JMIR Publications

ABSTRACT

Background:

Cerebral Palsy (CP) is the most frequent motor disability in childhood, with a higher prevalence in low- and middle-income countries where access to essential early rehabilitation is limited. Generative Artificial Intelligence (GenAI) emerges as a disruptive technology with potential to address these challenges. This scoping reviews maps the current landscape of GenAI applications in CP rehabilitation.

Objective:

To systematically review and synthesize literature on the use of GenAI in CP rehabilitation, analyzing its applications, reported benefits, technical/ethical challenges, and future research directions.

Methods:

A systematic search was conducted following PRISMA 2020 guidelines across five databases (PubMed/MEDLINE, Scopus, Web of Science, IEEE Xplore, Google Scholar) through October 2025. Studies utilizing generative models (LLMs, GANs, VAEs, diffusion models) for diagnosis, assessment, therapy planning, documentation, or education in CP were included. Screening and data extraction were performed independently by two reviewers.

Results:

From 487 initial records, 32 studies (2022-2025) were included, indicating a nascent field dominated by research in high-income countries. Large Language Models (LLMs) constituted 75% of applications. Four key application categories were identified: 1. Diagnosis/Assessment: LLMs enabled early CP detection from clinical notes (Sensitivity:82%); GANs synthesized movement data to improve GMFCS classification accuracy from 72% to 90%. 2. Therapy Planning: LLMs generated personalized exercise regimens (quality 7.8/10 vs. expert 8.9/10); AI-designed VR content increased therapy adherence by >40%. 3. Clinical Documentation: Automation reduced note-writing time by 55%; AI decision support showed 80% concordance with clinical guidelines. 4. Patient/Caregiver Education: Tailored educational materials significantly improved family knowledge scores. Reported benefits included enhanced personalization, efficiency, and accessibility. Critical challenges included hallucinations/factual errors, data privacy concerns, algorithmic bias, a lack of interpretability, and risks of dehumanization.

Conclusions:

GenAI presents significant potential to augment CP rehabilitation by scaling personalization and improving efficiency. However, current evidence is primarily proof-of-concept. Responsible implementation necessitates: (1) robust clinical trials focusing on functional outcomes, (2) development of domain-specific models, (3) ethical frameworks addressing bias and accountability, (4) strategies for equitable global access, and (5) professional training for AI-augmented practice. GenAI should amplify, not replace, the therapist's expertise and the human therapeutic connection. Our collective choices will determine its ultimate impact on care.