Functional magnetic resonance imaging research in China

Abstract Functional magnetic resonance imaging (fMRI) non‐invasively measures the activity of the human brain and provides a unique technological tool for investigating aspects of the human brain including cognition, development, and disorders. As one of the main funding agencies for basic research in China, the National Natural Scientific Foundation of China (NSFC) has initiated various research programs during the last two decades that are related to fMRI research. In this review, we collected and analyzed the metadata of the projects and published studies in research fields using fMRI that were funded by the NSFC. We observed a trend of increasing funding amounts from the NSFC for fMRI research, typically from the General Program and Key Program. Leading research institutes from economically developed municipalities and provinces received the most support and formed close collaboration relationships. Finally, we reviewed several representative achievements from research institutions in china, involving data analysis methods, brain connectomes, and computational platforms in addition to their applications in brain disorders.

2018 was not fully recorded, we restricted our analysis to projects in amounts), which is a primary funding type that supports recipient scientists in freely selecting their research themes ( Figure 1B).
The second largest project type was the Young Scientists funding scheme (30.8%, 416% projects, and 9.8% in amounts, 88.5 million RMB), which is a kind of project allowing young researchers to freely engage in the study of key scientific research. Notably, although the Key Program only accounted for 3.7% (50 projects) of the projects, it provided 13.8% of the funding (125.0 million RMB). This corresponds to its responsibility to support researchers in conducting indepth, systematic, and innovative research in the direction of sound research. Finally, the influential talent-oriented projects that aim to support adept scholars and research teams, including the Excellent Young Scientist Fund, National Science Fund for Distinguished Young Scholars and Science Fund for Creative Research Groups, supported a total of 41 projects (110.9 million RMB) related to fMRI research.

| Topic and disciplines of NSFC grants for fMRI research
To examine the key topic of the fMRI research funded by NSFC, we conducted a word frequency analysis of the title, abstract, and keyword of the project. The top 15 words with the highest frequency included cognition, brain network, behavior, multimodal, development, functional connectivity, resting-state, intervention, memory, emotion, gene, study, major depressive disorder, heritability, and treatment effect (Figure 2A,B). These widely distributed topics represent a broad scope of sponsorship by the NSFC funding schemes covering almost all subfields of fMRI research, from the basic fMRI analytical methodology to application research into brain mechanisms for F I G U R E 1 Funding from the NSFC in the fMRI field. (A) The numbers and amounts of fMRI-related projects founded by the NSFC from 2001 to 2017. (B) The numbers and amounts of different categories of programs in the NSFC cognition and behavior, normal development, and neuropsychiatric disorders. To better illustrate the NSFC support in specific research areas or topics, including data analysis methods, brain disorders, brain development, and neuroimaging techniques, we further analyzed these fMRI research projects' title, keyword, and abstract.
We found that 73 projects (78.0 million RMB) were related to computation and analysis methods, 932 projects (629.1 million RMB) were supported for brain disorders, and 212 projects (139.7 million RMB) were involved in brain development. However, only eight projects (6.81 million RMB) aimed at developing imaging theory and equipment. With support from the NSFC, fMRI research in China has developed rapidly. We conducted a publication search in the Web of Science Core Collection (http://apps.webof knowl edge.com/) using all of the grant numbers of the NSFC projects. A total of 3732 articles were published in peer-reviewed journals, with a significant linear yearly increase of β = 53.9 (articles published in 2020 were not included due to their incomplete records) ( Figure 2C). Over 20% of the articles were published in journals with an impact factor (IF) higher than 5, among which 76 articles (accounting for 2.0% of the total number) were published in top journals with an IF higher than 10 ( Figure 2D). Regarding the fund categories, the General Program, were the highest among all fund categories (Table S1). According to the discipline categories defined by the Web of Science, these articles covered a broad scope of disciplines, particularly in neuroscience, neuroimaging, radiology nuclear medicine, psychiatry, clinical neurology, and multidisciplinary sciences ( Figure 2E). Notably, in the To provide a better understanding of the interaction between these universities and institutes regarding fMRI research, we conducted a graph-theoretical analysis of the article co-occurrence network. Briefly, we first constructed a co-occurrence network from the obtained 3732 fMRI-related published papers, in which the nodes were defined as the research institutes, and the weights of edges were defined as the number of papers that had coauthors in the two research institutes. To simplify the analysis and focus on the top institutes, we restricted the following analysis within a subnetwork compromising institutions with more than 50 cooccurrences. We then calculated degree centrality for each node.
A higher degree centrality of a given node indicates a higher level of cooperation for that specific institute. Accordingly, several universities and institutions were identified as network hubs (above the mean plus 1.5 standard deviations in degree centrality) that  (Table S4). We further applied the Louvain community detection algorithm 3 to identify cooperation clusters among the institutions. We observed a modularity Q = 0.243 for this cooperation network, suggesting that relatively independent but interrelated cooperative communities existed in the network.
Specifically, as illustrated in Figure 3E

| Representative achievement of fMRI studies in China
Under the support of the NSFC, scholars from China have obtained many achievements in fMRI research fields that have received increasing international attention, particularly regarding data analysis methods of resting-state fMRI, brain connectome, and computational platforms in addition to their application in brain disorders. Here, we summarize several representative works. Aiming to characterize the spontaneous activity signal of resting-state fMRI data, Zang and colleagues from IA-CAS proposed the amplitude of low-frequency fluctuation (ALFF), which calculates the time series of a single voxel at a specific frequency. 4 They found significantly higher ALFF values in the medial frontoparietal cortices compared to the average level of the whole brain, where higher regional cerebral blood flow and local oxygen metabolism have been found in previous studies. This suggests that ALFF values may be closely related to the metabolism of spontaneous brain activity. Then, the research group noticed that some brain areas with obvious physiological noise, such as the ventricles and large blood vessels, also have high ALFF values. To suppress the low-frequency amplitude of the ventricle and improve the gray matter sensitivity of the measurement, they further proposed the fractional amplitude of low-frequency fluctuation (fALFF), which is calculated as the ratio of the low-frequency amplitude to the fullfrequency amplitude of the signal at each voxel. 5 Additionally, to estimate the synchronization of local signals, Zang et al. proposed a regional homogeneity metric, which calculates Kendall's coefficient of concordance between a given voxel and its neighboring voxels. 6 These three metrics have been widely recognized by domestic and international fMRI research teams; moreover, they have been widely used in scientific research as they have been cited over 3000 times. For example, researchers from the Shenzhen Institute of Neuroscience used ALFF to study the evolutional and developmental patterns of the left inferior frontal gyrus (IFG) in humans and macaques. They showed that the ALFF of the IFG subregions were significantly increased during evolution and were significantly higher in senior-aged groups. 7 In the last decade, research into large-scale functional connectomes derived from fMRI data has become a hotspot in the field of neuroscience. Several research groups in China have made a series of representative achievements in developing investigational methods and assessing the physical substrates of the functional connectome.
For example, the research group from BNU systematically evaluated the effects of several key factors, such as different brain parcellation and connection definitions, on the topological measurements of the functional brain networks derived from resting-state fMRI data; the results of this evaluation have guided the choice of strategies for functional brain network analysis. The researchers found that the topology of functional networks had significant differences when different atlases were used to define network nodes or different null models were adopted. 8,9 They also proposed several frameworks to characterize the dynamic and directed functional connectivity of networks, including the variable parameter regression model combined with the Kalman filtering method 10 and the convergent cross mapping approach. 11 The researchers further systematically evaluated the test-retest reliability (TRT) of resting-state functional brain networks using repeated measured scans. Their results showed that most functional hubs exhibited fair to good TRT reliability using intraclass correlation coefficients and suggested that a 6-minute scan duration was required to reliably detect these functional hubs. 12 The TRTs of network metrics were sensitive to the scanning orders and intervals, with fair to excellent long-term reliability (>6 weeks) but lower short-term reliability (~20 min). 13 Finally, the researchers also explored the physiological basis of these brain network properties by elaborating on the relationship between highly connected hub regions of large-scale functional brain networks and the underlying metabolic consumption patterns of the brain. Using multimodal BOLD fMRI and arterial spin labeling perfusion MRI data, the researchers showed a significant positive correlation between the nodal centralities of whole-brain functional networks and regional cerebral blood flow, especially in the long-range functional hubs of brain networks. This relationship appeared to be significantly strengthened with increasing working memory task load. 14 The research group from UESTC made a series of improvements to the estimation of the directed connections in functional brain networks. They developed a kernel canonical correlation based on multivariate nonlinear Granger causality to explore local directed network connectivity at different scales. 15,16 Using multivariate Granger causality analysis, they constructed a whole brain-directed influence brain network and found that the directed functional network followed a small-world configuration with several medial and lateral frontoparietal regions acting as network hubs; these hubs were either affected by or exerted an influence on other regions. 17 The confounding effect of a hemodynamic response function (HRF) and the conditioning of a large number of variables in the presence of short time series are two core issues for reconstructing directed functional brain networks using fMRI data. Aiming to overcome these issues, the researchers proposed a blind deconvolution approach to recover effective connectivity of brain networks from resting-state fMRI data, which considers resting-state fMRI data as "spontaneous event-related" to extract a region-specific HRF and use it in deconvolution. 18 The researchers further proposed a partially conditioned Granger causality to cope with the redundancy and dimensionality curse in evaluating effective connectivity from fMRI data. The researchers constructed a high-resolution directed functional brain network at the voxel level and depicted several voxelwise hubs of incoming and outgoing connections, which were mostly located in the default mode network (DMN). 18,19 Collaborated with researchers from CCMU, the research group from UESTC developed an approach to reliably identify homologous functional regions in each individual and showed that aligning data using the homologous functional regions derived from resting-state fMRI can significantly improve the study of resting-state functional connectivity, task fMRI activations, and brain-behavior associations. 20 Many of the proposed analysis methods have been embedded in several domestically toolboxes and platforms for neuroimaging processing, network computation, and visualization. Most of these toolboxes are published on neuroscience websites and are freely available to researchers using fMRI. For example, the graphtheoretical network analysis (GRETNA) toolbox enables the construction and topological analysis of networks based on MATLAB code with parallel acceleration (http://www.nitrc.org/proje cts/ gretna). 21 Based on statistical parametric mapping (SPM), the REST 22 and Data Processing Assistant for Resting-State fMRI (DPARSF) provides a pipeline for resting-state fMRI data analysis (http://www. nitrc.org/proje cts/dparsf). 23 BrainNet Viewer provides flexible functions for the visualization of measurements derived from neuroimages and topological architectures of brain networks (http:// www.nitrc.org/proje cts/bnv/). 24 These toolboxes have been widely used by more than 1000 laboratories worldwide, and they have been used in over 5000 published articles.
The application of fMRI analysis becomes increasingly important when either studying pathophysiology or exploring neuroimaging biomarkers associated with neuropsychiatric disorders. The research group from West China Hospital of SCU applied fMRI techniques to improve the understanding of psychiatric illnesses and treatment effects. The researchers proposed a new discipline in medical sciences, psychoradiology, which uses radiological technologies to unveil patterns of brain abnormalities in patients with psychiatric disorders. 25 They conducted neuroimaging studies with a considerable number of participants to identify alterations in brain features in patients with psychiatric disorders underlying the disruption of emotion and behavior. Using resting-state functional connectivity analysis, they revealed different disrupted functional circuits in patients with non-refractory and refractory depression. Refractory depression is associated with disrupted functional connectivity mainly in thalamocortical circuits, whereas non-refractory depression is associated with more distributed decreased connectivity in the limbic-striatalpallidal-thalamic circuit. 26 Cooperating with a research group from BNU, the research group from West China Hospital of SCU conducted the first whole-brain functional network analysis in drugnative, first-episode major depressive disorder patients. The patient group showed a shorter path length and a higher global efficiency, implying a shift toward randomization in their brain networks.
Furthermore, altered nodal centralities in depressed patients were predominately located in the DMN regions, primary sensory cortex, and caudate nucleus. 27 The investigators from HNU and IP-CAS collaborated with 17 sites over China and studied the functional connectivity within the DMN in patients with depression. They showed that the reduction of functional connectivity only in recurrent MDD, but not in first-episode drug-naïve MDD. The altered functional connectivity was associated with medication usage but not with disease duration. 28 In schizophrenia, the research group from West China Hospital of SCU found significant correlations between symptom scores for thought disturbance and temporo-putamen connectivity and between negative symptoms and temporo-precuneus connectivity. 29 Using ALFF values to characterize regional cerebral function, they found that antipsychotic therapy can increase regional spontaneous activity in widespread brain areas, providing further understanding of antipsychotic drugs at a complex system level. 30 Deep brain stimulation (DBS) is a powerful therapy strategy for movement disorders; however, its neural mechanisms remain unclear. Using simultaneous DBS-fMRI techniques, Researchers from Peking University revealed robust BOLD responses in the basal ganglia-thalamocortical network in a frequency-dependent manner in Parkinsonian rats. 31 The research group from Tsinghua University investigated the whole-brain functional effects of subthalamic nucleus stimulation in patients with Parkinson's disease. They showed that a circuit involving the globus pallidus internus, thalamus, and deep cerebellar nuclei was significantly activated, whereas another circuit involving the primary motor cortex, putamen, and cerebellum showed deactivation. These two distinct neurocircuits were related to divergent symptoms, providing a novel understanding of DBS's neural mechanisms in Parkinson's disease. 32 Developmental disorders can also be companied with brain alterations in adults. A study of the collaborations between BNU and PKU examined the alterations of cerebral blood flow and resting-state functional connectivity in adults with attention-deficit/hyperactivity disorder. They showed significantly decreased cerebral blood flow in the large-scale resting-state networks, including the ventral attention, somatomotor, and limbic networks in patients compared with the controls. 33 Regarding physiological pain, using task fMRI, researchers from Peking University found that the pain-induced activation in the dorsal anterior cingulate cortex and bilateral insula was significantly reduced after individuals performed altruistic actions, suggesting that incurring personal costs to help others can relieve painful feelings in human performers. 34

| CON CLUS ION
In general, both the number of projects and amount of funding invested in research using fMRI from the NSFC have increased substantially in the 21st century. These projects have promoted the rapid development of fMRI research in China and have fueled several representative achievements in fMRI computation methods and their applications in brain disorders. More importantly, several fMRI research teams with specialized expertise have been developed in China, and a highly cooperative environment has been created.
However, domestic research into neuroimaging techniques is still not enough, especially in developing imaging theory and equipment.
Most of the MRI scanners reported in the NSFC-funded projects are imported instruments from abroad. Nevertheless, the domestic manufacturers such as United Imaging have made great progress in MR scanners and could gain a place in fMRI research in future. It is worth noting that the clinical translational application of fMRI research needs to make further efforts. The China Brain Project is about to launch; this project covers basic research on neural mechanisms underlying cognition, brain development, translational research for the diagnosis and intervention of brain diseases, and brain-inspired intelligence technology. As a technological tool for basic research, fMRI will play an important role in many research topics. It is foreseeable that future research trends of fMRI will rely more on multicenter imaging big data with multidimensional biological variables (eg, imaging, genetic, demographic, cognitive, and clinical measures) and focus more on individual differences and individual uniqueness. A large collaborative project integrating multicenter fMRI datasets from ten domestic research institutes and clinical hospitals has revealed reproducible disruptions in the functional connectome in patients with major depressive disorder, 35,36 offering a promising paradigm for the understanding of the pathologies and exploration of biomarkers for clinical diagnosis and treatment of psychiatric disorders. Combining brain stimulation techniques and fMRI can offer a powerful tool for elucidating brain functions. For example, modulating the activity of the right temporoparietal junction can lead to changes in the social framing effect but not in nonsocial conditions. 37 Thus, future studies with simultaneous brain stimulation and neuroimaging could provide causal interpretations for the large-scale functional brain activity and connectivity observed by fMRI. Moreover, by searching keywords related to machine learning in the papers funded by NSFC fMRI projects, we observed a considerable increase from 2008 to 2019, particularly after 2017 ( Figure S1). This increase indicates that future fMRI research will integrate machine learning methods more to reveal the neuronal features that underlie individual behaviors and brain disorders and identify imaging biomarkers for predicting cognitive performance, normal development, disease diagnosis, and prognosis. These are also hot topics for fMRI research that the NSFC needs to pay attention in future.

ACK N OWLED G M ENT
We would like to thank the native English-speaking scientists of BioMed Proofreading Company for editing our manuscript.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest.

E TH I C A L A PPROVA L
This is a meta-analysis. Our hospital Ethics Committee has confirmed that no ethical approval is required.

CO N S E NT FO R PU B LI C ATI O N
Publication is approved by all authors and tacitly or explicitly by the responsible authorities where the work was carried out.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions