ADHD classification by feature space separation with sparse representation

In recent years, increasing attention has been paid to Attention Deficit Hyperactivity Disorder (ADHD) as one of the most common neurobehavioral diseases in school-age children. In this paper, sparse representation is used as an effective tool to analyze the corresponding functional brain connectivities from functional MRI data. Two diagnosis models, i.e., ADHD and healthy control models, are employed to deal with the ADHD patients and control subjects. In these models, we learn the feature spaces of ADHD and control components adaptively via dictionary learning. Moreover, we also perform an energy operation as a penalty term in the models, which forces the feature energy of subjects minimized in the wrong feature spaces. Following these models, we carefully design a framework for ADHD classification, where more sophisticated methods are adopted including synthetic minority oversampling technique and SVM. In the experiments on ADHD-200 dataset, our method can achieve a remarkable accuracy compared with several state-of-the-art classification methods.

I. Introduction

Attention Deficit Hyperactivity Disorder (ADHD) is viewed as one of the most common mental-health disorders, such that about 5% of school-age children and 2–4% of adults suffer this disease [1]. Though the diagnosis of ADHD usually depends on the judgment of skilled clinicians via a complicated procedure, Blood-Oxygen-Level Dependent functional Magnetic Resonance Imaging (BOLD-fMRI) provides another way to discover more information about ADHD. It shows brain abnormalities of ADHD exist in several regions, e.g., prefrontal, inferior frontal and sensorimotor cortices [2]. However, the underlying mechanism of ADHD cannot be fully exploited only from the above anatomical aspects, since any cooperative information among these brain regions is not referred. Fortunately, by using the temporal and spatial coherence of BOLD signal, a Functional Connectivity (FC) pattern of the brain is built from the fMRI data, and successfully identifies more fundamental differences between ADHD and control subjects. As a result, FC analysis becomes a popular tool in the ADHD diagnosis and classification.