twin paradigm, it should be considered a Digital Twin-oriented or BMI-adaptive embodied avatar rather than a fully individualized patient-specific anatomical Digital Twin. The Unity-based architecture includes head-mounted display tracking, wearable biosensors for ECG-derived heart rate and respiration, and a clinician control interface. Physiological signals are processed with low end-to-end latency to preserve embodiment stability. A structured embodiment modulation protocol implements progressive body morphing across sessions. A usability study was conducted in healthy volunteers (N = 15) to evaluate system performance and user experience. Results showed good usability (SUS = 76.4 ± 7.9), low-to-mild simulator sickness (SSQ = 13.6 ± 10.2), strong visuomotor synchrony (78.2 ± 9.6), and moderate embodiment induction (64.7 ± 13.5). The EBS module represents an innovative XR framework for multisensory embodiment modulation. Its potential clinical relevance for AN remains hypothetical and requires formal validation in clinical samples through controlled longitudinal studies.
