Utilizing Machine Learning Techniques to Identify Autism Spectrum Disorder Using fMRI Data

Abstract

This study extensively investigates Autism Spectrum Disorder (ASD) classification by analyzing functional connectivity derived from functional magnetic resonance imaging (fMRI) data. ASD is a neurological development disorder affecting 1 in 36 US children, underscoring the importance of early detection for effective intervention. Using Autism Brain Imaging Data Exchange (ABIDE) fMRI data, this research evaluates the predictive capabilities of the Automated Anatomical Labeling (AAL), cc200, and cc400 brain atlases. Functional connectivity is assessed through correlation, covariance, tangent, partial correlation, and precision measures. Support Vector Machines (SVMs) and a proposed Convolutional Neural Network (CNN) are employed for classifying ASD, with the CNN achieving comparable results: 68.11% accuracy, 73.45% AUC, 73.41% recall, and 69.27% precision. Notably, correlation, tangent, and covariance measures show robust performance across the assessed brain atlases. This research provides valuable contributions by thoroughly comparing various functional connectivity analyses for ASD classification, shedding new light on their comparative effectiveness.