Advancing autism prediction through visual-based AI approaches: integrating advanced eye movement analysis and shape recognition with Kalman filtering

Abstract

In the recent past, the global prevalence of autism spectrum disorder (ASD) has witnessed a remarkable surge, underscoring its significance as a widespread neurodevelopmental disorder affecting children, with an incidence rate of 0.62%. Individuals diagnosed with ASD often grapple with challenges in language acquisition and comprehending verbal communication, compounded by difficulties in nonverbal communication aspects such as gestures and eye contact. Eye movement analysis, a multifaceted field spanning industrial engineering to psychology, offers invaluable insights into human attention and behavior patterns. The present study proposes an economical eye movement analysis system that adroitly integrates Neuro Spectrum Net (NSN) techniques with Kalman filtering, enabling precise eye position estimation. The overarching objective is to enhance deep learning models for early autism detection by leveraging eye-tracking data, a critical consideration given the pivotal role of early intervention in mitigating the disorder’s impact. Through the synergistic incorporation of NSN and contrast-limited adaptive histogram equalization for feature extraction, the proposed model exhibits superior scalability and accuracy when compared to existing methodologies, thereby holding promising potential for clinical applications. A comprehensive series of experiments and rigorous evaluations underscore the system’s efficacy in eye movement classification and pupil position identification, outperforming traditional Recurrent Neural Network approaches. The dataset utilized in the aforementioned scholarly article is accessible through the Zenodo repository and can be retrieved via the following link: [https://zenodo.org/records/10935303?preview=1].