Automated Autism Detection Based on Characterizing Observable Patterns From Photos
Alice Z. Guo
Abstract
Autism spectrum disorder (ASD) is a developmental disorder that affects the communication and behavior. People with ASD show atypical attentions to social stimuli and gaze at human faces and complex scenes in an unusual way, and their facial expressions are often atypical as well. This article investigates the feasibility of developing an automated method to analyze the visual cues of autism using the photos taken by people with ASD, comparing to photos taken by people without ASD, in different scenarios. It was inspired by a recent study based on manual inspection of the photos. The key challenge is what and how to characterize the photos taken by people with ASD, to facilitate an automated separation from normal people. Several features are proposed to characterize the observable behaviors for ASD with experimental validations. This is the first work to perform an automatic analysis of the photos taken by people with ASD, achieving a prediction accuracy of 85.8 percent.