Classifying Eye Gaze Patterns of ASD And Typically Developing Children Using Convolutional Neural Network

This research shows that utilising several datasets, autism spectrum disorder (ASD) may be detected at an early stage. The first of the many goals of this thesis is to describe how to use an eye tracking dataset in conjunction with convolutional neural network data to classify children as having autism. The Saliency4ASD Visual centre demonstration for ASD provided the eye tracking data that was acquired at IEEE ICME'19. For 300 pictures, the collection includes focus maps for kids with ASD and TD. It is possible to determine if an observer has ASD by looking at their fixation maps, which reveal the areas of the visual cortex that respond to certain stimuli. To achieve this goal, we will utilise Convolution Neural Networks (CNN) to detect whether an individual has autism spectrum disorder or normal development (TD) just by looking at their photograph. Autism is a neurological syndrome that is marked by difficulties in socialising, communicating through both verbal and nonverbal means, repetitive activity, not playing with others, varying degrees of sensitivity to more or less sensual motives, and a strict adherence to routine. Using the Eye Gaze Image dataset, the model demonstrates its great performance in distinguishing between instances of ASD and TD. This is shown by the model's high accuracy when it correctly identifies two true positives without producing any false positives. When applied to this specific scenario, it would seem that the model constantly generates categories that are the most suitable. With the best accuracy (92.8%), precision (93%), recall (95.6%), and F1 score (93%), the proposed model outperforms both CNN and RF, demonstrating its superior performance when comparing children who are normally developing (TD) to those who have autism spectrum disorder (ASD).