Eye Tracking As a Spectrum of Biomarkers in Children with ASD

Background: Eye tracking has become a core methodology in autism research and is quickly moving towards relevance as a biomarker for ASD. Yet, the power of eye tracking lies not in its quantitative, objective, technical nature, but in the design of appropriate measures and paradigms that can tap into constructs with clinical and mechanistic relevance.

Objectives:

To critically examine, through a series of example studies, a spectrum of eye tracking data metrics, paradigms, and analysis approaches, with the ultimate goal of charting the trajectory of eye tracking as a potential biomarker for ASD.

Methods:

We aim to categorize eye tracking biomarker development efforts in ASD in terms of their (1) intended use; (2) selection of measures; and (3) theoretical approach. Using illustrative data from infants and toddlers on paradigms such as Chawarska et al., 2012 (including subsets of children with ASD (n=206), developmental delay (DD; n=51), children at high risk for autism without autism (HRNA; n=115) and typical development (TD; n=184)) we describe tasks and approaches designed to capture features of autism, describe mechanisms, and fulfill a practical role in early diagnostics or monitoring. We then turn towards a discussion of measures and contrast high-level region-based approaches to data-, computational-, and statistically-driven approaches.

Results:

We show that standard region-of-interest (ROI) based eye-tracking techniques replicate high-level deficits observed in ASD, such as diminished orienting towards social information (p<.001). We show that a single paradigm can be flexibly used to explore genetic mechanisms associated with autism (sex-linked differences between girls at high risk for ASD and other groups at 6 months, p<.06), to identify early markers of later developing autism (>80% predictive accuracy), and quantify features associated with clinical phenotypes (correlations with autism severity, p<.05). We discuss the utility of missing data and the distributional properties of fixations, saccades, and looking time, highlighting the importance of understanding the fundamental nature of eye tracking metrics and the errors that can result from lack of clarity. We show that data-driven methods, computationally-based approaches, and statistical techniques designed to quantify “atypicality” complement high-level approaches and provide additional precision towards our goal of defining “biomarkers” (p<.01). Stepping back and examining the field critically, we highlight the disconnect between eye tracking as a research methodology and as a clinical tool. Similarly, we reframe our progress as a field in terms of our goals for a viable biomarker for autism. We point towards cumulative evidence of the field of eye tracking in ASD research as a positive and the power of large data sets and new approaches to provide illumination.

Conclusions:

Eye tracking is a powerful tool that has provided us with a way of quantifying atypical behavior and social preferences in children with ASD. Viewed as a tool with multiple axes of flexibility and a spectrum of forms, it empowers us to make great progress towards identifying viable biomarkers for ASD symptoms and dimensions impacting quality of life. Yet, in order to do so, our research questions must be framed and appropriate methodology brought to bear.