Modeling gaze behavior with continuous-time markov chains to investigate social attention dynamics in autism

This study introduces a novel methodological framework combining continuous-time Markov chains and principal component analysis (PCA) to model and investigate gaze behavior in young children observing naturalistic social interactions. By quantifying transition propensities between areas of interest (AOIs), this approach enables a dynamic, data-driven analysis of gaze patterns beyond static fixation metrics. We applied this framework to eye-tracking data from children with autism spectrum condition (ASC) and neurotypical (NT) peers as they watched scenes of a child and an adult engaged in interactive play, involving turn-taking and reciprocal imitation. The stimuli, designed to ensure ecological validity, depicted sensory social routines (SSRs) with songs and shared play with musical instruments, allowing exploration of gaze dynamics in both dyadic and triadic social contexts. Results revealed distinct gaze transition profiles in ASC children, characterized by more frequent disengagement from socially salient AOIs and reduced re-orientation to faces following non-social fixations. In contrast, NT children exhibited frequent gaze alternation between faces and triangulation with objects, supporting joint attention and reciprocal engagement. Additionally, ASC participants were more likely to enter and persist in non-social states, especially during object-centered trials, whereas NT peers showed consistent transitions toward socially meaningful targets. These findings highlight the relevance of capturing the temporal patterns of visual engagement in autism, revealing how moment-to-moment gaze transition dynamics reflect underlying differences in social motivation, attentional control, and sensory processing. The proposed framework provides a powerful tool for characterizing individual differences in gaze organization and holds promise for advancing biomarker identification in neurodevelopmental research.