Node-Centric Graph Learning From Data for Brain State Identification

Abstract

Data-driven graph learning models a network by determining the strength of connections between its nodes. The data refers to a graph signal which associates a value with each graph node. Existing graph learning methods either use simplified models for the graph signal, or they are prohibitively expensive in terms of computational and memory requirements. This is particularly true when the number of nodes is high or there are temporal changes in the network. In order to consider richer models with a reasonable computational tractability, we introduce a graph learning method based on representation learning on graphs. Representation learning generates an embedding for each graph node, taking the information from neighbouring nodes into account. Our graph learning method further modifies the embeddings to compute the graph similarity matrix. In this work, graph learning is used to examine brain networks for brain state identification. We infer time-varying brain graphs from an extensive dataset of intracranial electroencephalographic (iEEG) signals from ten patients. We then apply the graphs as input to a classifier to distinguish seizure vs. non-seizure brain states. Using the binary classification metric of area under the receiver operating characteristic curve (AUC), this approach yields an average of 9.13 percent improvement when compared to two widely used brain network modeling methods.

Decoding Intracranial EEG With Machine Learning: A Systematic Review

“…Fewer studies employed the Bonn dataset, where deep learning and a vector machine algorithm achieved greater accuracies compared to a standard multilayer perceptron (MLP) with a single hidden layer (Raghu and Sriraam, 2017 ; Vidyaratne and Iftekharuddin, 2017 ; Gong et al, 2020 ). Studies using EPILEPSIAE data all employed standard ML but specificity metrics were not reported for comparison (Manzouri et al, 2018 ; O'Leary et al, 2018 ; Ghoroghchian et al, 2020 ). Those using the Mayo-UPenn show similar accuracies with deep learning and a combination classifier of standard supervised algorithms (Hosseini et al, 2017 , 2018 ; Truong et al, 2018 ).…”

Improving J-divergence of brain connectivity states by graph Laplacian denoising

“…State-of-the-art graph learning methods have the limitation that they usually present over-simplified models for the signal on graph to overcome problems of computational and memory cost. Some recent works, such as [27], propose different strategies to deal with graph learning problems. Specifically, in the context of mental state identification, authors in [27] present a novel technique to create and modify embeddings associated to each graph node to efficiently compute the adjacency matrix.…”

“…Some recent works, such as [27], propose different strategies to deal with graph learning problems. Specifically, in the context of mental state identification, authors in [27] present a novel technique to create and modify embeddings associated to each graph node to efficiently compute the adjacency matrix. Since FC computation requires a lot of time and computational power, one possibility consists in clustering FC into relevant communities of synchronous components.…”

Missing Data Estimation in Temporal Multilayer Position-aware Graph Neural Network (TMP-GNN)

“…Graph neural networks (GNN) has recently been used as a standard in developing machine learning methods for graphs. Graphs that have been formed from transportation network [1], [2], [3], [4], brain network [5], social media community networks, etc. The GNN architecture has effectively combined the node / edge features and graph topology to build distributed representation.…”

“…The GNN architecture has effectively combined the node / edge features and graph topology to build distributed representation. The resulting representation can be used to solve node-level, edge-level [6] and graph-level prediction tasks [5].…”