Synergy of Artificial Intelligence and Extended Reality in Patient-focused Health and Well-Being Applications—A Systematic Review
Tim Schwirtlich, Cheolmin Matthew Lee, Molly Beestrum, David C. Mohr · 2026 · ACM Transactions on Computing for Healthcare · doi:10.1145/3793544
Summary
This systematic review, conducted by researchers at Northwestern University's Feinberg School of Medicine, maps the landscape of applications that combine artificial intelligence and extended reality (XR) for direct patient health and well-being outcomes. The review follows PRISMA guidelines and was pre-registered in PROSPERO. Seven databases were searched through July 2024, yielding 4,612 articles after deduplication. After two rounds of independent screening against a Patient Intervention Outcome (PIO) framework, 64 articles met the eligibility criteria. The scope was specifically limited to patient-facing applications where AI and XR interact as integrated components — excluding provider tools, diagnostic aids, and cases where data is collected in XR but analysed by AI separately. The heterogeneous corpus spans conference papers (42%), journal articles (44%), and book chapters (14%), with study types ranging from conceptual papers and pilot studies (57.8%) through to experimental studies and a small number of randomised controlled trials. Applications were categorised across eight healthcare domains. Rehabilitation and physical therapy attracted the most studies (n=23), closely followed by mental health and psychological well-being (n=22), with additional clusters in paediatric and developmental care (n=12), physical fitness (n=7), telehealth (n=4), pain management (n=4), elderly care (n=4), and sleep management (n=2). The AI landscape covered machine learning, deep and reinforcement learning, and large language model / NLP methods roughly equally; VR was the dominant XR modality, used in over three-quarters of studies. The review concludes by proposing a taxonomy of four computational AI-XR paradigms linking technology types to user outcomes: engagement, personalisation, trustworthy flexibility, and sustainable affordability.
Key findings
Four patient benefit categories emerged across healthcare domains: (A) improved clinical outcomes through adaptive, personalised environments (n=25); (B) enhanced accessibility via remote care options (n=25); (C) increased engagement through gamification and real-time feedback (n=24); and (D) affordability as a replacement for expensive in-person services (n=21). Benefit A was most prominent in rehabilitation and mental health; benefit B was equally distributed across domains. AI type analysis showed that Analytical and Causal AI (movement recognition, interaction classification, adaptive difficulty) dominated rehabilitation applications, while Generative AI (LLM-powered conversational avatars) drove mental health and elderly care use cases. A surge in publications beginning in 2022 correlates with the peak of machine/deep learning adoption and, from 2023 onwards, with the launch of ChatGPT driving a wave of conversational AI applications in mental health XR. Three categories of challenges were documented. Technical challenges appeared in 62.5% of studies and centred on early-stage development, limited hardware capability, and insufficient validation data. Adoption challenges appeared in 60.1% of studies and included equipment cost, motion sickness, connectivity requirements, and patient reluctance to share mental health concerns with AI. Societal/ethical challenges — covering privacy, bias, accountability, and over-reliance on technology — appeared in 21.9% of studies. Only one randomised controlled trial was identified, treating childhood social anxiety with VR; the majority of evidence remains conceptual or from small pilot studies.
Relevance
This review is directly relevant to accessibility practitioners working at the intersection of AI, XR, and disability. It documents both the promise and the barriers of AI-XR interventions for patient populations who struggle to access traditional in-person care — including people with physical disabilities, autism spectrum disorder, developmental conditions, chronic pain, and mental health disorders. The finding that accessibility (remote care flexibility) ranks as a top-tier patient benefit, cited across rehabilitation, mental health, and telehealth domains, underscores the potential of these technologies to reduce participation barriers for disabled people. Critically, the review also surfaces accessibility failures: the paper explicitly lists HMD discomfort, motion sickness, high device costs, internet dependency, and specific barriers for older adults and people with physical or cognitive disabilities as adoption challenges in 60% of studies. For practitioners, the absence of large-scale validation and the scarcity of RCTs means that current AI-XR health applications should be treated as promising pilots rather than proven interventions. The proposed AI-XR taxonomy — centred on engagement, personalisation, trustworthy flexibility, and affordability — offers a useful framework for evaluating whether a proposed XR accessibility tool genuinely serves users or merely displaces cost and complexity onto them.
Tags: extended reality · artificial intelligence · rehabilitation · mental health · accessibility · systematic review · machine learning · conversational AI · healthcare · telehealth · human-computer interaction