Automated recognition of spontaneous facial expression in individuals with autism spectrum disorder: parsing response variability

Background Reduction or differences in facial expression are a core diagnostic feature of autism spectrum disorder (ASD), yet evidence regarding the extent of this discrepancy is limited and inconsistent. Use of automated facial expression detection technology enables accurate and efficient tracking of facial expressions that has potential to identify individual response differences. Methods Children and adults with ASD (N = 124) and typically developing (TD, N = 41) were shown short clips of funny videos. Using automated facial analysis software, we investigated differences between ASD and TD groups and within the ASD group in evidence of facial action unit (AU) activation related to the expression of positive facial expression, in particular, a smile. Results Individuals with ASD on average showed less evidence of facial AUs (AU12, AU6) relating to positive facial expression, compared to the TD group (p < .05, r = - 0.17). Using Gaussian mixture model for clustering, we identified two distinct distributions within the ASD group, which were then compared to the TD group. One subgroup (n = 35), termed over-responsive, expressed more intense positive facial expressions in response to the videos than the TD group (p < .001, r = 0.31). The second subgroup (n = 89), (under-responsive), displayed fewer, less intense positive facial expressions in response to videos than the TD group (p < .001; r = - 0.36). The over-responsive subgroup differed from the under-responsive subgroup in age and caregiver-reported impulsivity (p < .05, r = 0.21). Reduced expression in the under-responsive, but not the over-responsive group, was related to caregiver-reported social withdrawal (p < .01, r = - 0.3). Limitations This exploratory study does not account for multiple comparisons, and future work will have to ascertain the strength and reproducibility of all results. Reduced displays of positive facial expressions do not mean individuals with ASD do not experience positive emotions. Conclusions Individuals with ASD differed from the TD group in their facial expressions of positive emotion in response to funny videos. Identification of subgroups based on response may help in parsing heterogeneity in ASD and enable targeting of treatment based on subtypes.