VISUAL PROCESSING ANALYSIS AND MACHINE LEARNING IN THE EARLY DIAGNOSIS OF ASD USING EYE TRACKING

Abstract

This non-clinical study explores the effectiveness of holistic processing in facial recognition and the application of eye-tracking systems in the diagnosis of Autism Spectrum Disorder (ASD). Three approaches are adopted: Composite Faces, Part-to-Whole and Flip Effect, highlighting the importance of holistic analytics for efficient facial recognition.

The research utilizes camera-based eye-tracking systems, which are notable for their non-invasive approach and accuracy in detecting specific eye movements. OGAMA® software and data mining tools such as Orange Canvas are used to analyze eye metrics. The methodology includes the identification, storage, and processing of oculometric variables using supervised learning algorithms to predict behavioral patterns in individuals with ASD.

The experiments carried out demonstrated the effectiveness of the proposed methodology. Reference data were used to validate the findings, and machine learning techniques were employed to differentiate individuals with ASD, with Neural Networks standing out as the most effective algorithm.

It is concluded that the combination of eye tracking with data mining offers valuable insights for the diagnosis and understanding of ASD, opening up new possibilities for research in holistic processing and contributing significantly to the fields of psychology, medicine and assistive technologies.