Mental health care sees rising interest in AI-integrated virtual reality

Mental health disorders remain among the world's largest health burdens, affecting quality of life, health systems and families. New evidence suggests that AI-integrated virtual reality (VR) could support anxiety detection, stress monitoring, PTSD assessment, depression screening and adaptive exposure therapy. However, a new review published in the Journal of Clinical Medicine finds that the field remains early, fragmented and not yet ready for broad clinical adoption.

The study, titled "Artificial Intelligence-Integrated Virtual Reality in Mental Health Care: A Scoping Review of Evidence, Clinical Applications, and Future Directions" reviewed English-language studies published between January 2020 and February 2026 to map how AI and VR are being combined in mental health care, what clinical uses have emerged, and what evidence is still missing.

AI and VR converge as mental health systems face treatment gaps

Global care systems continue to face persistent gaps in access and capacity, creating pressure for scalable tools that can support assessment, treatment and monitoring. VR is already being explored in mental health because it can create controlled and repeatable simulations of feared, stressful or clinically relevant situations. In anxiety care, for example, virtual environments can expose users to heights, public speaking or other triggers in a graded setting. AI adds another layer: it can analyze physiological, behavioral or symptom-response data, classify distress states, predict treatment response and help adapt virtual scenarios to individual needs.

The combined technology is potentially useful because many psychiatric symptoms are dynamic and context-dependent. Anxiety, trauma responses, stress and avoidance patterns often change in response to environment, arousal and perceived threat. Immersive VR can recreate those contexts, while AI can interpret biosignals or behavior recorded during exposure.

The review included 18 eligible studies. The evidence base was small and geographically uneven, with South Korea contributing the largest share of studies, followed by China and several single-study countries. Most studies used immersive or hybrid VR environments combined with machine learning methods, biosensors, physiological monitoring, EEG, heart rate, electrodermal activity or movement tracking.

The strongest concentration of research focused on anxiety-related conditions, trauma and post-traumatic stress symptoms, stress, panic disorder, agoraphobia, phobias and social anxiety disorder. Some studies examined adjacent areas, including subjective cognitive decline, psychological comfort in virtual hospital settings and neurocognitive rehabilitation. The review treats these as relevant to mental health care but not as direct proof of psychiatric treatment efficacy.

A recurring pattern emerged: many studies were designed to test feasibility, classification accuracy or predictive modeling rather than clinical effectiveness. In other words, the field is showing that AI can be connected to VR-based mental health workflows, but it has not yet shown that AI-integrated VR is better than standard VR or usual care.

Anxiety, PTSD and stress dominate early clinical applications

The review identifies four main uses for AI-integrated VR in mental health care: assessment and screening, prediction and monitoring, treatment augmentation through adaptive VR or biofeedback, and usability or implementation testing.

• Anxiety detection was one of the clearest early applications. Several studies used VR exposure settings and physiological signals to classify anxiety or stress. In acrophobia-related scenarios, researchers used heart rate, electrodermal activity and movement data to detect fear or anxiety responses. Other studies used EEG or biosensor data during VR stress and relaxation sessions to estimate stress levels or classify stress states.
• Social anxiety disorder also featured prominently. Some studies used VR public speaking or social situations to collect autonomic signals and predict specific anxiety symptoms or VR sickness risk. These findings suggest that AI may help clinicians understand which patients are likely to experience more severe symptoms, tolerate VR poorly or respond differently to exposure-based therapy.
• Panic disorder and agoraphobia were another target area. A VR-based multimodal approach used physiological measures and machine learning to distinguish patients from controls, while another study used simulated VR experiences to predict early treatment response in panic disorder. These applications point toward a future in which VR sessions could generate data useful for treatment planning and risk stratification.

The review also found emerging work on PTSD and post-traumatic stress symptoms. Some studies used physiological habituation responses inside immersive virtual environments to classify PTSD-related patterns, including in military personnel and veterans. Others used machine learning to predict the response to VR-based stabilization therapy among people with post-traumatic stress symptoms.

Depression screening appeared in newer research. One study used a VR-based multimodal framework with machine learning to screen adolescents for depression, combining behavioral and physiological signals. The review describes the results as preliminarily promising but not yet validated enough for routine screening.

Adaptive and biofeedback-supported VR formed another major theme. Some studies tested VR systems that used AI or machine learning to detect arousal and adjust feedback during exposure therapy. Others explored intelligent agents or reinforcement learning to adapt tasks for older adults with cognitive decline. A more recent study tested an AI-supported adaptive VR exposure program for spider fear and reported preliminary improvements across sessions.

The review also highlights the rise of AI-guided self-reflection tools. One formative study used a large language model-enabled virtual human in an immersive avatar-based self-talk environment for psychological counseling and self-reflection. Participants found the approach acceptable, but clinical outcomes were not yet established.

Across these studies, the technological results were often stronger than the clinical evidence. Some models reported high classification accuracy for stress, anxiety or movement-based fear responses, but the review cautions that small samples, single-site designs, limited external validation and varied outcome measures make it difficult to translate technical performance into clinical value.

Evidence gap delays clinical rollout

The study warns that AI-integrated VR should be treated as promising but unproven. Available studies are heterogeneous, generally small and often focused on proof-of-concept performance. Large randomized trials, pragmatic multicenter studies and direct comparisons with standard VR or usual care remain limited.

Standard VR therapy already has a role in anxiety and exposure-based care. The key question is whether AI improves outcomes by personalizing scenarios, detecting distress earlier, predicting response, reducing dropout, enhancing engagement or helping clinicians make better decisions. The review finds that this question has not yet been answered.

Safety reporting is another weakness. VR can trigger cybersickness, emotional overload, distress, dissociation or symptom worsening in vulnerable patients. AI systems can also introduce risks through bias, poor transparency, overfitting, weak validation or misleading outputs. When these technologies are combined, the risks become more complex because the system may process highly sensitive data, including physiology, movement, behavior, voice, symptoms and emotional response.

The review point out that clinician oversight remains crucial. AI-integrated VR should not be treated as an autonomous mental health intervention. Its more realistic role, based on current evidence, is as a support tool for assessment, monitoring, personalization, biofeedback or treatment planning under human supervision.

Implementation barriers are also significant. Even when studies report acceptability, real-world rollout depends on hardware cost, sensor reliability, staff training, data governance, maintenance, workflow fit, patient accessibility and cultural relevance of virtual environments. Many studies used off-the-shelf VR systems and sensors, but routine care settings may struggle with calibration, signal quality, privacy rules and integration into clinical records.

The study identifies several research priorities. Future studies need larger and more representative samples, standardized VR and AI reporting, clearer safety monitoring, long-term follow-up, stronger equity analysis, and direct measurement of whether AI adds clinical value beyond VR alone. Research should also test whether these tools work across cultures, age groups, diagnoses and service settings.

The review calls for better reporting standards aligned with emerging AI and clinical trial guidance, including clearer descriptions of AI components, human-AI interaction, input data quality, validation methods and risk of bias. It also urges future work to track clinical outcomes alongside usability, adoption, fidelity, sustainability and cost.