Multimodal behavioral analysis of child play interactions for early detection of Autism Spectrum Disorder

Autism Spectrum Disorder (ASD) is characterized by early challenges in social communication and behavioral flexibility, making timely behavioral analysis critical for early detection. We present a comprehensive framework for analyzing ASD-related child behaviors using a novel Play Behavior Protocol tailored to the Indian sociocultural context. Video recordings of semi-structured play interactions between the child, caregiver, and a clinically trained examiner were collected from children aged 0-36 months, including 34 children with ASD, 10 with developmental delay (DD), and 23 typically developing (TD) children. A comprehensive set of twenty-six behaviors, including facial, social, and play-based behaviors, were analyzed for frequency, duration, and variability across age and sex. Statistical analyses indicated that children with ASD exhibited reduced eye contact, social smiling, and joint attention, along with increased no affect, repetitive play, and solitary play compared to TD children. Additionally, children with ASD demonstrated lower levels of eye contact compared to children with DD. Furthermore, participant-level ratios, such as the eye-contact-to-gaze ratio and the social-smile-to-positive-affect ratio, emerged as potentially important markers for distinguishing ASD from DD and TD, with both ratios being lower in the ASD group compared to the DD and TD groups. Automated multimodal analysis of facial affect and gaze cues revealed that children with ASD exhibited reduced human-directed gaze, lower levels of happy and surprise expressions, and increased sad expressions compared to children with DD and TD. By integrating culturally contextualized data collection, systematic statistical analysis, and automated multimodal inference, this work presents the first comprehensive behavioral analysis framework of its kind for early ASD assessment in Indian children.