Markerless motion capture to assess upper extremity movements in individuals with dyskinetic cerebral palsy: an accuracy and validity study

AimTo evaluate clinical utility of markerless motion capture (MMC) during an reaching-sideways-task in individuals with dyskinetic cerebral palsy (DCP) by determining (1) accuracy of key points tracking in individuals with DCP and typically developing (TD) peers, (2) concurrent validity by correlating MMC towards 3D-motion analysis (3DMA) and (3) construct validity by assessing differences in MMC features between a DCP and TD group. MethodMMC key points were tracked from frontal videos and accuracy was assessed towards human labelling. Shoulder, elbow and wrist angles were calculated from MMC and 3DMA (as gold standard) and correlated. Additionally, execution time and variability features were calculated from key points. MMC features were compared between groups. ResultsFifty-one individuals (30 DCP;21 TD; age:5-24 years) participated. An accuracy of approximately 1.5 cm was reached for key point tracking. While significant correlations were found for wrist ({rho}=0.810;p<0.001) and elbow angles ({rho}=0.483;p<0.001), MMC shoulder angles were not correlated ({rho}=0.247;p=0.102) to 3DMA. Wrist and elbow angles, execution time and variability features all differed between groups (Effect sizes 0.35-0.81;p<0.05). InterpretationVideos of a reaching-sideways-task processed by MMC to assess upper extremity movements in DCP showed promising accuracy and validity. The method is especially valuable to assess movement variability within DCP without expensive equipment.