A Comparative Analysis of Brain Atlases and DNN-Based Autism Spectrum Disorder Detection Using IMPAC Dataset

An autism spectrum disorder (ASD) is a physiological condition that impairs an individual's ability to acquire skills, engage in social interactions, empathize, and socialize. Early intervention is crucial to alleviate intellectual, behavioral, and functional challenges related to ASD. The significance of early ASD assessment cannot be overstated, as it is rapidly addressed through cognitive progress, yielding equally favorable outcomes. This research work developed a machine learning (ML) model for early assessments of autism spectrum disorder (ASD) utilizing functional magnetic resonance imaging (fMRI) data and deep neural networks (DNNs). A comprehensive analysis is carried out utilizing seven distinct brain atlases from the IMPAC dataset. The results indicate that among the seven atlases, Craddock 200 (CC200) attains the highest ROC AUC of 85.091% and an impressive accuracy of 82.759<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, which outperforms numerous current state-of-the-art models. These findings emphasize the potential of fMRI-based machine learning models, especially deep neural networks, in improving the precision and efficacy of early autism spectrum disorder diagnosis. Such enhancements offer potential for enhancing the lives of individuals with ASD, as prompt treatments can profoundly influence their developmental trajectory and general well-being.