Frontiers in Neuroscience (Nov 2022)

RJAfinder: An automated tool for quantification of responding to joint attention behaviors in autism spectrum disorder using eye tracking data

  • Jie Zhang,
  • Jie Zhang,
  • Ziyi Li,
  • Yige Wu,
  • Adam Yongxin Ye,
  • Lei Chen,
  • Xiaoxu Yang,
  • Qixi Wu,
  • Liping Wei

DOI
https://doi.org/10.3389/fnins.2022.915464
Journal volume & issue
Vol. 16

Abstract

Read online

Deficits in responding to joint attention (RJA) are early symptoms of autism spectrum disorder (ASD). Currently, no automated tools exist for identifying and quantifying RJA behaviors. A few eye tracking studies have investigated RJA in ASD children but have produced conflicting results. In addition, little is known about the trajectory of RJA development through developmental age. Here, a new video was designed including 12 clips of an actor pointing to or looking at an object. Eye tracking technology was used to monitor RJA in three groups: 143 ASD children assessed with the Autism Diagnostic Interview-Revised (ADI-R) and the Autism Diagnostic Observation Schedule (ADOS) (4–7 years old), 113 age- and gender-matched typically developing children (TDC), and 43 typically developing adults (TDA) (19–32 years old). RJAfinder was developed in R and MATLAB to quantify RJA events from the eye tracking data. RJA events were compared among the three groups. Spearman correlation coefficients between total number of RJA events in ASD and the Social Responsiveness Scale (SRS) scores were calculated. A logistic regression model was built using the average valid sampling rate and the total number of RJA events as two predictive variables to classify ASD and TDC groups. ASD children displayed statistically significantly less RJA events than the TDC and TDA groups with medium-to-large-sized effects. ASD and TDC children both displayed more RJA events in response to pointing stimuli than to looking stimuli. Our logistic regression model predicted ASD tendency with 0.76 accuracy in the testing set. RJA ability improved more slowly between the ages of 4–7 years old in the ASD group than in the TDC group. In ASD children, RJA ability showed negative correlation with SRS total T-score as well as the scores of five subdomains. Our study provides an automated tool for quantifying RJA and insights for the study of RJA in ASD children, which may help improve ASD screening, subtyping, and behavior interventions.

Keywords