Predicting ASD Diagnosis in Children with Synthetic and Image-based Eye Gaze Data.

Area of Science:

• Neurodevelopmental Disorders
• Computational Neuroscience
• Pediatric Medicine

Background:

• Early intervention significantly improves outcomes for children with autism spectrum disorder (ASD).
• Atypical visual attention is a recognized characteristic of ASD, observable even in early childhood.
• Eye gaze data offers a potential biomarker for early ASD risk identification.

Purpose of the Study:

• To develop and evaluate machine learning methods for the automatic classification of ASD risk using eye gaze data.
• To enable early identification of ASD, facilitating timely intervention before the full spectrum of symptoms manifests.

Main Methods:

• Two machine learning approaches were developed: a synthetic saccade approach and an image-based approach.
• The synthetic saccade method uses generative models of eye movement patterns combined with real scan-path data.
• The image-based method employs deep learning (CNNs/RNNs) on fixation maps and visual stimuli.

Main Results:

• The proposed methods achieved 67.23% accuracy on the validation dataset.
• The models demonstrated 62.13% accuracy in predicting ASD risk on the independent test dataset.
• These results indicate the feasibility of using eye gaze analysis for early ASD screening.

Conclusions:

• Machine learning analysis of eye gaze data presents a viable strategy for early ASD risk detection.
• Automated screening tools based on visual attention patterns can support early diagnosis and intervention.
• Further research can refine these methods for improved accuracy and clinical application.