Deep Learning for Autism Detection Using Eye Tracking Scanpaths
R Supritha, Bharathi Mohan G
Abstract
Autism spectrum disorder (ASD) is a neurodevelopmental condition marked by difficulties in social interaction, communication, and limited repetitive behaviors, with symptoms varying significantly among individuals. Eye tracking holds promise in autism detection due to its unique ability to capture and analyze visual attention patterns, providing insights into the cognitive processes and atypical visual behaviors associated with ASD. Eye-tracking technology offers a unique perspective, allowing the observation and quantification of visual attention patterns, which may reveal distinctive features associated with ASD. These visualizations represent how individuals, particularly those with ASD, explore and engage with stimuli. In this research, we propose a novel approach using deep learning models, specifically DenseNet-201, EfficientNet B7, ResNet-50, and MobileNetV2, to analyze eye-tracking scan paths for ASD detection. The dataset focuses on visualizations of eye-tracking scan paths, primarily involving individuals with ASD. The study yielded promising results, with the deep learning models achieving accuracies of 94.97%, 94.74%, 84.21%, and 92.45%, respectively. DenseNet-201 demonstrated the highest accuracy at 94.97%. The research contributes to advancing early diagnosis and intervention strategies for individuals with ASD.