Convolutional Neural Network for Prediction of Autism based on Eye-tracking Scanpaths

Autism Spectrum Disorder (ASD), difficulty in socialization, can be detected by observation of atypical visual attention of children. Eye tracking is one of the most important techniques used in providing information on visual behavior as a statistically-motivated step towards the accurate diagnosis of such disorder. The scanpath, sequences of fixations of the eyes on image, provides data related to the locations and durations of the gazes that can be used to develop visual patterns to analysis the visual behavior of children. The aim of this paper is to develop a deep learning model implementing a convolutional neural network (CNN) to classify children to autistic and typically-developing according to eye tracking scanpaths. The model was applied on 29 autistic and 30 normal children and achieved 98% testing accuracy.