Joint probabilistic modeling analysis of head and eye behaviors in autism spectrum disorder children based on a social interaction paradigm

Autism Spectrum Disorder (ASD), as a complex neurodevelopmental disorder, is closely associated with attention deficits that manifest through eye and head movements. Previous studies on the eye gaze and head posture of children with ASD have been somewhat limited by the contexts in which the children were observed, Exploring head and eye coordination in natural environments is crucial for developing effective intervention strategies applicable to everyday life. This paper aims to examine the perceptual and behavioral responses of children with ASD in simulated real-life social environments. Using a social interaction paradigm based on real-life scenarios, we propose a joint probabilistic modeling method for head and eye behaviors. This method includes the influence of head posture on gaze direction in eye-tracking studies within social interaction contexts. For each participant, we establish a data-driven Markov chain model based on individual data, preserving the temporal nature of eye movement behavior and the highly individualized nature of visual behavior. We conducted experiments on a video dataset of children with ASD that we collected, achieving a classification accuracy of 79.66%, demonstrating the feasibility and effectiveness of our proposed method. Additionally, we found that one manifestation of attention deficits in children with ASD is an increased occurrence of head-eye counter movement. This finding provides new reference indicators for the early diagnosis and screening of ASD.