Modifying Social Attention with Gaze-Contingent Eye Tracking for Toddlers and Children with ASD

Background: Toddlers with ASD, compared to typically developing (TD) and developmentally delayed (DD) peers, show diminished responses to dyadic engagement (Chawarska et al, 2012) and limited attention to the activities of others when viewing a naturalistic play scene (Shic et al, 2011). This decreased attention likely reflects not only the culmination of atypical experience-dependent knowledge regarding scenes and people, but also suggests that future access to observational learning may be limited.

Objectives: 1) To explore the feasibility of automated gaze-modification strategies for toddlers with ASD, 2) To assess short-term malleability of visual attentional patterns. 3) To identify whether measures of attentional redirection can be used in the direction of intervention.

Methods: In (1) Normative Data Collection phone, TD controls (n=41, age = 37.4±16.6 months) free-viewed scenes (of an actress emulating interactions with the viewer) while their gaze was eye tracked. This data was used as a normative reference to create a gaze-contingent (GC) adaptive cue system which redirected gaze upon deviation from TD looking patterns. Redirection was accomplished by highlighting TD controls looking locations. In the 2) Experimental phase, toddlers with ASD were randomly assigned to either the adaptive cue condition (cue, n=16, age = 32.3±8.7 months) or a control condition (no-cue, n = 19, age = 35.4±9.3 months).

Proportion of time looking at the screen (%valid) and proportion of time spent looking at the actress’s face (%face) were analyzed in a group (3) x block (3) linear mixed model, covarying for Mullen nonverbal DQ. Pearson correlations were applied between change in %face after GC training and the Mullen verbal and nonverbal DQ.

Results: There was a main effect of group for %face (F(2,47)=6.42, p=.003) and a significant group*block interaction (F(4, 163)=3.06, p=0.018). At pre-training there was no difference between groups (ps>0.1). During training, TDs had higher %face than the ASD no-cue (p=0.001), and the ASD cue group was not different from the TDs (p=.062) or ASD no-cue group (p=.11). Post training, the TD group still had more %face than the ASD no-cue (p<0.001) and ASD cue (p = .009) groups, but the ASD cue group had higher %face than ASD no-cue (p=.015). Only the ASD no-cue group had a significant decrease of %face from baseline to training (p=0.028) and post-training (p=0.001), with no trend in the TD and ASD cue groups (ps>0.1). A negative correlation between the Mullen nonverbal DQ and the difference between pre- to post-training of %face was observed for the ASD cue group (r=-.64, p=.013), but not significant for the ASD no-cue (r=0.38, p=0.18) or TD (r=0.11, p=0.65) groups.

Conclusions: Compared with the ASD no-cue group, gaze-contingent training was effective at mitigating decreases of attention towards the face of the dyadic social partner in the ASD cue group. After adaptive training the attention towards the social partner was significantly higher in the ASD cue group than the no-cue group. The negative association with the Mullen nonverbal DQ implies that this approach may have particular relevance as an attentional training tool for more impaired toddlers with ASD.