Adults with Autism Display Increased Gaze to Low-Level Visual Features When Viewing Dynamic Social Videos

Background:  Abnormalities in social attention emerge early in individuals with autism spectrum disorders (ASD), and represent a core aspect of the phenotype across the lifespan.  The vast majority of experiments that have examined social attention in ASD have used highly constrained tasks and stimuli in order to attempt to isolate specific cognitive and attentional abnormalities.  These types of experiments, though essential and have provided important insight into attentional abnormalities, result in artificial scenarios that often fail to capture some of the key characteristics of social stimuli as they exist in the real world (e.g., dynamic, multi-modal, complex, subtle, continuous, etc.).  An alternative approach, and the one used here, is to use stimuli that better approximates that which would be encountered in the real world.

Objectives:  To determine the factors underlying gaze abnormalities in ASD when viewing dynamic social scenes.  

Methods:  We recorded eye movements from 20 adult ASD participants and 34 neurotypical control participants at 300Hz while they watched and listened to a full episode of the television show The Office (“Pilot”; © NBC Universal; 22 minutes in length, divided into 3 clips).  This stimulus was chosen for several reasons, which included long moments uninterrupted by camera cuts, multiple actors on the screen at once, often subtle emotional expressions, and socially awkward interactions (social faux pas).  Participants were instructed simply to watch the video and pay attention.  Initial analyses described below were carried out on data from the first clip.  We only included data from a participant if there was less than 2° of error across the entire screen both before and after the video was shown.  Similarity across individuals was quantified using the Normalized Scanpath Saliency method along with a leave-one-out-approach.  

Results:  We found that ASD and control groups differed overall in how they viewed the video.  The gaze pattern in controls was more similar to other controls, and less similar to those with autism (p <0.001), while the gaze pattern in individuals with ASD was less similar to controls and more similar to others with ASD, suggesting a common pattern of abnormal gaze in ASD.  Group differences in gaze persisted regardless of the number of people in the frame (0, 1, 2, or 3 people) (all p < 0.05), suggesting perhaps non-social factors might account for group differences.  Using a well-described computational model of visual saliency (Itti-Koch), we found that when gaze diverged between groups, individuals with ASD were more likely to fixate pixels with higher intensity (i.e., brightness).  This finding was replicated using data from the second clip. 

Conclusions:  These findings replicate prior results that individuals with ASD display abnormal attention to social stimuli, but we further demonstrate that abnormal patterns are common across individuals with ASD, with a bias toward looking at high-intensity areas of the scene.  Studies aimed at further elucidating the mechanisms underlying social (and non-social) attentional differences in ASD will be essential to understand how they might give rise to aspects of the autistic phenotype, and for developing novel targets of intervention.