Using computer vision to quantify mirror movements in children with unilateral cerebral palsy
MetadataVis full innførsel
Mirror movements (MM) are common in children with unilateral spastic cerebral palsy (USCP). In the clinic and in research, MM are most often assessed qualitatively using the scoring system proposed by Woods and Teuber (W&T), whereas quantification of MM is more rarely reported - probably since it has been dependent on special equipment. AIM: This study investigated, whether computer based video analysis could be used to quantify MM. METHOD: 37 children and adolescents with CP, and 22 typically developing (TD) children and adolescents were instructed to perform hand movements while their hands were video-recorded. The films were used to assess MM qualitatively using the score according to W&T. In addition, the movements of the passive hand were quantified using computer based video analysis. RESULTS: 25 (68%) of the participants with USCP had MMs in their non-affected, and 22 (59%) had MM in their affected hand when movements were performed at slow speed. At fast speed, 26 (70%) had MM in their non-affected, and 27 (73%) had MM in their affected hand as assessed with W&T. Among the TD participants, 6 (27%) had MM in their dominant hand, and 4 (18%) had MM in their nondominant had assessed with the score according to W&T when movements were performed at fast speed. One (5%) had MM in the non-dominant hand, and none had MM in the dominant hand when movements were perfomed at slow speed. Boxplots and correlation analyses suggested that quantity of motion (QoM) mean was the variable from the computer based analysis that was most suitable as a proxy for MM. In the total population including TD participants the correlation of QoM mean with W&T scores was moderate to good as indicated by correlation coeffcients between 0.59-0.74 (p < 0.01). Within the group of participants with USCP, correlation coeffcients ranged between 0.70 and 0.88 (p < 0.01). Particular high correlation coefficients (0.70 - 0.88) were observed among participants where the setup of the video-recordings was strictly adhered to. The correlations were lower in the subsample of participants with USCP, where deviations from this setup were observed. Height and width of motion (HoM and WoM) standard deviation had correlation coefficients 0.63 -0.76 (p < 0.01). For the above mentioned subgroup of participants, the correlation coefficients were 0.72-0.84 (p < 0.01). CONCLUSION: We found a moderate to good correlation between the computer based video analysis and the clinical assessment, suggesting that this method may be used to assess MM quantitatively. However, the method seems to be sensitive to deviations in setup of the video recordings.