Abstract:

One of the core symptoms of Autism Spectrum Disorder (ASD) is social communication impairment, with atypical facial emotion expressions (FEEs) being a prominent feature. FEEs have the potential to serve as a biomarker for early ASD screening. Following the PRISMA guidelines, this study systematically reviewed the literature up to 2024, identifying six studies that investigated atypical FEEs in children with ASD aged 8 months to 6 years. The review aimed to characterize these atypical expressions and evaluate the application of computer vision technology for ASD identification.

The findings reveal that children with ASD exhibit three main atypical characteristics in FEEs: (1) Predominance of neutral expressions and reduced positive expressions. Children with ASD often display neutral or minimal emotional facial expressions during daily interactions and in response to emotional stimuli, reflecting challenges in emotional perception, social context comprehension, and emotional regulation. While the frequency of positive expressions increases with age, it remains significantly lower than that of neutral expressions. (2) Low frequency of social smiles. Social smiles, a hallmark of early social behavior, appear less frequently in children with ASD compared to typically developing (TD) peers. This difference is evident as early as infancy and persists throughout development. (3) Deficits in facial expression imitation. Compared to TD children, children with ASD show reduced intensity and frequency of imitation when observing others' facial expressions, particularly in recognizing and imitating complex emotional expressions. These deficits are closely linked to social cognition impairments and difficulties in emotional processing.

The growing demand for early ASD screening has driven advancements in computer vision and artificial intelligence technologies, providing new tools for the automatic recognition of FEEs. Compared to traditional methods, such as manual evaluations and electromyography (EMG), computer vision-based approaches offer significant advantages: (1) Non-invasive assessment. These techniques use cameras for data collection without disrupting the child, enabling the capture of natural facial expressions. (2) Multimodal data integration. By combining facial expression data with behavioral and physiological signals, these methods improve the accuracy and efficiency of emotion recognition. (3) Scalability. Computer vision automatic recognition technology overcome the efficiency limitations of traditional tools, supporting large-scale screening and facilitating early intervention.

Despite these advancements, challenges remain. Future research should prioritize the following: (1) Developing emotion-inducing paradigms that mimic naturalistic scenarios to enhance ecological validity; (2) Exploring the diverse features of FEEs in ASD across varying contexts and emotional valences to identify unique expression patterns; and (3) Improving the accuracy and sensitivity of computer vision automatic recognition to ensure their applicability across different age groups and cultural backgrounds. Addressing these challenges will provide robust support for early screening and intervention for children with ASD.

Key words: autism spectrum disorder, facial expression, computer automatic recognition