ARTIFICIAL INTELLIGENCE-BASED MOTION ANALYSIS COMBINED WITH CONVENTIONAL PHYSIOTHERAPY FOR IMPROVING UPPER LIMB FUNCTION FOLLOWING STROKE: A RANDOMIZED CONTROLLED TRIAL

Abstract

Background: Upper limb hemiparesis is a common, disabling consequence of stroke. While conventional physiotherapy is the standard of care, it is limited by therapist availability and the qualitative nature of observational assessment. Markerless computer-vision-based motion analysis using artificial intelligence (AI) provides objective, real-time kinematic feedback during therapy, but rigorous randomized evidence evaluating its efficacy remains sparse.

Objective: To evaluate whether an 8-week intervention of conventional physiotherapy combined with real-time AI-based motion analysis feedback yields superior improvements in upper limb motor impairment measured by the Fugl-Meyer Assessment for Upper Extremity (FMA-UE) compared to conventional physiotherapy alone in stroke survivors with upper limb hemiparesis.

Methods: A parallel-group, assessor-blinded randomized controlled trial was conducted at an outpatient neurorehabilitation center. Sixty adults with post-stroke upper limb hemiparesis were randomly allocated (1:1) to either conventional physiotherapy plus AI-based motion analysis feedback (experimental group, n=30) or conventional physiotherapy alone (control group, n = 30). Both groups received 24 sessions over 8 weeks (3 sessions/week). The primary outcome was the change in FMA-UE score from baseline to week 8. Secondary outcomes included the Action Research Arm Test (ARAT), Box and Block Test (BBT), Wolf Motor Function Test (WMFT), Modified Ashworth Scale (MAS), Stroke Impact Scale (SIS) hand function domain, and Motor Activity Log (MAL). Data were analyzed using linear mixed-effects models under intention-to-treat principles.

Results: The experimental group showed a mean Fugl Meyer Assessment for Upper Extremity improvement of approximately 15 points compared with approximately 19 points in the control group from baseline to week eight, corresponding to an illustrative between group mean difference of roughly 5 to 6 points with a 95% confidence interval excluding zero and an effect size in the moderate to large range. Illustrative secondary outcome trends favored the experimental group on the Action Research Arm Test, Box and Block Test, and Wolf Motor Function Test, with no meaningful illustrative between group difference on the Modified Ashworth Scale

Conclusion: Integrating real-time AI motion analysis feedback into conventional physiotherapy leads to superior gains in motor recovery, dexterity, and functional arm use compared to conventional physiotherapy alone, without altering muscle spasticity. Markerless AI tracking offers a scalable, objective adjunct for clinical neurorehabilitation.