A Deep Neural Network Approach for Autism Spectrum Disorder Detection Using Multiple Brain Atlases with Resting-State Functional Magnetic Resonance Imaging Data

The exploration of Deep Neural Network (DNN) classifier models for Autism Spectrum Disorder (ASD) detection has attracted considerable interest, given its potential for early diagnosis and intervention. In recent developments, research has shifted its focus to using multisite brain imaging datasets to increase clinical validity. In this paper, we propose a framework that utilizes the Autism Brain Imaging Data Exchange (ABIDE)-1 dataset, comprising resting-state functional magnetic resonance imaging (rs-fMRI) data for ASD Detection. The ABIDE-1 dataset has rs-fMRI data of 1035 individuals. We extract features from the functional connectivity of each individual's rs-fMRI using various brain atlases (Automated Anatomical Labeling – AAL, Craddock 200 – CC200, Craddock 400 – CC400) along with subject phenotype informations. We used Multi-Layer Perceptrons (MLPs) and Convolutional Neural Networks (CNNs) for ASD detection. The MLP focused on discerning patterns within the data, while the CNN captured complex spatio-temporal patterns in rs-fMRI data. Our results showed that the proposed framework using CNN classifier with CC200 atlas has an accuracy of 93.3%, 96.83% precision, 89.7% recall, and a 92.92% F1 score. Our framework demonstrates substantial improvements, increasing accuracy by 6%. This study highlights the potential of neuroimaging data and advanced computational techniques for enhancing early ASD detection and understanding.