ABSTRACTFunctional network connectivity (FNC) obtained from resting-state functional magnetic resonance imaging (fMRI) data have been commonly used to study mental disorders in neuroimaging applications. Likewise, generative adversarial networks (GANs) have performed well in multiple classification benchmark tasks. However, the application of GANs to fMRI is relatively rare. In this work, we proposed an FNC-based GAN for classifying brain disorders from healthy controls (HCs), in which FNC matrices were calculated by correlation of time courses derived from non-artefactual fMRI independent components (ICs). The proposed GAN model consisted of one discriminator (real FNCs) and one generator (fake FNCs), each has four fully-connected layers, and feature matching was implemented between each other to improve classification performance. An average accuracy of 70.1% with 10-fold cross-validation was achieved for classifying 269 major depressive disorder (MDD) patients from 286 HCs, at least 5.9% higher compared to other 6 popular classification approaches (54.5-64.2%). In another application to discriminating between 558 schizophrenia patients and 542 HCs from 7 sites, the proposed GAN model achieved 80.7% accuracy in leave-one-site-out prediction, outperforming support vector machine (SVM) and deep neural net (DNN) by 3-6%. To the best of our knowledge, this is the first attempt to apply GAN model based on fMRI data for mental disorder classification. Such a framework promises wide utility and great potential in neuroimaging biomarker identification.
Dynamic Granger causality based on Kalman filter for evaluation of functional network connectivity in fMRI data