Quantifying the Dynamic of Visual Exploration of Complex Social Scenes in Children with Autism Spectrum Disorder (ASD) without Any a Priori: An Eye-Tracking Study

Background: Numerous eye-tracking studies using predefined areas of interest (AOIs) have highlighted the atypical visual exploration pattern in individuals with ASD (e.g. tendency to look less at the eyes and faces compared to the group of typically developing (TD) individuals (for a review, see Klin et al. 2003)). The approach using AOIs is highly adapted for addressing specific research questions, like the one of relative dominance between different groups of AOIs (e.g. eyes vs. mouth). However, while it can inform us on what is more attracting for one group/person, it is less suitable for capturing the subtlety of the dynamic of visual exploration employed during viewing naturalistic social interactions, and how visual exploration of such scenes develops and changes with age.

Objectives: Our aim was to develop a method that will be able to define age-appropriate dynamic norms of visual exploration of complex social scenes, based on a group of TD children, for quantitative comparison with children with ASD.

Methods: A 3-minute “Trotro” cartoon was displayed on a Tobii eye-tracker device for 37 ASD males (aged 3.95±1.25) and 28 TD males (aged 3.10±1.30). Inspired by the concept of a heatmap, for each frame of the video we intended to create the “normative” gaze pattern distribution. This normative gaze distribution was obtained from TD subjects, employing kernel density distribution estimation on the raw gaze data of TD individual for each frame of the video (Botev et al., 2009). Once the “norm” was defined for each patient we calculated the “distance” of his/hers gaze coordinates form this “norm” in a frame-by-frame manner. For each patient we obtained one measure per frame of Proximity from the “norm”, and thus obtained measure was averaged for the duration of the video. Higher values indicate the visual exploration of the individual is being more similar to the one of TD subjects. The proximity measure was further correlated with direct measures of expressive, receptive language, cognition and imitation skills (PEP-3); indirect measures of communication and socialization skills (VABS-II) and severity of autistic symptoms (ADOS-2).

Results: We found positive correlation between proximity from the “norm” and direct measure of verbal & nonverbal cognition (R² = 0.19; p =. 008), language understanding (R² = 0.18; p =. 01), language expression (R² = 0.21; p =. 007) and imitation skills (R² = 0.15; p =. 02). We have also observed positive correlation between this measure and indirect measure of communication (R² = 0.24; p =. 004) and socialization skills (R² = 0.27; p =. 001), as parents reported them. Finally, a tendency toward negative correlation between the proximity from the “norm” and the severity of autistic symptoms was observed (p=. 0507).

Conclusions: Our results suggest that our method can be used as a valid measure for quantifying the dynamic visual exploration of complex social scenes based on a group of TD children, providing an age-appropriate manner to measure deviances in social cognition development in age-matched children with ASD. We further intend to use this measure in a longitudinal design.