The Discriminant Power of the Combination of Oculometric and Pupillometric Parameters during the Exploration of Objects and Faces between Autism Spectrum Disorders and Typical Development

Background: Autism Spectrum Disorders (ASD) are characterized by impairments in social communication and atypical social interactions. These impairments could be related to an alteration of face and social stimuli processing. Many studies have used eye-tracking in order to identify key parameters specific to ASD visual exploration, particularly of faces and social situations. However, if several parameters have been identified (e.g. time spent on the eyes region during face observation, preference for non-biological over biological movement), none of them allow to distinguish between ASD and Typical Development (TD) subjects with more than 70-80% discrimination power.

Objectives: The aim of this study was to explore how the combination of several eye-tracking parameters could improve the discrimination between ASD and TD children compared to each parameter individually.

Methods: Eye position and pupil diameter were recorded with an eye-tracking system (FaceLAB®) in 87 ASD and 96 typical children (3-12 years old) during a passive observation paradigm in which images of objects and faces were presented. These stimuli were interleaved with grey images containing a black cross (baseline). At the beginning of the paradigm, black and white images were also presented in order to elicit a pupillary light reflex.

A total of 18 parameters were extracted from the different conditions (baseline, objects, faces, black/white images) and the two measurements: pupil (diameter, speed) and gaze (total tracking time, time spent on the screen or the stimulus, fixation duration, number of fixations, latency to the first fixation).

A Receiver Operating Curve (ROC) analysis was conducted in order to determine the discrimination power of individual parameters, estimated by the Area Under a Receiver Operating Curve (AUC). We then used a data-mining approach combining all oculometric and pupillometric parameters using a logistic regression model. A ROC analysis was conducted to estimate the discrimination power of the model.

Results: The analysis required to determine the 18 parameters for all the subjects without any missing data. In total, 18 ASD children and 33 TD children were included in the analysis.

Some individual parameters reached a discrimination power between the ASD and TD groups of about 80%, e.g. the time spent on faces for children older than 8 years-old.

The combination of the 18 parameters reached a >95% discrimination power for children under 8 years-old (with a sensitivity and a specificity of >95%). The results stayed really good when the 9 and 10 years-old were also included (AUC 93%, sensitivity 85%, specificity 88%), and lowered to 82% when the 11 and 12 years-old were added.

Conclusions: These results suggest that eye-tracking parameters are very effective to discriminate ASD and TD children when combined, particularly for younger children. Analyses still have to be pursued in order to determine which parameters are crucial to describe the evolution of the exploration strategies with age, both in ASD and TD children. These results will need to be replicated with a new and large group of patients, but are very promising to help the clinical diagnosis and follow the evolution of individual patients.