Scan Pattern Deviations in Toddlers with ASD: A Framework Based on Cohesion

Background: Standard region-of-interest (ROI) analysis assumes a specific spatial context in which interpretation is made.  Yet, these assumptions are often challenged by atypical attentional selections made by individuals with ASD. Autism researchers who employ eye-tracking in their work are also increasingly relying on the delivery of complex, naturalistic, dynamic scenes; ROI analysis of such stimuli can be ambiguous and time-consuming.

Objectives: To employ a region-free analysis method that leverages the spatiotemporal cohesion of typically developing (TD) toddlers in order to examine the atypical visual scanning of complex, social scenes by toddlers with ASD.

Methods: Two analyses were conducted on previous datasets from our lab (Chawarska et al., 2011; Campbell et al., 2011). In the first analysis participants were 20 month-old-toddlers with ASD (N=57), DD (N=31), and TD (N=74). In the second analysis, the ASD group was broken down via hierarchical clustering into three groups based on their response to an eye-tracking dyadic social probe: a group with typical scanning trajectories (ATYP, N=23), a group that did not monitor the speaker (ASPK, N=16), and a group that did not monitor the scene (ASCE, N=18). These groups were compared against TD controls.  A dynamic 3-minute scene consisting of multiple probes was shown to the toddlers, and for analysis was broken down into successive 500 ms segments. A third of the segments where TD individuals exhibited the greatest cohesion were isolated by taking successive medians on the distances between TD individuals. Median distances from TD participants in these segments were computed for each participant.  Scan pattern velocities were also calculated, in order to provide further region-free measures. Analyses were conducted both on the collection of isolated segments as well as the segments broken down into episodes including Sandwich-making and DyadicCues.

Results: Toddlers with ASD scanning patterns deviated from both DD and ASD groups overall (p<.05,p<.001), but most prominently in the DyadicCue condition (p<.001,p<.001). Atypical scanning patterns were, overall, associated with greater difficulties with eye-contact and social affect; in the DyadicCue condition it was also associated with lower receptive language. Velocities in the DyadicCue condition were also higher for the ASD group than other groups. Subgroup analysis indicated that ASCE showed the most atypical scanning patterns in all conditions (p<.001); the ASPK group showed atypical attention only in the DyadicCue condition (p<.001). Interestingly, the ATYP group only viewed the Sandwich condition atypically (p<.05), though prior ROI analysis had shown them to be equivalent in this condition. In this group, greater scene exploration (higher velocities) was associated with better receptive language.

Conclusions: This study presents uses region-free methods of eye-tracking analysis that also arrives at results similar to standard ROI analysis. Discrepancies between findings of atypical scan patterns in ATYP and findings of typical region scanning suggest more subtle spatiotemporal synchronization issues in the ATYP group given equivalence in overall region looking times. Increased velocities combined with greater atypicalities in scanpaths in ASD in the DyadicCue condition suggests depressed synchrony with socially salient cues not evidence in individuals without ASD.