Deriving ADHD biomarkers with sparse coding based network analysis

Sparse coding has been increasingly used to explore brain networks using functional magnetic resonance imaging (fMRI). However, modeling and comparing brain network based on sparse coding is still challenging, especially in clinical applications. In this study, we propose a novel temporal sparse coding method to identify functional connectivity biomarkers in patients with Attention-Deficit/Hyperactivity Disorder (ADHD). Specifically, a group-wise temporal sparse coding method was proposed to localize corresponding brain regions of interest (ROIs) in rsfMRI data. The localized common ROIs were then used as brain network nodes for further functional connectivity analysis. By using a publicly available ADHD-200 dataset, we demonstrated that our method can identify functional connectivity biomarkers with improved performance in patient-healthy controls classification compared with the widely used independent component analysis (ICA).