Deep Learning Approached Features for ASD Classification using SVM

Autism spectrum disorder (ASD) is a neurodevelopmental spectrum disorder having impacts on a person's adaptive functioning, such as communication difficulties and social interactions. The quantitative analysis of ASD is reasonably challenging, and research is enduring to find out an applicable method for ASD diagnosis. Functional magnetic resonance imaging (fMRI) plays an imperative role in identifying a conceivable diagnosis system using machine learning approaches integrated with deep learning techniques. In this work, we build a model comprising Restricted Boltzmann Machine (RBM) for extracting features from fMRI data and Support Vector Machine (SVM) for classifying ASD subjects from healthy controls. Before that, we perform several data processing steps involving appropriate slice time correction as well as normalization. We apply our method on a dataset including 105 Typical control (TC) and 79 ASD subjects from a renowned database called ABIDE. The outcomes demonstrate that the proposed framework performs outstandingly to classify ASD using the grid-search cross-validation. The findings also indicate that the amalgamation of RBM and SVM methods may be applied as a future tool to diagnose ASD.