Autism Spectrum Disorder Screening Through Behavioral Gaze Trajectories and Electroencephalography Signal Classification

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that needs accurate early screening for early intervention. Early diagnosis is crucial for infants when their neuroplasticity allows treatment that could effectively hinder long-term developmental outcomes. Indeed, traditional methods are employed such as the Modified Checklist for Autism in Toddlers (M-CHAT), Autism Diagnostic Interview-Revised (ADI-R), and Autism Diagnostic Observation Schedule (ADOS). Yet, these approaches are subjective, which makes them susceptible to bias and misdiagnosis. Therefore, this research aims to identify an objective and accurate non-invasive biomarker for early diagnosis of ASD.

Firstly, a machine learning system that analyses Electroencephalography (EEG) signals and outputs binary classification results will be used. This system was developed using EEG Representation learning and an EEG feature extractor that extracts unique and relevant features of EEG signals. In addition, Gaze trajectories, which detect the patient's eye movements, were used. A gaze estimation network was created by obtaining yaw and pitch angles and using eye-tracking technology.

Our proposed system for EEG-based classification recorded an accuracy of 98.29%, sensitivity of 98.29%, precision of 98.34%, and F1-score of 98.32%. This was significantly higher than the baseline CNN model, which achieved 94.54% accuracy, supporting the effectiveness of the proposed system. In addition, the gaze-based classifier achieved 99% classification accuracy.

These findings indicate that EEG representation learning and gaze trajectory analysis provide an accurate and objective approach for ASD screening. The system minimizes subjectivity while allowing rapid binary classification for early diagnosis. This shows strong potential for clinical application in ASD detection by integrating biomarkers.