Detecting Autism Based on Eye-Tracking Data from Web Searching Tasks

Abstract

The ASD diagnosis requires a long, elaborate, and expensive procedure, which is subjective and is currently restricted to behavioural, historical, and parent-report information. In this paper, we present an alternative way for detecting the condition based on the atypical visual-attention patterns of people with autism. We collect gaze data from two different kinds of tasks related to processing of information from web pages: Browsing and Searching. The gaze data is then used to train a machine learning classifier whose aim is to distinguish between participants with autism and a control group of participants without autism. In addition, we explore the effects of the type of the task performed, different approaches to defining the areas of interest, gender, visual complexity of the web pages and whether or not an area of interest contained the correct answer to a searching task. Our best-performing classifier achieved 0.75 classification accuracy for a combination of selected web pages using all gaze features. These preliminary results show that the differences in the way people with autism process web content could be used for the future development of serious games for autism screening. The gaze data, R code, visual stimuli and task descriptions are made freely available for replication purposes.