JMIR Publications

ABSTRACT

Background:

Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder in school-aged children. The lack of objective biomarkers for assessment often results in missed and misdiagnosed cases, giving rise to incorrect or untimely interventions. Eye-tracking technology provides an objective method to measure children’s neuropsychological behavior.

Objective:

The purpose of this research was to develop an impartial and trustworthy auxiliary diagnostic system for ADHD assessment using eye-tracking technology. This system could be beneficial for community and campus screening and could offer objective biomarkers for clinical diagnosis.

Methods:

This was a case-control study of children with ADHD and typically developing (TD) children. We designed an eye-tracking assessment paradigm based on the core cognitive deficits of ADHD and extracted a range of effective digital biomarkers to comprehensively represent the behaviors of participants. Various biomarkers were compared between the ADHD and TD groups, as well as their developmental patterns with aging. Machine learning (ML) was implemented to validate whether the extracted eye-tracking biomarkers could accurately predict ADHD. The performance of ML models was evaluated using k-fold cross-validation.

Results:

This study involved 216 participants. Among them, 94 of them were children with ADHD, and 122 were TD. The ADHD group significantly underperformed the TD group in the pro-, anti-, and delayed-saccade tasks with respect to accuracy and completion time. Additionally, there were noticeable differences between the groups in digital biomarkers, such as pupil diameter fluctuation, regularity of gaze trajectory, fixations on uninterested areas, etc. Although the ADHD group improved with aging in terms of accuracy and task completion speed, their eye movement patterns remained irregular. The 5-6-year-old TD group outperformed the 9-10-year-old ADHD group, and remained relatively stable across subsequent ages, indicating a unique developmental pattern in the ADHD group. The ML model was effective in discriminating the groups, achieving an area under the curve (AUC) of 0.965 and an accuracy of 0.908.

Conclusions:

The eye-tracking biomarkers proposed in this study could effectively demonstrate the differences in eye movement patterns between ADHD and TD groups in various aspects. In addition, the machine learning model constructed using these digital biomarkers showcased an accurate and reliable performance in identifying ADHD. The developed system was expected to facilitate early screening in schools and communities, and to provide clinicians with objective biomarkers as a reference.