Graph Autoencoders for Embedding Learning in Brain Networks and Major Depressive Disorder Identification

Abstract

Brain functional connectivity (FC) reveals biomarkers for identification of various neuropsychiatric disorders. Recent application of deep neural networks (DNNs) to connectomebased classification mostly relies on traditional convolutional neural networks using input connectivity matrices on a regular Euclidean grid. We propose a graph deep learning framework to incorporate the non-Euclidean information about graph structure for classifying functional magnetic resonance imaging (fMRI)derived brain networks in major depressive disorder (MDD). We design a novel graph autoencoder (GAE) architecture based on the graph convolutional networks (GCNs) to embed the topological structure and node content of large-sized fMRI networks into low-dimensional latent representations. In network construction, we employ the Ledoit-Wolf (LDW) shrinkage method to estimate the high-dimensional FC metrics efficiently from fMRI data. We consider both supervised and unsupervised approaches for the graph embedded learning. The learned embeddings are then used as feature inputs for a deep fully-connected neural network (FCNN) to discriminate MDD from healthy controls. Evaluated on a resting-state fMRI MDD dataset with 43 subjects, results show that the proposed GAE-FCNN model significantly outperforms several state-of-the-art DNN methods for brain connectome classification, achieving accuracy of 72.50% using the LDW-FC metrics as node features. The graph embeddings of fMRI FC networks learned by the GAE also reveal apparent group differences between MDD and HC. Our new framework demonstrates feasibility of learning graph embeddings on brain networks to provide discriminative information for diagnosis of brain disorders.

A Network Science Perspective of Graph Convolutional Networks: A Survey

“…Second, GCNs serve as the basis for many other graph deep learning approaches, including graph autoencoders, graph reinforcement learning, and graph adversarial methods. For example, graph autoencoders leverage GCN-based encoders to learn meaningful embeddings for the nodes or graphs [172]. GCNs are also used to extract useful features from the graph, which are subsequently used by reinforcement learning algorithms to make decisions [173].…”

Graph Autoencoder-Based Embedded Learning in Dynamic Brain Networks for Autism Spectrum Disorder Identification

“…However, these studies mainly consider group-level network topology using phenotypic-based information, or use supervised graph-level embedding learning with pre-calculated population graph for static brain networks classification. Our recent work [28] developed an alternative approach based on graph autoencoders (GAE) that can efficiently learn graph representations of static brain networks to facilitate individual-level network classification, and results on the identification of major depressive disorders show superior performance over other DNN approaches. Our aim is to extend the GAE to learn representations from graph-based data to capture brain network topology that changes dynamically over time.…”

Deep Representations for Time-varying Brain Datasets

“…Method utilizing regularized pooling with GNN to identify fMRI biomarkers is also proposed [16]. Another work [18] embeds both topological structures and node signals of fMRI networks into low-dimensional latent representations for better identification of depression. However, the first two works [9,16] use time-averaged fMRI, losing rich dynamics in the temporal domain.…”

“…These works also did not incorporate structural modality that can provide extra connectivity information missing in the functional modality. The modeling in [18] combines nodes' temporal and feature dimensions instead of handling them separately, leading to a suboptimal representation (as discussed in section 3.2). To overcome these issues, we propose ReBraiD (Deep Representations for Time-varying Brain Datasets), a graph neural network model that jointly models dynamic functional signals and structural connectivities, leading to a more comprehensive deep representation of brain dynamics.…”