Eye Tracking on an Unmodified Ipad for Visual Attention of Children with and without ASD: A Feasibility Test

Background:  Eye tracking is widely used in autism research to study visual social cognition. However, most studies depend on expensive commercial eye-tracking devices, which dramatically restricts accessibility. In this study, we used video-based eye tracking-algorithms to implement and validate eye tracking on an unmodified tablet, an advance which could increase the accessibility of eye tracking for researchers and families.

Objectives:  1) To implement eye tracking on an unmodified iPad, 2) To validate calibration and looking time by using smooth pursuit, 3) To compare looking behaviors of children with Autism Spectrum Disorder (ASD) with controls.

Methods:  Participants were 21 children (4 to 8 years old: 6 typically developing (TD), 4 developmentally delayed (DD), and 11 children with autism spectrum disorder (ASD)) and 6 TD adults. Participants watched three 5-minutes videos on an iPad; each video video contained three 5-point calibrations and two animated smooth pursuit trajectories. Using front facing camera recordings, we applied image processing to track head and eye movement for use in eye tracking calculations. Calibration errors and proportion of valid looking on screen (Valid%) were calculated. Percentage of looking during smooth pursuit was calculated as a validation and measure of sustained attention. Repeated measures ANOVA was conducted to examine between group differences.

Results:  Adults failed 5.6% of calibrations (failure defined as looking at fewer than 3 of 5 calibration target points), and had an average calibration error (1.1 deg) and Valid% (81.7% ±4.0%) significantly better than children (p = .005, F = 4.364; p<.001, F = 21.6). Similarly adults looked more during smooth pursuit trajectory validations compared to children (83.3% ±2.9%, p<.001, F =12.3).

The TD group failed 11.1% of calibrations, DD failed 30% and ASD 22.2%. However, in passed calibrations errors were comparable between the three groups: TD 1.5 deg, DD 1.3 deg and ASD 1.4 deg. Valid% was not different between the three groups of children (58.7% ± 4% in TD, 52.3% ± 4.7% in DD with p = .303, and 53.2% ± 3% in ASD with p=.280 between TD and ASD). Both TD (56.0% ± 3%, p = .049) and DD (59.8%±3.5%, p = .007) groups exhibited significantly higher proportions of following the smooth pursuit trajectory than the ASD children (48.6%±2.3%). There was no performance difference between DD and TD (p=.397).

Conclusions:  The preliminary data found that children with ASD and DD looked at less at calibration points than their TD peers, and that the ASD group showed lower percentage of valid looking while following smooth pursuit trajectories than TD and DD groups. This study provides evidence for the application of iPad eye tracking as a viable option for tracking eye movements and studying visual attention in children with and without ASD, but also points towards tablet-based eye tracking as a method for understanding potentially broader issues in sustained attention in children with ASD.