Graph neural networks for fMRI functional brain networks: A survey.

Brain network analysis is one of the most effective methods for brain disease diagnosis. Existing studies have shown that exploring information from multimodal data is a valuable way to improve the effectiveness of brain network analysis. In recent years, deep learning has received more and more attention due to its powerful feature learning capabilities, and it is natural to introduce this tool into multi-modal brain networks analysis. However, it would face two challenges. One is that brain network is in non-Euclidean domain, so the convolution kernel in deep learning cannot be directly used to brain networks. The other is that most of existing multi-modal brain network analysis methods cannot extract full use of complementary information from distinct modalities. In this paper, we propose a multi-modal non-Euclidean brain network analysis method based on community detection and convolutional autoencoder, which can solve the above two problems simultaneously in one framework (M2CDCA). First, we construct the functional and structural brain network, respectively. Second, we design a multi-modal interactive community detection method that exploits structural modality to guide functional modality for detecting community structure, then readjusts the nodes distribution so that the adjusted brain network can preserve the potential community information and is more suitable for convolution kernels. Finally, we design a dual-channel autoencoder model with self-attention mechanism to capture hierarchical and highly non-linear features, then comprehensively use two modalities information for classification. We evaluate our method on an epilepsy dataset, the experimental results show that our method outperform several state-of-the-art methods.

Functional brain network (FBN) analysis aims to enhance the understanding of brain organization and support the diagnosis of neurological and psychiatric disorders. Prior studies have shown that FBNs exhibit small-world topology, where brain regions form functional clusters, and abnormalities in these clusters are strongly associated with disease. However, current learning-based methods either ignore this special topological structure or impose it as a post-hoc step outside the learning process, limiting both performance and interpretability. In this paper, we propose Learning Optimal Spectral Clustering (LOSC), a new framework that integrates the FBN generation, clustering, and classification with a novel graph theory grounded loss to fully exploit the small-world topology. Firstly, LOSC learns brain connectivity in a nonlinear spatio-spectral embedding space, guided by our proposed Rayleigh Quotient Loss (RQL), to preserve the small-world properties in generated FBNs. Then, the FBNs are partitioned into clusters of functionally synchronized regions, and both intra- and inter-cluster relations are utilized for brain network classification. Our contributions are threefold: (1) Improved brain network classification accuracy: by leveraging small-world functional clusters, LOSC achieves consistent gains of 2.0%, 3.6%, and 2.6% on the ABIDE, ADHD-200, and HCP datasets compared with state-of-the-art models, respectively; (2) Theoretical grounding: with our proposed RQL, LOSC bridges the gap between the graph theory and learning-based FBN analysis; and (3) Interpretability: the discovered functional clusters align with known neuropathology and contribute to the discovery of new functional community biomarkers.

Fundamentals of Brain Network Analysis

Schizophrenia (ScZ) is a chronic mental disorder affecting the function of the brain, which causes emotional, social, and cognitive problems. This paper explored the functional brain network and deep learning methods to detect ScZ using electroencephalogram (EEG) signals. Functional brain network analysis was proposed and implemented using a multivariate autoregressive model and coherence connectivity algorithm. The three machine learning techniques and 3D-convolutional neural network (CNN) models were applied to classify the ScZ patients and health control subjects, and then the public LMSU database was utilized to assess the performance. The proposed 3D-CNN method achieved the performance of a 98.47 ± 1.47 % in accuracy, 99.26 ± 1.07 % in sensitivity, and 97.23 ± 3.76 % in specificity. Moreover, in addition to the default mode network region, the temporal and posterior temporal lobes of both right and left hemispheres were found as the significant difference areas in ScZ brain network analysis.

Brain network analysis can help reveal the pathological basis of neurological disorders and facilitate automated diagnosis of brain diseases, by exploring connectivity patterns in the human brain. Effectively representing the brain network has always been the fundamental task of computeraided brain network analysis. Previous studies typically utilize human-engineered features to represent brain connectivity networks, but these features may not be well coordinated with subsequent classifiers. Besides, brain networks are often equipped with multiple hubs (i.e., nodes occupying a central position in the overall organization of a network), providing essential clues to describe connectivity patterns. However, existing studies often fail to explore such hubs from brain connectivity networks. To address these two issues, we propose a Connectivity Network analysis method with discriminative Hub Detection (CNHD) for brain disease diagnosis using functional magnetic resonance imaging (fMRI) data. Specifically, we incorporate both feature extraction of brain networks and network-based classification into a unified model, while discriminative hubs can be automatically identified from data via ℓ1-norm and ℓ2,1-norm regularizers. The proposed CNHD method is evaluated on three real-world schizophrenia datasets with fMRI scans. Experimental results demonstrate that our method not only outperforms several state-of-the-art approaches in disease diagnosis, but also is effective in automatically identifying disease-related network hubs in the human brain.

Brain network analysis has been proved to be one of the most effective methods in brain disease diagnosis. In order to construct discriminative brain networks and improve the performance of disease diagnosis, many machine learning–based methods have been proposed. Recent studies show that combining functional and structural brain networks is more effective than using only single modality data. However, in the most of existing multi-modal brain network analysis methods, it is a common strategy that constructs functional and structural network separately, which is difficult to embed complementary information of different modalities of brain network. To address this issue, we propose a unified brain network construction algorithm, which jointly learns both functional and structural data and effectively face the connectivity and node features for improving classification. First, we conduct space alignment and brain network construction under a unified framework, and then build the correlation model among all brain regions with functional data by low-rank representation so that the global brain region correlation can be captured. Simultaneously, the local manifold with structural data is embedded into this model to preserve the local structural information. Second, the PageRank algorithm is adaptively used to evaluate the significance of different brain regions, in which the interaction of multiple brain regions is considered. Finally, a multi-kernel strategy is utilized to solve the data heterogeneity problem and merge the connectivity as well as node information for classification. We apply the proposed method to the diagnosis of epilepsy, and the experimental results show that our method can achieve a promising performance.

Network science has become an invaluable tool for neuroimaging analysis and has advanced our understanding of the brain's complex structural and functional topology. Network science, an interdisciplinary offshoot of graph theory, conceives of the brain as a system of nodes, representing brain regions or voxels, and edges, representing the structural or functional connection between these regions or voxels. Such a model is considered appealing in neuroscience as it describes a system with various interacting regions that produces complex behaviors. This conception of the brain has revealed that the brain exhibits “small-world” properties, namely that brain networks are systems with regional specialization and efficient global information transfer. More importantly, studies in brain networks have furthered our understanding of diseases and disorders affecting the brain. Assessing group differences in representative samples of brain networks has become increasingly important, yet appropriate methodological developments have remained relatively sparse. Additionally, there is increased interest in understanding dynamic changes, or how a network changes over time. Currently, most network analysis tools are designed to analyze a static representation, thus changes that occur over time are often overlooked or ignored. As the field continues to grow, it is becoming increasingly important to develop methods for understanding the complexity of these data. This research topic features selected contributions that explores group analysis and dynamics in brain network studies including methodological and practical applications employing various imaging modalities. This snapshot of the field represents the growing interest to expand and validate new methods for use in the field of brain network neuroscience. “Defining nodes in complex brain networks” is a review article discussing the methodological choices for defining nodes, particularly how the choice of atlas-based or voxel-based methods affects the computed network (Stanley et al., 2013). “Routing in the brain” is a commentary on network routing focusing on how current methods and models may not be adequate for brain networks. Further exploration of the packet switching network model may be more realistic and appropriate for the brain (Graham, 2014). “A multimodal approach for determining brain networks by jointly modeling functional and structural connectivity” is a methods article introducing a Bayesian model for estimating functional networks, from functional magnetic resonance imaging (fMRI) data, leveraging complementary structural diffusion tensor imaging (DTI) data (Xue et al., 2015). “A permutation testing framework to compare groups of brain networks” is a methods article describing a permutation testing framework to statistically compare groups of functional brain networks while incorporating topological features inherent in each individual network (Simpson et al., 2013). “Assessing dynamics, spatial scale, and uncertainty in task-related brain network analyses” is a methods article describing a statistical approach for computing uncertainty in static and dynamic functional networks and aggregating network measures in task-related electrocorticography (ECoG) data (Stephen et al., 2014). “Hierarchical vector auto-regressive models and their applications to multi-subject effective connectivity” is a methods article introducing a generalization of the vector auto-regressive (VAR) model for comparing effective connectivity between experimental conditions while accounting for between-subject heterogeneity (Gorrostieta et al., 2013). “Statistical network analysis for functional MRI: summary networks and group comparisons” is a review article discussing the construction of summary networks and how to test for topological differences in groups of networks when these groups also exhibit significant differences in density (Ginestet et al., 2014). “Detecting functional connectivity change points for single-subject fMRI data” is a methods article that extends Dynamic Connectivity Regression (DCR), a data-driven technique which detects change points in functional connectivity between brain regions to determine state changes over the course of an experimental task, by introducing a novel algorithm aimed at increasing the estimation of networks for individual subject data (Cribben et al., 2013). “The Laplacian spectrum of neural networks” is an original research article examining the use of the Laplacian spectrum in anatomic networks of the macaque, cat and Caenorhabditis elegans. This method describes network structure at a systems level, assessing systemic infrastructural properties of the entire network as opposed to properties of specific nodes or connections (De Lange et al., 2014). “Quantifying network properties in multi-electrode recordings: spatiotemporal characterization and inter-trial variation of evoked gamma oscillations in mouse somatosensory cortex in vitro” is an original research article on network dynamics in multi-electrode array data focusing on changes in the functional connectivity of evoked gamma oscillations in cortical circuits (Carmeli et al., 2013).

Automatic diagnosis of brain diseases based on brain connectivity network (BCN) classification is one of the hot research fields in medical image analysis. The functional brain network reflects the brain functional activities and structural brain network reflects the neural connections of the main brain regions. It is of great significance to explore and explain the inner mechanism of the brain and to understand and treat brain diseases. In this paper, based on the graph structure characteristics of brain network, the fusion model of functional brain network and structural brain network is designed to classify the diagnosis of brain mental diseases. Specifically, the main work of this paper is to use the Laplacian graph embed the information of diffusion tensor imaging, which contains the characteristics of structural brain networks, into the functional brain network with hyper-order functional connectivity information built based on functional magnetic resonance data using the sparse representation method, to obtain brain network with both functional and structural characteristics. Projection of the brain network and the two original modes data to the kernel space respectively and then classified by the multi-task learning method. Experiments on the epilepsy dataset show that our method has better performance than several state-of-the-art methods. In addition, brain regions and connections that are highly correlated with disease revealed by our method are discussed.

The fusion of structural and functional brain network analysis has been widely applied in the analysis of various brain diseases, especially in identifying the progression stages of Alzheimer’s disease. However, most existing multimodal brain network analysis methods separately construct functional and structural networks, making it difficult to incorporate complementary information from several modalities. To tackle this difficulty, we propose AMFusion(Adaptive Multimodal Fusion), a unified brain network construction framework that jointly learns from functional and structural images, thereby efficiently resolving connectivity and node feature learning problems. Our approach begins with a designed quantization encoder to extract structural features from DTI, while dynamic functional connectivity (FC) is constructed from fMRI. Subsequently, a multilevel brain network fusion module is employed to learn and integrate brain connections, considering interactions across different temporal and spatial scales. Finally, a classifier guides further optimization of the brain network, facilitating multi-class disease classification. The performance of the proposed AMFusion was validated using the authentic dataset from ADNI, and experimental results demonstrated the effectiveness of AMFusion. By considering both connection patterns and node features, our method overcomes the limitations of existing approaches, providing a more comprehensive and robust framework for analyzing and diagnosing AD.

Patients with traumatic brain injury show clear impairments in behavioural flexibility and inhibition that often persist beyond the time of injury, affecting independent living and psychosocial functioning. Functional magnetic resonance imaging studies have shown that patients with traumatic brain injury typically show increased and more broadly dispersed frontal and parietal activity during performance of cognitive control tasks. We constructed binary and weighted functional networks and calculated their topological properties using a graph theoretical approach. Twenty-three adults with traumatic brain injury and 26 age-matched controls were instructed to switch between coordination modes while making spatially and temporally coupled circular motions with joysticks during event-related functional magnetic resonance imaging. Results demonstrated that switching performance was significantly lower in patients with traumatic brain injury compared with control subjects. Furthermore, although brain networks of both groups exhibited economical small-world topology, altered functional connectivity was demonstrated in patients with traumatic brain injury. In particular, compared with controls, patients with traumatic brain injury showed increased connectivity degree and strength, and higher values of local efficiency, suggesting adaptive mechanisms in this group. Finally, the degree of increased connectivity was significantly correlated with poorer switching task performance and more severe brain injury. We conclude that analysing the functional brain network connectivity provides new insights into understanding cognitive control changes following brain injury.

In recent years, the field of network science has enabled researchers to represent the highly complex interactions in the brain in an approachable yet quantitative manner. One exciting finding since the advent of brain network research was that the brain network can withstand extensive damage, even to highly connected regions. However, these highly connected nodes may not be the most critical regions of the brain network, and it is unclear how the network dynamics are impacted by removal of these key nodes. This work seeks to further investigate the resilience of the human functional brain network. Network attack experiments were conducted on voxel-wise functional brain networks and region-of-interest (ROI) networks of 5 healthy volunteers. Networks were attacked at key nodes using several criteria for assessing node importance, and the impact on network structure and dynamics was evaluated. The findings presented here echo previous findings that the functional human brain network is highly resilient to targeted attacks, both in terms of network structure and dynamics.

BackgroundNeural stem cells (NSCs) are believed to have the most therapeutic potential for neurological disorders because they can differentiate into various neurons and glial cells. This research evaluated the safety and efficacy of intranasal administration of NSCs in children with cerebral palsy (CP). The functional brain network (FBN) analysis based on electroencephalogram (EEG) and voxel-based morphometry (VBM) analysis based on T1-weighted images were performed to evaluate functional and structural changes in the brain.MethodsA total of 25 CP patients aged 3–12 years were randomly assigned to the treatment group (n = 15), which received an intranasal infusion of NSCs loaded with nasal patches and rehabilitation therapy, or the control group (n = 10) received rehabilitation therapy only. The primary endpoints were the safety (assessed by the incidence of adverse events (AEs), laboratory and imaging examinations) and the changes in the Gross Motor Function Measure-88 (GMFM-88), the Activities of Daily Living (ADL) scale, the Sleep Disturbance Scale for Children (SDSC), and some adapted scales. The secondary endpoints were the FBN and VBM analysis.ResultsThere were only four AEs happened during the 24-month follow-up period. There was no significant difference in the laboratory examinations before and after treatment, and the magnetic resonance imaging showed no abnormal nasal and intracranial masses. Compared to the control group, patients in the treatment group showed apparent improvements in GMFM-88 and ADL 24 months after treatment. Compared with the baseline, the scale scores of the Fine Motor Function, Sociability, Life Adaptability, Expressive Ability, GMFM-88, and ADL increased significantly in the treatment group 24 months after treatment, while the SDSC score decreased considerably. Compared with baseline, the FBN analysis showed a substantial decrease in brain network energy, and the VBM analysis showed a significant increase in gray matter volume in the treatment group after NSCs treatment.ConclusionsOur results showed that intranasal administration of NSCs was well-tolerated and potentially beneficial in children with CP.Trial registration: The study was registered in ClinicalTrials.gov (NCT03005249, registered 29 December 2016, https://www.clinicaltrials.gov/ct2/show/NCT03005249) and the Medical Research Registration Information System (CMR-20161129-1003).

In recent years, driven by scientific and clinical concerns, there has been an increased interest in the analysis of functional brain networks. The goal of these analyses is to better understand how brain regions interact, how this depends upon experimental conditions and behavioral measures and how anomalies (disease) can be recognized. In this paper, we provide, first, a brief review of some of the main existing methods of functional brain network analysis. But rather than compare them, as a traditional review would do, instead, we draw attention to their significant limitations and blind spots. Then, second, relevant experts, sketch a number of emerging methods, which can break through these limitations. In particular we discuss five such methods. The first two, stochastic block models and exponential random graph models, provide an inferential basis for network analysis lacking in the exploratory graph analysis methods. The other three addresses: network comparison via persistent homology, time-varying connectivity that distinguishes sample fluctuations from neural fluctuations, and network system identification that draws inferential strength from temporal autocorrelation.

For decades, task-based functional magnetic resonance imaging (tfMRI) has been a powerful noninvasive tool to explore the organizational architecture of human brain function. Researchers have developed a variety of brain network analysis methods for tfMRI data, including the general linear model (GLM), independent component analysis (ICA) and sparse representation methods. However, these shallow models are limited in faithful reconstruction and modeling of the hierarchical and temporal structures of brain networks, as demonstrated in more and more studies. Recently, recurrent neural networks (RNNs) exhibit great ability of modeling hierarchical and temporal dependency features in the machine learning field, which might be suitable for tfMRI data modeling. To explore such possible advantages of RNNs for tfMRI data, we propose a novel framework of deep recurrent neural network (DRNN) to model the functional brain networks for tfMRI data. Experimental results on the motor task tfMRI data of Human Connectome Project 900 subjects data release demonstrated that the proposed DRNN can not only faithfully reconstruct functional brain networks, but also identify more meaningful brain networks with multiple time scales which are overlooked by traditional shallow models. In general, this work provides an effective and powerful approach to identifying functional brain networks of multiple time scales from tfMRI data.

This paper studies the functional connection of brain regions by using network analysis method. Through the analysis of network attributes, the influence of different emotions on brain functional network connection and the construction method of brain functional network are learned. In this paper, we used Electroencephalogram (EEG) signals to construct the binary brain networks based on correlation and phase synchrony. Firstly, the global network attributes under different emotions were studied, and the results obtained by different network construction methods were compared. We find that the brain functional networks with different definitions share common characteristics in the global attribute change. Secondly, the node attributes were analyzed. The results show that the node attributes of brain functional network, such as degree, betweenness centrality and local efficiency, can reflect the change of emotion. This paper provides clues for the study of emotional brain network analysis.