Dataset Information

Understanding Graph Isomorphism Network for rs-fMRI Functional Connectivity Analysis.

ABSTRACT: Graph neural networks (GNN) rely on graph operations that include neural network training for various graph related tasks. Recently, several attempts have been made to apply the GNNs to functional magnetic resonance image (fMRI) data. Despite recent progresses, a common limitation is its difficulty to explain the classification results in a neuroscientifically explainable way. Here, we develop a framework for analyzing the fMRI data using the Graph Isomorphism Network (GIN), which was recently proposed as a powerful GNN for graph classification. One of the important contributions of this paper is the observation that the GIN is a dual representation of convolutional neural network (CNN) in the graph space where the shift operation is defined using the adjacency matrix. This understanding enables us to exploit CNN-based saliency map techniques for the GNN, which we tailor to the proposed GIN with one-hot encoding, to visualize the important regions of the brain. We validate our proposed framework using large-scale resting-state fMRI (rs-fMRI) data for classifying the sex of the subject based on the graph structure of the brain. The experiment was consistent with our expectation such that the obtained saliency map show high correspondence with previous neuroimaging evidences related to sex differences.

SUBMITTER: Kim BH

PROVIDER: S-EPMC7344313 | biostudies-literature | 2020

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Understanding Graph Isomorphism Network for rs-fMRI Functional Connectivity Analysis.

Kim Byung-Hoon BH Ye Jong Chul JC

Frontiers in neuroscience 20200630

Graph neural networks (GNN) rely on graph operations that include neural network training for various graph related tasks. Recently, several attempts have been made to apply the GNNs to functional magnetic resonance image (fMRI) data. Despite recent progresses, a common limitation is its difficulty to explain the classification results in a neuroscientifically explainable way. Here, we develop a framework for analyzing the fMRI data using the Graph Isomorphism Network (GIN), which was recently p ...[more]

PMID: 32714130

Similar Datasets

Project description:BackgroundLocal network connectivity disruptions in Alzheimer's disease patients have been found using graph analysis in BOLD fMRI. Other studies using MEG and cortical thickness measures, however, show more global long distance connectivity changes, both in functional and structural imaging data. The form and role of functional connectivity changes thus remains ambiguous. The current study shows more conclusive data on connectivity changes in early AD using graph analysis on resting-state condition fMRI data.Methodology/principal findings18 mild AD patients and 21 healthy age-matched control subjects without memory complaints were investigated in resting-state condition with MRI at 1.5 Tesla. Functional coupling between brain regions was calculated on the basis of pair-wise synchronizations between regional time-series. Local (cluster coefficient) and global (path length) network measures were quantitatively defined. Compared to controls, the characteristic path length of AD functional networks is closer to the theoretical values of random networks, while no significant differences were found in cluster coefficient. The whole-brain average synchronization does not differ between Alzheimer and healthy control groups. Post-hoc analysis of the regional synchronization reveals increased AD synchronization involving the frontal cortices and generalized decreases located at the parietal and occipital regions. This effectively translates in a global reduction of functional long-distance links between frontal and caudal brain regions.Conclusions/significanceWe present evidence of AD-induced changes in global brain functional connectivity specifically affecting long-distance connectivity. This finding is highly relevant for it supports the anterior-posterior disconnection theory and its role in AD. Our results can be interpreted as reflecting the randomization of the brain functional networks in AD, further suggesting a loss of global information integration in disease.

Project description:Major depressive disorder (MDD) is a complex mood disorder characterized by persistent and overwhelming depression. Previous studies have identified abnormalities in large scale functional brain networks in MDD, yet most of them were based on static functional connectivity. In contrast, here we explored disrupted topological organization of dynamic functional network connectivity (dFNC) in MDD based on graph theory. One hundred and eighty-two MDD patients and 218 healthy controls were included in this study, all Chinese Han people. By applying group information guided independent component analysis (GIG-ICA) to resting-state functional magnetic resonance imaging (fMRI) data, the dFNCs of each subject were estimated using a sliding window method and k-means clustering. Network properties including global efficiency, local efficiency, node strength and harmonic centrality, were calculated for each subject. Five dynamic functional states were identified, three of which demonstrated significant group differences in their percentage of state occurrence. Interestingly, MDD patients spent much more time in a weakly-connected State 2, which includes regions previously associated with self-focused thinking, a representative feature of depression. In addition, the FNCs in MDD were connected differently in different states, especially among prefrontal, sensorimotor, and cerebellum networks. MDD patients exhibited significantly reduced harmonic centrality primarily involving parietal lobule, lingual gyrus and thalamus. Moreover, three dFNCs with disrupted node properties were commonly identified in different states, and also correlated with depressive symptom severity and cognitive performance. This study is the first attempt to investigate the dynamic functional abnormalities in MDD in a Chinese population using a relatively large sample size, which provides new evidence on aberrant time-varying brain activity and its network disruptions in MDD, which might underscore the impaired cognitive functions in this mental disorder.

Project description:The assessment of task-independent functional connectivity (FC) after a lesion causing hemianopia remains an uncovered topic and represents a crucial point to better understand the neural basis of blindsight (i.e. unconscious visually triggered behavior) and visual awareness. In this light, we evaluated functional connectivity (FC) in 10 hemianopic patients and 10 healthy controls in a resting state paradigm. The main aim of this study is twofold: first of all we focused on the description and assessment of density and intensity of functional connectivity and network topology with and without a lesion affecting the visual pathway, and then we extracted and statistically compared network metrics, focusing on functional segregation, integration and specialization. Moreover, a study of 3-cycle triangles with prominent connectivity was conducted to analyze functional segregation calculated as the area of each triangle created connecting three neighboring nodes. To achieve these purposes we applied a graph theory-based approach, starting from Pearson correlation coefficients extracted from pairs of regions of interest. In these analyses we focused on the FC extracted by the whole brain as well as by four resting state networks: The Visual (VN), Salience (SN), Attention (AN) and Default Mode Network (DMN), to assess brain functional reorganization following the injury. The results showed a general decrease in density and intensity of functional connections, that leads to a less compact structure characterized by decrease in functional integration, segregation and in the number of interconnected hubs in both the Visual Network and the whole brain, despite an increase in long-range inter-modules connections (occipito-frontal connections). Indeed, the VN was the most affected network, characterized by a decrease in intra- and inter-network connections and by a less compact topology, with less interconnected nodes. Surprisingly, we observed a higher functional integration in the DMN and in the AN regardless of the lesion extent, that may indicate a functional reorganization of the brain following the injury, trying to compensate for the general reduced connectivity. Finally we observed an increase in functional specialization (lower between-network connectivity) and in inter-networks functional segregation, which is reflected in a less compact network topology, highly organized in functional clusters. These descriptive findings provide new insight on the spontaneous brain activity in hemianopic patients by showing an alteration in the intrinsic architecture of a large-scale brain system that goes beyond the impairment of a single RSN.

Dataset Information

Understanding Graph Isomorphism Network for rs-fMRI Functional Connectivity Analysis.

Publications

Understanding Graph Isomorphism Network for rs-fMRI Functional Connectivity Analysis.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets