Distance Metric Learning Using Graph Convolutional Networks: Application to Functional Brain Networks

Abstract

Abstract. Evaluating similarity between graphs is of major importance in several computer vision and pattern recognition problems, where graph representations are often used to model objects or interactions between elements. The choice of a distance or similarity metric is, however, not trivial and can be highly dependent on the application at hand. In this work, we propose a novel metric learning method to evaluate distance between graphs that leverages the power of convolutional neural networks, while exploiting concepts from spectral graph theory to allow these operations on irregular graphs. We demonstrate the potential of our method in the field of connectomics, where neuronal pathways or functional connections between brain regions are commonly modelled as graphs. In this problem, the definition of an appropriate graph similarity function is critical to unveil patterns of disruptions associated with certain brain disorders. Experimental results on the ABIDE dataset show that our method can learn a graph similarity metric tailored for a clinical application, improving the performance of a simple k-nn classifier by 11.9% compared to a traditional distance metric.

Dynamic Edge-Conditioned Filters in Convolutional Neural Networks on Graphs

“…The ability to compute node embeddings has given rise to many application, such as link prediction (Schlichtkrull et al, 2017), (semi-supervised) node classification in citation networks (Kipf and Welling, 2016a), performing logical reasoning tasks (Li et al, 2016b), finding correspondences across meshes (Monti et al, 2017) or point cloud segmentation (Chapter 4). Graph embeddings has also been useful in many tasks, for example measuring similarity of brain networks by metric learning (Ktena et al, 2017), suggesting heuristic moves for approximating NP-hard problems (Dai et al, 2017), predicting satisfaction of formal properties in computer programs (Li et al, 2016b), classifying chemical effects of molecules (Chapter 3) and regressing their physical properties (Gilmer et al, 2017), or point cloud classification (Chapter 3).…”

Graph Neural Network for Interpreting Task-fMRI Biomarkers

“…Significant progress has been made using functional magnetic resonance imaging (fMRI) to characterize the brain remodeling in ASD [9]. Recently, emerging research on Graph Neural Networks (GNNs) has combined deep learning with graph representation and applied an arXiv:1907.01661v2 [cs.LG] 12 Jul 2019 integrated approach to fMRI analysis in different neuro-disorders [11]. Most existing approaches (based on Graph Convolutional Network (GCN) [10]) require all nodes in the graph to be present during training and thus lack natural generalization on unseen nodes.…”

Simple 1-D Convolutional Networks for Resting-State fMRI Based Classification in Autism

“…In this context, [15], [20] achieved a maximum accuracy of 70.4% and 70.86%, using as feature vector(s) either the similarity matrix or the mean time series for each of the 111 regions of the Harvard-Oxford atlas [12]. Importantly, another work, [19], based on GCN reported a high variability in the accuracy achieved across sites (50% to 90%). Finally, the highest accuracy in classifying participants in the ABIDE database so far was obtained by [18] by using an ensemble learning strategy on GCN.…”