Classification of Autism Based on fMRI Data with Feature-Fused Convolutional Neural Network

Abstract

Autism spectrum disorder (ASD) is a neurological development disorder. Due to the cause of the disease is not clear, the diagnosis of ASD mainly depends on the interactions between individuals and clinical professionals. Functional magnetic resonance imaging (fMRI) has been widely used in the study of brain function in patients with ASD, which provides a new way for the diagnosis of ASD. In this paper, we propose a convolutional neural network (CNN) classification method to classify ASD. The proposed method fuses two kinds of brain functional features, namely brain functional connectivity (FC) and amplitude of low frequency fluctuation (ALFF). Firstly, the two types of feature data which reflect different brain functions are extracted from the fMRI data of ASD patients and normal subjects. Then, CNN is utilized to fuse the two types of data and to predict the classification results. Finally, several experiments are carried out on the ABIDE (Autism Brain Imaging Data Exchange) datasets to test the performance of our proposed method. The fused feature CNN model is compared with the CNN model with only FC or ALFF features; it is also compared with three traditional machine learning methods. The results show that the feature-fused CNN classification model can improve the classification performance to a certain extent; it can be used for computer-aided diagnosis of ASD.