ADET MODEL: Real time autism detection via eye tracking model using retinal scan images.

Abstract

Background: Deficits in concentration with social stimuli are more common in children affected by autism spectrum disorder (ASD). Developing visual attention is one of the most vital elements for detecting autism. Eye tracking technology is a potential method to identify an early autism biomarker based on children's abnormal visual patterns.

Objective: Eye tracking retinal scan path images can be generated by eyeball movement during the time of watching the screen and capture the eye projection sequences, which helps to analyze the behavior of the children. The Shi-Tomasi corner detection methodology uses open CV to identify the corners of the eye gaze movement in the images.

Methods: In the proposed ADET model, the corner detection-based vision transformer (CD-ViT) technique is utilized to diagnose autism at an early stage. Generally, the transformer model divides the input images into patches, which can be fed into the transformer encoder process. The vision transformer is fine-tuned to resolve binary classification issues once the features are extracted via remora optimization. Specifically, the vision transformer model acts as the cornerstone of the proposed work with the help of the corner detection technique. This study uses a dataset with 547 eye-tracking retinal scan path images for both autism and non-autistic children.

Results: Experimental results show that the suggested ADET frameworkachieves a better classification accuracy of 38.31%, 23.71%, 13.01%, 1.56%, 18.26%, and 44.56% than RM3ASD, MLP, SVM, CNN, SVM, and our proposed ADET methods.

Conclusions: This screening method strongly suggests that it be used to assist medical professionals in providing efficient and accurate autism detection.