Functional connectivity of the default mode, dorsal attention and fronto-parietal executive control networks in glial tumor patients

Abstract

Purpose

Resting state functional magnetic resonance imaging (rsfMRI) is an emerging tool to explore the functional connectivity of different brain regions. We aimed to assess the disruption of functional connectivity of the Default Mode Network (DMN), Dorsal Attention Network(DAN) and Fronto-Parietal Network (FPN) in patients with glial tumors.

Methods

rsfMRI data acquired on 3T-MR of treatment-naive glioma patients prospectively recruited (2015–2019) and matched controls from the 1000 functional-connectomes-project were analyzed using the CONN functional toolbox. Seed-Based Connectivity Analysis (SBCA) and Independent Component Analysis (ICA, with 10 to 100 components) were performed to study reliably the three networks of interest.

Results

35 patients with gliomas (17 WHO grade I-II, 18 grade III-IV) and 70 controls were included. Global increased DMN connectivity was consistently found with SBCA and ICA in patients compared to controls (Cluster1: Precuneus, height: p < 10⁻⁶; Cluster2: subcallosum; height: p < 10⁻⁵). However, an area of decreased connectivity was found in the posterior corpus callosum, particularly in high-grade gliomas (height: p < 10⁻⁵). The DAN demonstrated small areas of increased connectivity in frontal and occipital regions (height: p < 10⁻⁶). For the FPN, increased connectivity was noted in the precuneus, posterior cingulate gyrus, and frontal cortex. No difference in the connectivity of the networks of interest was demonstrated between low- and high-grade gliomas, as well as when stratified by their IDH1-R132H (isocitrate dehydrogenase) mutation status.

Conclusion

Altered functional connectivity is reliably found with SBCA and ICA in the DMN, DAN, and FPN in glioma patients, possibly explained by decreased connectivity between the cerebral hemispheres across the corpus callosum due to disruption of the connections.

Introduction

Gliomas are one of the most common primary brain tumors and are associated with significant morbidity and mortality [1, 2]. The addition of molecular testing for these tumors such as isocitrate dehydrogenase (IDH) mutation status and MGMT methylation have furthered our understanding of high-grade gliomas. In addition, the established treatment paradigms of surgical resection, radiation and chemotherapy are advanced by the availability of more precise surgical techniques and the prospect of novel targeted agents and immunotherapy approach [3, 4]. Neuroimaging for brain tumors has also evolved. Currently, radiologists evaluate not only morphological aspects of the tumor as well as their relation to the main areas and networks corresponding to functional hubs of the brain [5].

The exploration of brain activation via Blood-Oxygen-Level-Dependent (BOLD) sequences has revolutionized our understanding of brain functioning [6, 7]. While task-based functional MRI, now an integral part of presurgical mapping protocols, detects the activation of brain regions using motor or language tasks, resting-state functional magnetic resonance imaging (rsfMRI) was recently developed to study functional networks at rest [8]. Compared to task-based fMRI, rsfMRI can evaluate the preoperational functional integrity of intrinsically segregated regions without asking patients to perform any specific task and therefore has potential to increase the translation of fMRI into the clinical routine and care.

Literature in this field has grown at an exponential rate during the past two decades [9]. The Default Mode Network (DMN) is a task-negative network, meaning it is the most prominent network activated during rest and internally-driven processes such as mind wandering or future planning [10, 11]. Impaired DMN is involved in various conditions, including major depressive disorder [12], autism spectrum disorder [13], stroke [14] and brain tumors [15]. Two of the main attention networks, the Dorsal Attention Network (DAN) and the Fronto-Parietal Network (FPN), are also involved in multiple diseases, including glial tumors. These two networks are extensively interconnected with the DMN to regulate goal-directed cognition: the DAN is involved in orienting one’s focus to a particular task while the FPN is a control network linked to complex problem solving and working memory [16, 17]. Several studies already reported altered functional connectivity in patients with brain tumors but the methodology used across the articles and the results have been heterogeneous [18]. The understanding of resting state networks alteration by the tumor could lead to improved knowledge of brain functioning in patients with glial lesions, which could be related to neuron-glioma interaction.

A specific concern about the results of rsfMRI is their validity with for example two rsfMRI studies highlighting aberrant findings [19, 20]. The analytic technique performed also significantly impacts the reproducibility of results across studies [21]. Most of the previous studies used only one technique to assess the disruption of functional networks, often either seed-based correlation-analysis (SBCA) [22] or independent-component-analysis (ICA) [23]. One limitation of using SBCA is that there is no consensus regarding the ideal number of Independent Components (ICs) to use given the anatomical distortions due to brain masses.

The aim of this study was to assess the potential disruption of functional connectivity of the DMN, DAN, and FPN in treatment-naïve patients with glial tumors comparing the data with non-glioma controls, and to correlate these disruptions with histological grade and IDH1-R132H-mutation status, using both SBCA and ICA.

Methods

Patients

From March 2015 to January 2019, 48 patients with histologically confirmed glial lesions (including gliomas and glioneural tumors) were prospectively recruited. Imaging obtained prior to any surgery or other therapy was evaluated (patients with recurrences, prior brain lesions, or other neurologic disorders as well as excessive head motion during resting state BOLD acquisition were excluded).

Age- and sex-matched healthy controls were included, at a rate of 2 controls-per-subject, using data from the 1000 functional connectomes project [24].

Characteristics from patients including sex, age, personal and family medical history were collected. Tumor dimension was measured by a radiologist (MT) and a fellowship-trained neuro-radiologist with 18 years of experience (RJ).

This study was approved by the Institutional Review Board (IRB:i1700006) and informed consent was obtained from included subjects.

Data acquisition

Images were acquired on a 3T-Siemens Skyra MR-system. The protocol included T1 pre- and post-gadolinium(TR = 2400 ms, TE = 2.24 ms) and T2(TR = 3200 ms, TE = 563 ms) sequences. Resting functional imaging data was acquired using a two-dimensional echo-planar imaging sequence with the following imaging parameters: TR = 2.52 s, TE = 29 ms, flip angle = 90°, voxel size = 3 × 3 × 3.3 mm³, 40 slices and 200 time frames for a total imaging times of approximately 8.4 min (eyes closed). Controls included had similar acquisition parameters [TR = 2 or 2.5 s, number of time points = 123–265]. The differences of acquisition parameters (including varying TR) were taken into account during the processing of the data.

Image analysis

Conn functional connectivity toolbox for SPM: version 19(www.nitrc.org/projects/conn, RRID:SCR_009550) based on MATLAB 2019 was used to perform functional analysis [25]. Results were reproduced by two independent operators (one radiologist [MT] and one post-doctoral fellow [PKG]).

Pre-processing of the data used the default pre-processing pipeline of Conn, including functional realignment and unwarping, slice-timing correction, detection of the outliers, functional and structural normalization in the MNI-space (normalization of the co-registered T1 image and EPI volumes with a voxel size of 2 × 2 × 2 mm) and segmentation of the data (segmentation of grey-matter, whiter-matter and cerebrospinal fluid), functional smoothing (at 8 mm full width at half-maximum). Denoising was processed using the usual covariates (white-matter and cerebrospinal fluid signals) and band-pass filtering with the default Conn values (0.008–0.09 Hz). A component-based noise reduction (Com-pCor) is used in Conn, avoiding regression of the global signal. This method performs principal component analysis to estimate the physiological noise from white matter and cerebrospinal fluid for each participant and corrects for biases related to non-neural sources (such as respiration or cardiac activity). Pre-processed data were visually checked to control the quality of the segmentation and normalization in brains with tumors.

ICA was performed using Calhoun’s group-level ICA approach [26] to decompose the data into spatial independent components (IC), meaning networks, and into temporal components, meaning shared patterns within networks. First, a three-step principal component analysis (PCA) decomposed the data set into Independent Components (ICs), estimated using the Infomax algorithm. Then, we used the group ICA3 (GICA3) to back-reconstruct the individual ICs. ICs were recognized as functional networks based on visual examination and spatial match to template (spatial correlation and spatial overlap of suprathreshold areas/Dice coefficient). ICA was tested with different numbers of ICs from 10 to 100 IC (every 10 ICs: 10, 20, 30etc.) in order to check the reliability of the results with different numbers of ICs.

SBCA was also performed. For each seed, functional connectivity data was averaged by the software across all subjects in each group. Four seeds were used for the DMN, from the predefined network atlases implemented in the CONN toolbox: one in the medial prefrontal cortex (MPFC; coordinates (1;55;−3)), one in the posterior cingulate cortex (PCC; (1;−61;38)) and one in each lateral parietal cortex (left and right; (−39;−77;33) and (47;−67;29)). For the DAN, the seeds were situated in the frontal eye fields (FEF; (−27;−9;64) and (30;−6;64)) and right and left intraparietal sulci (IPS; (−39;−43;52) and (39;−42;54)). Seeds for the FPN were placed in the bilateral lateral prefrontal cortex (LPFC; (−43;33;28) and (41;38;30)) and the posterior parietal cortex (PPC; (−46;−58;49) and (52;−52;45)). A second SBCA was also performed: first level analysis maps were visually inspected for each patient. In those where the network of interest was not recognized as typical by trained evaluators, patients were excluded from this second analysis to take into account a potential bias induced by the tumor involving (or displacing) the seed or anatomical variants in patients or controls. Thus, only brain tumor patients and controls with typical appearance of DMN, DAN or FPN were included.

Pathological examination

Histological diagnosis of gliomas and IDH-mutation status were collected. High grades tumors were defined as tumors WHO-grades III and IV and low-grade tumors as WHO-grades I and II.

Statistical analysis

For all the analyses (ICA and SBCA), two-sample t-tests were performed. The voxel level threshold was for p < 0.001 (uncorrected height threshold) and among the resulting suprathreshold clusters, only those with cluster-extent FDR-corrected p-values below a p < 0.05 threshold were reported as significant.

Subgroup analyses were performed in patients according to WHO tumor grades (categorized as low- versus high-grade gliomas, low or high-grade gliomas versus controls), locations as well as by tumor IDH1-R132H status (IDH mutant or IDH wild type) and tumors’ size (>3 cm or <3 cm).

Results

Subject characteristics

Forty-eight glial tumor patients were imaged at our center during the study duration. Thirteen of these patients were excluded: two patients had previous neurological disorders (multiple sclerosis and stroke), ten tumors were recurrences and one patient presented excessive head motions during resting state fMRI acquisition. The remaining 35 patients were included in this analysis. The mean age was 44.6 years (range = [15–79]) with 18 females and 17 males. Among them, 6 patients were left-handed.

Seventeen patients had low-grade lesions (WHO-grade I = 7, grade II = 10) and 18 patients had high-grade tumors (WHO-grade III = 7, grade IV = 11). 24 tumors were localized in the left hemisphere and 11 in the right hemisphere (Table S1). Mean greatest diameter of the tumor (including solid and necrotic components but not peri-tumoral edema) was 33.7 mm (range = [6–67]). Of these, 12 tumors contained the canonical IDH1-R132H mutation and 16 were immunostain negative. Of note, mutation was not assessed for low-grade lesions such as dysembryoplastic neuroepithelial tumor.

Data of 70 age- and sex-matched participants from the 1000 functional connectomes project (36 females and 34 males) served as control data in our study, with mean age of 45.4 years (range = [18–79]).

Individualization of the networks with ICA and SBCA

Different components were identified as DMN parts using ICA with 10 to 100 ICs (every 10 ICs: 10, 20, 30 etc.): the main components were the anterior, posterior and superior DMN. At 60 ICs, we observed a split from the anterior and posterior DMN components. Only the superior DMN was consistently found across different analyses (Figure S1). Correlation coefficients were higher for ICA with a low number of ICs (Table S2). The highest correlation coefficients were noted for ICA with 20 ICs, for which two components (posterior and superior) were visually identified as part of the DMN (0.37 and 0.39). A single component was identified as the DAN using ICA with 10 to 40 ICs, and a separation into two components was noted starting at 50 ICs. A component corresponding to the right FPN was visualized using ICA with 10 and 20 ICs; the left FPN appeared when using 30 ICs and a split with the anterior part of the FPN was noted at 50 ICs (3 components).

For the SBCA, DMN was visually consistent using the 4 seeds (Figure S2-A), with high correlations between the time series of voxels in the MPFC, PCC, precuneus and angular gyri. The highly correlated areas were also consistent for the seeds of the DAN (FEF and IPS) and for the FPN (LPFC and PPC) (Figure S2-B;C).

Functional connectivity of the DMN

Global increased connectivity was found when examining the DMN in patients compared to controls (Fig. 1 and S3), with consistency using the 4 seeds for SBCA and varying numbers of ICs (anterior and superior DMN). Areas with significantly increased connectivity with DMN in patients versus the control groups were the subcallosal cortex, the precuneus, the frontal and occipital lobes and the cerebellum (Table 1; MPFC seed: main cluster including the subcallosal cortex, height: p < 10⁻⁶; PCC seed: main cluster including the precuneus, height: p = 0.006). However, an area of decreased connectivity was found using SBCA in the corpus callosum, precuneus, and the posterior cingulate cortex (Cluster 1: 358 voxels corresponding to the corpus callosum, height: p < 10⁻⁵), as shown in Fig. 1. This area of decreased connectivity was increased in size when comparing high-grade gliomas with controls and absent when comparing low-grade gliomas and controls. This cluster of decreased connectivity was also consistently detected using ICA when comparing the superior DMN of glioma patients versus controls (Table 2), when performing ICA with higher number of ICs. The results were also similar with SBCA when excluding patients in which the network of interest was not detected during first-level analysis.

Fig. 1.

DMN connectivity in glioma patients compared to controls showed increased functional connectivity in subcallosal, precuneus, occipital lobes and the cerebellum and decreased connectivity in the corpus callosum. The color bar represents T-values with increased connectivity (red/yellow) and decreased connectivity (blue/purple)

Table 1.

Seed-based correlation analysis of the default mode network

Anatomical structure	Cluster (x; y; z)	±	Size (cluster)	Size p-FDR	Peak p-FWE	Beta
MPFC
Subcallosum	(+00; +30; −16)	+	2048	<10−6	<10−6	0.30
Right Temporal gyri	(+64; −28; −30)	+	1927	<10−6	0.006	0.18
PCC
Precuneus	(−04; −42; +54)	+	5067	<10−6	0.0005	0.20
Subcallosum	(+06; +32; −12)	+	1492	<10−6	10−6	0.17
LPL
Subcallosum	(+02; +32; −16)	+	1761	<10−6	<10−6	0.20
Precuneus	(+00; −42; +56)	+	625	10−5	0.09	0.20
RPL
Subcallosum	(+02; +30; −16)	+	989	<10−6	0.0009	0.18
Precuneus/Left Post-central gyrus	(−20; −40; +68)	+	923	<10−6	0.38	0.18

(Patients versus Controls) showing the two clusters the most significant (=Most significant heights) using Voxel threshold: p < 0.001 (uncorrected); cluster threshold: p < 0.05 (FDR-corrected); ± : Positivity (increased connectivity in patients) or negativity (decreased connectivity in patients) in this area; Coordinates in the MNI space; MPFC medial pre-frontal cortex, PCC posterior cingulate cortex, LPL/RPL Left/Right Parietal Lobule, Beta beta values (Subjects>Controls) representing the average Fisher-transformed correlations with the seed within each cluster

Table 2.

Independent-Component analysis of the superior DMN

Area	Cluster (x; y; z)	±	Size	Peak p-FWE
IC10
Right LOC	(+54; −68; −10)	+	2810	0.0002
Left MTG	(−46; −14; +20)	+	2134	<10−5
IC20
Cerebellum(Crus2)	(+14; −78; −42)	+	1485	0.002
Frontal Orbital Cortex	(−04; +16; −26)	−	645	0.004
IC30a
Precuneus/PCC/CC	(+06; −44; +20)	−	2302	0.02
IC40
Precuneus/PCC/CC	(+04; −44; +20)	−	2036	0.02
Left Parietal Operculum	(−50; −38; +28)	+	593	0.01
IC50
PCC/CC	(−02; −20; +20)	−	551	0.29
Precuneus	(+10; −64; +28)	−	251	0.65
IC60
Left Frontal Pole/MFG	(−30; +08; +56)	+	2193	0.0007
PCC/CC	(−06; −24; +20)	−	474	0.19
IC70
Precuneus/PCC/CC	(+06; −44;+20)	−	2165	0.01
Left Frontal Pole	(−46; +26; +22)	+	372	0.71
IC80
Precuneus/PCC/CC	(+06; −44; +22)	−	3652	0.001
Left Frontal Pole	(−48; +42; +14)	+	407	0.05
IC90
Precuneus/PCC/CC	(+08; −64; +26)	−	2108	0.08
Left Frontal Pole	(−48; +42; +12)	+	333	0.25

(Patients versus Controls) showing the 2 clusters the most significant (=Most significant heights) using Voxel threshold: p < 0.001 (uncorrected); cluster threshold: p < 0.05 (FDR-corrected); ± : Positivity (increased connectivity in patients) or negativity (decreased connectivity in patients) in this area

^aOnly one significant cluster; Coordinates in the MNI space. LOC lateral occipital cortex, PCC Posterior Cingulate Cortex, CC Corpus Callosum, MFG Middle Frontal Gyrus, MTG Middle Temporal Gyrus

Functional connectivity of the DAN and FPN

DAN demonstrated consistent areas of increased connectivity in patients in the occipital region and small areas of increased connectivity in the medial prefrontal cortex (multiple clusters with height: p < 10⁻⁶) using both SBCA and ICA (Fig. 2a and S4). Decreased connectivity in the subcallosal cortex and anterior cingulate gyrus was found with SBCA (seed in the IPSs; Table S3) and ICA (10 and 70 ICs).

Fig. 2.

Dorsal Attention Network (a) and Fronto-Parietal Network (b) connectivity using Seed-Based Correlation Analysis (Left) and Independent Component Analysis (right); DAN connectivity in glioma patients (a) compared to controls showed increased functional connectivity (red) in occipital lobes and decreased connectivity (blue) in the subcallosum. The FPN connectivity in glioma patients (b) showed increased connectivity in the precuneus, posterior cingulate cortex and frontal lobes. The color bar represents T-values with increased connectivity (red/yellow) and decreased connectivity (blue/purple)

For the FPN, increased connectivity was noted in patients in the precuneus, posterior cingulate gyrus, and frontal cortex with SBCA and ICA (Fig. 2b and S5; Table S4). These areas of increased connectivity were similar to anatomical locations of the DMN.

Tumor grade, IDH1-R132H mutation status and functional connectivity

The decreased connectivity in the corpus callosum when exploring the DMN was more pronounced in high-grade glioma patients compared to controls than in other glioma patients. However, no difference in the connectivity of the networks of interest (for both SBCA and ICA) was demonstrated between low- and high-grade tumors for DMN, DAN and FPN. Furthermore, no difference was found as pertaining to their IDH1-R132H mutation status.

Tumor lateralization and size

For both SBC and ICA, there was no difference between tumors arising from right (N = 11) or left hemisphere (N = 24) when exploring the DMN, DAN and FPN. No difference was found between high and low-grade tumors of the left hemisphere. Furthermore, connectivity of tumors from the left frontal lobe didn’t differ from connectivity of tumors in the left temporal lobe. Finally, there was no difference in the connectivity of these 3 networks when comparing tumors >3 cm and <3 cm.

Discussion

Altered functional connectivity in glioma patients was found for the DMN, FPN, and DAN using both SBCA and ICA.

This study showed that the functional connectivity of the DMN was altered in patients with gliomas, with increased connectivity in the frontal and occipital regions and decreased connectivity in the corpus callosum and posterior cingulate cortex. Furthermore, increased global connectivity of the DAN and FPN was noted in all glioma patients. RsfMRI is commonly used to study networks in the gray-matter but can also detect interacting networks in the white-matter [27, 28]. These white-matter networks correlate with signals from functional gray-matter networks and include interhemispheric commissural bridges traversing the corpus callosum. Decreased connectivity in the corpus callosum could be explained by the decreased connection between the two cerebral hemispheres due to Wallerian degeneration (in the early stages) related to neuronal destruction by the tumor or disruption of the connections in the corpus callosum. Similar findings have already been depicted in brain tumors that do not infiltrate the corpus callosum in a previous study evaluating Diffusion Tensor Imaging [29]. A case of glioblastoma of the primary motor cortex associated with Wallerian degeneration of corticospinal tract (with no previous surgery) was described in the literature [30]. The white matter fibers from the prefrontal regions course through the anterior subregion of the rostrum and the genu while the temporo-parieto-occipital fibers project into the splenium [31]. Thus, an involvement of the DMN could be the result of disruption of the connections across corpus callosum. Another hypothesis could be that this disrupted functional connectivity results from the neuron-glioma interaction, bidirectional, which impacts tumor progression [32, 33]. Thus, gliomas increase neuronal activity while peritumoral neurons may stimulate tumor cell proliferation and exacer-bate tumoral progression. Previous study depicted that intra-operative electrocorticography demonstrate increased cortical excitability in glioma-infiltrated brain [34] This could explain increased functional connectivity in brain tumor patients. Future experiments could test if rsfMRI could be used as predictor of altered neurobiology and if altered connectivity may be predictor of impending tumor growth.

Our results also demonstrated that correlation coefficients used in ICA to identify networks (spatial match-to-template) were higher when the number of ICs was low. For example, the correlation coefficient to identify the superior DMN was 0.39 for ICA with 20 components, 0.26 with 60 components, and was only 0.16 with 90 components. These findings could be explained by overlapping components when using lower number of ICs, with increased noise included in these components and less reliable results than would be obtained with higher ICs. However, one or multiple splits in the network probably occur with higher number of ICs. Results of ICA should probably be compared using varying IC values for better accuracy.

Previous studies have shown heterogeneous results of the effect of brain tumors on functional connectivity, with decreased functional connectivity within well-established networks and increases in atypical functional connectivity pattern [18]. SBCA can reliably detect common functional connectivity networks in patients with glioma, with wide variations of functional connectivity in the DMN in high grade gliomas [35]. Harris et al. found that presence of high grade tumors results in decreased DMN functional connectivity [15], while Liu et al. showed increased connectivity with the precuneus when studying the ventral DMN which was associated with increased intra or cross-network interaction [36]. We found a similar increase in precuneus neural activity when comparing DMN connectivity of patients to that in controls. Thus, increased functional connectivity could be related to intra-network reorganization or a potential compensatory mechanism, possibly due to Wallerian degeneration of fiber tracts.

Only two previous studies described DAN involvement in 5 and 8 patients with brain lesions, using SBCA alone [37, 38] and no studies have described the effect of glial lesions on the FPN. We found that areas of increased connectivity in patients when studying FPN were similar to the DMN nodes (precuneus, PCC and frontal cortex). Thus, disruption of the DMN could lead to less effective inhibition of executive networks at rest, with increased connectivity of the FPN. The differences in the connectivity of these networks suggest that they are involved in the global disruption of brain connectivity related to glial tumors.

Numerous studies have shown that the white matter and gray matter tracts, studied by Diffusion Tensor Imaging (DTI) are disrupted by brain lesions, including those secondary to multiple sclerosis [39] and stroke [40] as well as tumors [41]. DTI is an interesting diagnostic tool with moderate diagnostic performance to differentiate high-grade gliomas from brain metastases [42] but also to study global brain reorganization. In glioma patients, DTI of the normal-appearing white matter exhibited decreased fractional anisotropy and axial diffusivity compared to controls, likely as a result of microstructural impairment [43]. DTI also showed contralateral brain reorganization in high-grade gliomas [44], which could be related to our findings.

This study has several limitations. First, the controls were taken from the 1000-functional connectome projects which had slightly different protocols of acquisition. Data were acquired with eyes closed but resting state condition (eyes closed or open) is an important variable with differences in brain activity that may limit the generalizability of our results [45]. Furthermore, these data were acquired after task-based fMRI while resting state networks can be affected when performed after tasks [46, 47]. We used SBCA which is subject to anatomic distortions caused by glial tumors. However, we also performed a second SBCA while excluding from group-SBCA the patients in which a network was not detected during the first level analysis using the seed of interest. The results of this second SBCA were similar to the initial SBCA, confirming that anatomic distortions didn’t affect the conclusions of this study. Another limitation is that the group level analyses performed in this study used heterogeneous group of glial tumors (different locations, grades). Lastly, the absence of appreciable differences in connectivity between low- and high-grade tumors or between IDH1-mutant or wildtype gliomas may be due to a small number of patients in each group and insufficient statistical power.

In conclusion, disrupted functional connectivity at rest of the DMN, DAN, and FPN was found in patients with glial tumors. A future study correlating DTI results with rsfMRI in patients with glial lesions would be interesting to correlate degeneration effects and areas of decreased connectivity. Furthermore, a new study evaluating these networks as part of standard imaging during neurosurgical planning could be of great interest to reduce cognitive impairment and behavioral changes after brain surgery.

Abbreviations

DAN

Dorsal attention network

DMN

Default mode network

FPN

Fronto-parietal executive control network

ICA

Independent component analysis

ICs

Independent components

IDH

Isocitrate dehydrogenase

rsfMRI

Resting state functional magnetic resonance imaging

SBCA

Seed-based correlation analysis