Identifying Autism Gaze Patterns in Five-Second Data Records.

Area of Science:

• Neuroscience
• Developmental Psychology
• Biomedical Engineering

Background:

• Diagnosing autism spectrum disorder (ASD) often requires extensive data collection, posing challenges, especially with children.
• Current diagnostic methods for ASD can be time-consuming and data-intensive.

Purpose of the Study:

• To investigate novel classifiers for ASD using short eye-tracking data sets (5 seconds).
• To introduce and evaluate a new metric for distinguishing ASD gaze patterns from typically developed (TD) individuals.
• To compare the proposed metric against traditional eye-tracking metrics and state-of-the-art AI classifiers.

Main Methods:

• Utilized 5-second eye-tracking data from children with ASD and TD children.
• Introduced a novel metric for analyzing gaze patterns.
• Compared the new metric's performance against established eye-tracking metrics and a leading AI classifier.

Main Results:

• The new metric achieved 93% accuracy in classifying eye gaze trajectories between ASD and TD children.
• The 5-second metric surpassed traditional eye-tracking metrics in classification accuracy.
• Performance was comparable to state-of-the-art AI benchmarks, even those trained on longer time series.

Conclusions:

• A simple, 5-second eye-tracking metric can effectively differentiate ASD gaze patterns.
• This method offers a less data-intensive and more straightforward alternative to current diagnostic approaches.
• While not a substitute for medical diagnosis, it aids in identifying potential visual attention differences in ASD.