Conditional Probabilities of Dynamic Visual Scanning in School-Age Children with ASD

Background: When viewing scenes of naturalistic social interaction, individuals with Autism Spectrum Disorders (ASD) allocate visual attention differently than typically-developing (TD) children. However, studies have not examined the temporal sequence of fixations: how the probability of a current fixation location depends on prior fixations. Fixating on a given location at one moment in time may result in learning that impacts where one looks in the future.  Likewise, a failure to fixate on that location may result in different future fixations.  Examining the temporal sequence of fixations may reveal what is learned or what is missed by children with ASD when viewing scenes of social interaction.

Objectives: This study aims to: (1) identify, within movie scenes of social interaction, onscreen fixation locations that strongly predict future fixation locations in TD children; (2) identify fixation locations that strongly predict future fixations in children with ASD; and (3) measure whether fixation locations predict the same or different future fixations in children with ASD compared to TD children.

Methods: School-age children with ASD (n=26) and TD children (n=23) were matched on chronological age and verbal IQ. Children watched videos of age-appropriate social interaction while eye-tracking data were collected. Our analysis focused on fixation locations perceived as highly salient to viewers (as indexed by timepoints when the majority of viewers looked at the same location). These fixation locations were identified separately for TD and ASD groups (referred to as ‘TD targets’ and ‘ASD targets’). To determine whether the probability of fixating on a current target depends on fixation on a prior target we calculated probability ratios: the proportion of viewers who fixated on target X and target X-1 relative to those who fixated on target X but not X-1). Probability ratios were calculated for all pairs of targets identified by TD viewing patterns (aims 1,3) and for all pairs of targets identified by ASD viewing patterns (aim 2). 

Results: Aim 1: Preliminary results show that in TD viewers, 59.5% of TD targets showed higher conditional dependency than expected by chance (p<.05). This indicates that 59.5% of the targets fixated by TD children were more likely to be fixated if those children had looked at specific locations earlier in the movie, but were less likely to be fixated if those children had not looked at those earlier locations. Aim 2: By contrast, in ASD viewers, 38% of ASD targets showed higher conditional dependencies than expected by chance (p<.05). Aim 3: In ASD, the probability of fixating on a TD target did not depend on viewing prior TD targets, suggesting that both groups arrive at the same target location via a different pattern of prior fixations. 

Conclusions: In navigating the social world, current insights depend on past experience. Failure to learn about information conveyed in certain locations may result in later visual fixations that are more divergent from normative viewing patterns. Studies examining the temporal sequence of fixations offer important insights into visual scanning strategies in children with ASD, and the way in which those strategies shape subsequent learning.