PREDICTING AUTISM DIAGNOSIS USING IMAGE WITH FIXATIONS AND SYNTHETIC SACCADE PATTERNS.

Area of Science:

• Neuroscience
• Computer Science
• Developmental Psychology

Background:

• Autism spectrum disorder (ASD) signs appear early, but diagnosis is delayed, missing crucial intervention windows.
• Early intervention significantly improves outcomes for young children with ASD.
• A need exists for tools to identify ASD risk in infancy, enabling earlier support.

Purpose of the Study:

• To develop and evaluate machine learning methods for early ASD risk classification using infant scanpath data.
• To explore novel approaches for automatic ASD identification from visual attention patterns.

Main Methods:

• Two machine learning approaches were proposed: a synthetic saccade method and an image-based convolutional neural network method.
• Scanpath data from infants viewing natural images were used as input.
• The synthetic saccade approach incorporated generative models and auxiliary data for deep learning classification.

Main Results:

• The proposed methods achieved 65.41% accuracy in classifying ASD risk on a validation dataset.
• Both the synthetic saccade and image-based approaches demonstrated potential for automated ASD screening.

Conclusions:

• Machine learning, particularly using eye-tracking data, shows promise for early ASD risk detection in infants.
• These computational methods could facilitate earlier diagnosis and intervention, improving developmental trajectories for children with ASD.