|
Abstract
|
Aim
Eye tracking can measure the duration of eye contact and the direction of gaze movements, providing measurable indicators of social communication deficits. An important aspect of social communication is keeping eye contact for children with Autism Spectrum Disorders (ASD). Emotion recognition behavior may vary among children with major neurodevelopmental disorders such as ASD, and Attention Deficit Hyperactivity Disorder (ADHD). We designed an experimental environment for emotion recognition through eye gaze tracking (EGT) data. Application log data (APL) was utilized to examine the differences among the study groups using machine learning (ML) techniques.
Methods
Eighteen ASD, 35 ADHD, and 15 control children participated in the study at the Marmara University Child and Adolescent Psychiatry Clinics. All participants had an Intelligence Quotient score above 70, and children in the ASD group who could read and write were included in the study. Data were collected using SMI Eye Tracking Glasses for the EGT fixation data, and the web-based application TrackEmo, which consists of 40 emotive human face images from the Cohn-Kanade database. APL data consisted of the response and reaction time of the participants. Random Forest, Support Vector Machine, and Logistic Regression classifiers were applied to the raw fixation data alongside the EGT log data as (ML) procedures. We used classification algorithms with the Tomek links noise removal method to detect the participant's diagnosis.
Results and conclusion
There were 12 ASD, 12 ADHD, and 10 participants in the control group. Mean ages were 9.20 ± 1.17 years (8 females, 16 males) for the study group and 9.50 ± 1.12 years (4 females, 6 males) for the control group. We obtained 5990 fixation points (FP) from 12 participants with ASD, 4897 FP from 12 participants with ADHD, and 3813 FP from 10 controls as raw fixation data. The average normalized pupil diameters of the participants with ADHD were less than the participants with ASD and the control group while they were looking at angry emotions (p < 0.0001). The pupil size of the participants with ASD was smaller than the control group (p < 0.0001). The highest classification accuracy results were reported as 86.36% for ASD vs. Control, 81.82% for ADHD vs. Control, and 70.83% for ASD vs. ADHD. This study provides evidence that EGT and APL data may have distinguishing features for diagnosing ASD and ADHD.
|
