An augmented reality and robotic rehabilitation platform with prognostic modeling for post-stroke hand spasticity
Year: 2026
Type: article
Study design: Clinical Trial
Abstract: Post-stroke hand impairment can limit manual dexterity and daily activities. We evaluated Rehabotics, an integrated augmented reality (AR) and robotic finger-stretching platform, in a 5-week single-arm intervention. Eighty patients with hand spasticity (time since stroke, 4.53 ± 1.04 months; baseline Ashworth Scale (AS), 2.28 ± 0.62) completed three supervised 30-minute sessions per week, consisting of AR serious-game training followed by robotic finger stretching. Clinical outcomes were assessed at baseline and post-intervention using the AS, Box and Block Test (BBT), and Active Range of Motion (AROM), expressed as Total Active Motion (TAM). We developed a novel exploratory prediction framework using linear mixed-effects models to derive task-specific features from first-session AR task performance, relative to patterns learned from repeated training-cohort data. These AR-derived digital markers were combined with baseline clinical variables in ridge regression to predict post-intervention outcomes. Internal validation used patient-level 5-fold grouped cross-validation (CV), with leave-one-subject-out (LOSO) CV performed as a sensitivity analysis. Following the intervention, AS scores decreased from 2.28 ± 0.62 to 1.34 ± 0.48 ( p = 3.22·10 − 15 ), BBT scores increased from 39.14 ± 2.45 to 45.14 ± 1.57 blocks ( p = 3.47·10 − 50 ), and AROM increased from 147.98 ± 19.27° to 224.61 ± 7.58° ( p = 3.37·10 − 62 ). Our findings reflect within-group pre–post changes in a single-arm cohort and do not establish superiority over conventional physiotherapy. The prediction model achieved R² = 0.615 ± 0.397 for BBT and 0.661 ± 0.271 for AROM, while AS classification achieved 0.850 ± 0.034 accuracy and quadratic weighted kappa (QWK) of 0.608 ± 0.187. LOSO CV yielded pooled R² values of 0.636 for BBT and 0.750 for AROM, with AS accuracy of 0.900 and QWK of 0.768, suggesting that early AR-derived markers may provide short-term prognostic information.
Source: Scientific Reports
Language: English
Subfield: Rehabilitation
Field: Medicine
Domain: Health Sciences
Keywords: stroke rehabilitation, augmented reality, serious games, prognostic model, linear mixed models, ridge regression, Ashworth scale, Box and Block Test, active range of motion, digital markers, hand function recovery, manual dexterity, physiotherapy comparison, leave-one-subject-out cross-validation
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7236688940
