Abstract

Systems and methods for diagnosing autism spectrum disorder (ASD) using only functional magnetic resonance imaging (fMRI) data are provided. Machine learning infrastructure can be used to identify reliable biomarkers of ASD in order to classify patients with ASD from among a group of typical control subjects using only fMRI. A sparse autoencoder (SAE) can be used, resulting in optimized extraction of features that can be used for classification. These features can then be fed into a deep neural network (DNN), which results in classification of fMRI brain scans more prone to ASD. The model can be trained to optimize the classifier while improving extracted features based on both reconstructed data error and the classifier error.