Heightened Attention to Nonsocial Images Is Not Driven By Feature Salience in Children with Autism

Background: Circumscribed interests are one form of the diagnostically relevant restricted, repetitive behavior observed in autism spectrum disorders (ASD), and they are characterized by an intense, narrow interest in a given subject area or object type. It is not known how circumscribed interests in ASD may be influencing or influenced by the social deficits that are also characteristic of ASD. Eye-tracking studies have assessed attention to social and non-social stimuli in ASD and have shown that individuals with ASD show a preference for looking at non-social images (e.g., objects) over social images (e.g., faces); however, there is little understanding of what is driving attention to non-social stimuli relative to social stimuli in individuals with ASD.

Objectives:  Our study aimed to determine if low-level features of the images could be driving the preferential looking to non-social stimuli (in particular, stimuli that are of high interest to people with ASD) over social stimuli in ASD.  

Methods: We used gaze data from a paired-preference viewing task, whereby; typically developing children and children with ASD were presented with images of faces paired with either an image of an object of high interest to individuals with ASD (HAI) or an object of low interest to individuals with ASD (LAI).  Gaze data (proportion of look time, average fixation duration, and first fixation duration) were analyzed in relation to the low-level visual features of the image (color, intensity, and orientation).

Results: Our results indicated that all low-level features of the images were predictive of look time to the HAI images for the typically developing children (Color: R = -.124, p = .019; Intensity: R = .234, p < .001; Orientation: R = .230, p <.001; Composite: R = .213, p < .001), but only the composite score for the children with ASD (R = .124, p = .020). There was very little relationship between the low-level features and look time for the LAI images and faces in both the typically developing children and the children with ASD. For typically developing individuals, low-level features were predictive of average fixation duration (Color: R = .180, p = .002; Intensity: R = .180, p = .002; Orientation: R = .130, p = .017; Composite: R = .233, p < .001) and first fixation duration (Color: R = .213, p < .001; Intensity: R = .216, p = .021; Orientation: R = .117, p = .028; Composite: R = .190, p = .001) only for faces that were paired with LAI images, but low-level features were not predictive of average fixation duration or first fixation duration in individuals with ASD.

Conclusions: Given that low-level visual features were minimally predictive of gaze patterns in individuals with ASD regardless of image type, our results suggest that preferential looking to non-social stimuli in ASD is likely due to more cognitively directed attention rather than stimulus-driven attention.