Sensor-integrated AI-assisted gait training in dementia and Parkinson's disease: A single-session pre–post study

To investigate whether a single session of AI-assisted gait training improves spatiotemporal gait parameters in individuals with dementia and PD.MethodsThis single-session pre-post study enrolled ambulatory adults aged 60-85 years with dementia or PD. Participants completed one gait-training session using the Best Shape Gait (BSG) platform, with a planned duration of 15-30 min and the actual duration adjusted according to individual physical tolerance. Primary outcome was gait-speed change. Secondary outcomes included step length, step count, and gait variability. Paired t tests were performed within four predefined subgroups: Dementia-Support (n = 67), Dementia-Escort (n = 43), PD-Support (n = 64), and PD-Escort (n = 43).ResultsGait speed significantly increased after training across all predefined subgroups. In Support-mode walking, both dementia and PD cohorts demonstrated reduced step count and increased step length, suggesting improved locomotor efficiency. In Escort-mode walking, gait speed increased and step count decreased significantly in both cohorts, whereas changes in step length and gait variability were not significant. Gait variability did not significantly change in either cohort. Gait-speed improvement was consistent across dementia subtypes, modified Hoehn and Yahr stages, age groups, and fall-history categories. Baseline MMSE score and levodopa equivalent daily dose were not significantly associated with gait-speed improvement.ConclusionsA single session of sensor-integrated, AI-assisted gait training was associated with immediate within-session improvements in gait speed in individuals with dementia and PD. These findings support the feasibility of standardized, minimally supervised gait training in geriatric neurorehabilitation settings.