fMRI Functional Connectivity Augmentation Using Convolutional Generative Adversarial Networks for Brain Disorder Classification

Recent applications of deep learning (DL) methods to brain disorder classification use functional connectivity (FC) from functional magnetic resonance imaging (fMRI) data as features. However, the classification performance has been limited by the small number of fMRI samples and over-fitting problem in the DL model training. We propose a novel framework based on deep convolutional generative adversarial network (DCGAN) to augment fMRI FC data for classifying altered brain networks. We develop a specialized DCGAN architecture for FC synthesis, which builds on multiple convolutional layers for hierarchical latent representation in both the generator and discriminator, in order to generate connections of signed-weighted FC networks, while preserving the spatial structure. We consider the 1-dimensional and 2-dimensional convolution of the DCGANs. The generated data are then used to improve the performance and generalizability of downstream FC classifiers. Results on major depressive disorder (MDD) identification using resting-state fMRI show substantial improvement in classification accuracy after data augmentation by the proposed models, outperforming several state-of-the-art FC classifiers without augmentation. The synthetic FCs also reveal close resemblance in structural patterns to the real data for both MDD and healthy subjects.