32244
Visualization-Guided Analysis of Eye Movements in Children with Autism Spectrum Disorder: Results from the ABC-CT Interim Analysis

Poster Presentation
Friday, May 3, 2019: 10:00 AM-1:30 PM
Room: 710 (Palais des congres de Montreal)
A. Atyabi1, M. C. Aubertine2, Y. Luo3, A. Naples4, S. J. Webb5, M. Murias6, C. Sugar7, R. Bernier5, G. Dawson8, S. Jeste7, J. McPartland4, C. A. Nelson9 and F. Shic10,11, (1)Seattle Children’s Research institute University of Washington, Seattle, WA, (2)Seattle Children's Hospital and Research Institute, Seattle, WA, (3)Child Health, Behavior & Devel, Seattle Children's Research Institute, Seattle, WA, (4)Child Study Center, Yale University School of Medicine, New Haven, CT, (5)Psychiatry and Behavioral Sciences, University of Washington, Seattle, WA, (6)Duke Center for Autism and Brain Development, Department of Psychiatry and Behavioral Sciences, Duke University, Durham, NC, (7)University of California, Los Angeles, Los Angeles, CA, (8)Department of Psychiatry and Behavioral Sciences, Duke Center for Autism and Brain Development, Durham, NC, (9)Boston Children's Hospital, Boston, MA, (10)Center for Child Health, Behavior and Development, Seattle Children's Research Institute, Seattle, WA, (11)Pediatrics, University of Washington School of Medicine, Seattle, WA
Background: Biomarker discovery in neuropsychiatric disorders often employs automated, statistical, data-driven, high-throughput machinery. Less explored are the ways human insight can be rapidly, scalably, and effectively leveraged. Computer visualization techniques can transform raw data into intuitive representations accessible by those without data expertise. An appropriate match between data and visualization can provide a common ground for allowing clinical insights to be utilized in facilitating data explorations associated with biobehavioral biomarker discovery. One area where visualization technology has advanced rapidly, and where the role of visualizations in bridging clinical insights with data mining expertise has great potential, is in the use of eye tracking for studying gaze differences between individuals with autism spectrum disorder(ASD) and typically-developing(TD) individuals.

Objectives: (1)To identify visualization strategies of promise that are appealing, accessible, and informative to autism clinical experts without significant data analysis or eye-tracking expertise. (2)To show visualizations of gaze patterns during an eye tracking experiment to clinicians, to obtain their feedback regarding between group or questions of clinical phenotype, to distill this feedback into testable hypotheses through qualitative data extraction, and then to statistical test these hypotheses as a template for a visualization-guided analysis of eye movements in children with ASD.

Methods: Visualizations:(1)gaze points represented by participant identifier; (2)300 ms historical gaze trajectory; (3)“heatmap” color representation of groups (i.e. gaze points convolved with Gaussian kernels); (4)combination of (1)+(3); (5)thresholded version of (3). Visualizations were applied to Activity Monitoring(AM) eye-tracking data from the interim dataset (Summer 2018; 6-to-12-year-old children(TD:n=64; ASD:n=161)) of the Autism Biomarkers Consortium for Clinical Trials (ABC-CT). AM-gaze visualizations were presented to clinicians at a metropolitan autism center (ARNP/RN n=11, BCBA n=2, MD n=1, Clinical Psychologist/Therapist n=8, Family services/CRA n=2; combined clinical experience= 287 years), and feedback regarding visualization preferences, as well as clinical insights from visualizations aimed at describing (1)ASD-TD between-group differences; and (2)Lower(LIQ; IQ<85) from Higher(HIQ; IQ>85) IQ, were requested.

Results: 14 out of 15 clinicians favored the “Threshold-HeatMap” visualization (5), see Figure. Qualitative data extraction revealed the following insights by clinicians, some of which were investigated statistically:

  1. in TD vs ASD,
  • TD>ASD on looking at people/faces on overall (p<.001,d=0.97),
  • TD>ASD on looking at peoples when
    • People reached for the object (p=.001,d=0.94),
    • Actors were not talking and/or in anticipation of speech or activity (p<.001,d=.91),
  • ASD>TD on looking at toys and the central activity(p<.001,d=-1.07;p<.001,d=-1.06 during speech;p<.001,d=-1.09 during non-speech),
  • ASD seemed more likely to reference faces after speech,
  • ASD was slower to disengage from objects/people.

2. Comparing lower versus higher IQ in ASD,

  • LIQ>HIQ on looking at distractors(p<.001,d=.76),
  • LIQ responded to conversation less in overall (p<.001,d=.82), and more slowly,
  • LIQ showed more scattered gaze patterns,
  • LIQ spent more time looking at background objects(p<.001,d=.59).

Conclusions: Results indicated that a simplified, thresholded visualization was preferred by clinicians. Clinicians were able to identify multiple hypotheses which were then confirmed analytically. This process may provide a template for future explorations that will increase accessibility to experimental data, allowing clinical expertise to be leveraged in biomarker discovery, moving us to a future where crowdsourcing may help us identify new analytical and data insights.