A Preliminary Study of Using Computer Vision to Quantify Trunk Recruitment during Bimanual Play and the Effect of Upper Extremity Interventions in Unilateral Spastic Cerebral Palsy

Abstract Background Children with unilateral spastic cerebral palsy (USCP) often rely on trunk compensation due to impaired upper limb control, but current clinical tools do not directly capture trunk involvement. Marker-based systems are challenging to use with children, while computer vision methods like OpenPose offer a promising, scalable alternative for kinematic analysis but need to be validated. Purpose We validated OpenPose for quantifying trunk recruitment during bimanual play in children with USCP and examined how the interventions Constraint-Induced Movement Therapy (CIMT) and Hand-Arm Bimanual Intensive Therapy (HABIT) influence trunk use. Methods We analyzed videos of children with USCP who underwent CIMT or HABIT. OpenPose was used to extract trunk displacement angle (TDA) and trunk rotation angle (TRA), which were compared to hand function scores. OpenPose was validated against a 3D motion analysis system in typically developing adults. Reach-phase kinematic variables were also assessed. Results OpenPose showed high validity for TDA and lower validity for multi-planar TRA. TDA and TRA did not correlate with baseline hand function. HABIT reduced TDA, while CIMT slightly increased it. No significant changes were found in velocity, movement time, or variability. Conclusions OpenPose is a viable tool for capturing gross trunk motion. Trunk recruitment patterns differed by intervention, supporting the need for personalized approaches.