Towards a computer-assisted assessment of imitation in children with autism spectrum disorder based on a fine-grained analysis

Imitation is considered a fundamental skill for learning in children with autism spectrum disorder (ASD). However, existing assessment methods usually provide a qualitative description of whether children imitate while ignoring individual variability in their imitation ability. Therefore, this study proposes a computer-assisted method for the refined assessment of imitation ability in children with ASD. First, 25 children with ASD and 25 age-matched typically developing (TD) children between 60 and 78 months old were recruited to imitate meaningful actions on objects and nonmeaningful gestures. In this process, several quantitative indicators were applied to quantify the children’s imitation ability based on a fine-grained analysis of their visual attention and motor execution patterns. Then, three classic machine-learning algorithms were employed to explore whether the indicators could efficiently classify children with imitation difficulties. The results indicated that the proposed indicators could provide a detailed description of individual differences in imitation impairment. Also, the high classification performance revealed that the proposed indicators could classify children with and without imitation difficulties. In addition, the findings contributed to research regarding the diagnostic evaluation of children with ASD.