A method for Early Autism Spectrum Disorder Screening based on Gaze Tasks

In this paper, an early intelligent screening method is proposed for autism spectrum disorder (ASD) based on eye-movement tasks. Currently, the prevalence of ASD in the United States is 2.8%, and the lack of early diagnosis is a significant factor. Our proposed method has three notable characteristics for ASD screening: 1) All data will be obtained through designed games, where user's eye movement information and game performance during gameplay will be collected, facilitating clinical implementation. 2) Task metrics related to gaze behavior are utilized to generate time series embedded with eye movement features, describing individual’s eye movement characteristics. 3) a Task-Driven Eye Movement Analysis Network (TDEA-Net) is designed to accept eye movement sequences, extracts contextual information at each time point, and inputs them into a classification network through multi-task feature vector splicing to enhance screening accuracy. In this study, experimental data were collected from 29 ASD and 30 typically developing (TD) children aged 5-8 years for model training and evaluation. The final classification accuracy achieved was 94.92%, with a sensitivity of 93.10% and specificity of 96.67%, which indicates that ASD and TD can be effectively distinguished in clinical diagnosis through games about gaze tasks.