Vision-Based Approach for Autism Diagnosis Using Eye Tracking based on Machine Learning and Deep Learning

Abstract

Although the early diagnosis of autism spectrum disorder (ASD) is highly sought, it is still a challenging task that requires a battery of cognitive tests and hours of clinical exams to be properly accomplished. ASD may manifest itself in various ways, which makes diagnosing it even more challenging. However, although diagnostic tests are generally devised by specialists, human mistakes are still possible. Computer-assisted technologies can be quite useful in this respect for aiding in the selection procedure. The use of eye tracking as a crucial component of ASD screening evaluation based on the distinctive characteristics of the eye gaze is continued in this article. This research adds to the growing body of evidence that eye tracking technologies may help in ASD screening. Eye tracking, visualization, using machine learning and deep learning are part of the suggested solution. Eye-tracking scan paths were first converted into a visual representation in the form of a series of pictures. After that, an image classification job was taught to a convolutional neural network. The findings of the experiments showed that using a visual representation simplified the diagnostic process while also achieving excellent accuracy. The convolutional neural network model, in particular, can attain promising classification accuracy. This shows that visuals are capable of encoding information about gaze movements and its underlying dynamics. Based on the greatest information coefficient, we examined potential relationships between autism severity and eye movement dynamics. The results demonstrate that using the great potential in combining eye tracking, visualization, machine learning, and deep learning to create an impartial tool to help in ASD screening. Generally, the approach we propose may be used to identify various diseases, including neurodevelopmental problems.