[Machine learning algorithms for identifying autism spectrum disorder through eye-tracking in different intention...

Area of Science:

• Neuroscience
• Computer Science
• Developmental Psychology

Context:

• Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by differences in social interaction and communication.
• Visual perception and attention patterns in children with ASD differ from those in typically developing (TD) children.
• Objective and reliable screening tools for ASD are crucial for early intervention.

Purpose:

• To compare visual perception differences between children with ASD and TD children using intention-based videos.
• To evaluate the effectiveness of machine learning algorithms in distinguishing between ASD and TD children based on eye-tracking data.
• To explore the potential of eye-tracking metrics for developing objective ASD screening methods.

Summary:

• Eye-tracking data from 58 children with ASD and 50 TD children were analyzed using various machine learning classifiers while viewing videos depicting joint and non-joint intentions.
• Initial models achieved up to 81.90% accuracy. A refined model using a decision tree classifier with feature reconstruction significantly improved performance.
• The optimized model demonstrated high accuracy (91.43%), specificity (89.80%), and sensitivity (92.86%), with an AUC of 0.909, indicating strong discriminative power.

Impact:

• Eye-tracking combined with machine learning offers a promising, objective method for ASD identification.
• This research provides a foundation for developing faster and more accurate ASD screening tools.
• The findings highlight the utility of computational approaches in understanding neurodevelopmental differences.