Static and dynamic functional connectivity supports the configuration of brain networks associated with creative cognition

Creative cognition is recognized to involve the integration of multiple spontaneous cognitive processes and is manifested as complex networks within and between the distributed brain regions. We propose that the processing of creative cognition involves the static and dynamic re-configuration of brain networks associated with complex cognitive processes. We applied the sliding-window approach followed by a community detection algorithm and novel measures of network flexibility on the blood-oxygen level dependent (BOLD) signal of 8 major functional brain networks to reveal static and dynamic alterations in the network reconfiguration during creative cognition using functional magnetic resonance imaging (fMRI). Our results demonstrate the temporal connectivity of the dynamic large-scale creative networks between default mode network (DMN), salience network, and cerebellar network during creative cognition, and advance our understanding of the network neuroscience of creative cognition.

The burgeoning field of network science and its applications to cognitive and clinical neuroscience has provided a framework to help understand the complex neurocognitive functions and reveal the pathological basis of neurological disorders. A network science approach helps us to understand the brain as a network through local and distributed processes [1][2][3][4] . Such an approach has been used in functional human neuroimaging (e.g., functional magnetic resonance imaging, fMRI) studies to understand Alzheimer's diseases [5][6][7][8][9] , epilepsy 10,11 and schizophrenia [12][13][14] , and to investigate cognitive functions associated with learning 15 , behavior 16 and task performance 17,18 . The relational and causal association of distributed brain regions with various cognitive functions have been mapped to reveal the connectome of the human brain 19,20 .
The neuroscientific literature mostly considers domain-general functions for which multiple regions and functions are engaged, resulting in diverse quantifications 19,21 . However, little has been reported of domain-specific cognitive functions associated with specific brain networks 22,23 . It is important to understand the domain-specific functions in the brain that result in activations in the specific brain regions 15,23 . Brain networks and the functional connections between brain regions involved in task execution behave dynamically with respect to time. Therefore, there is a need to explore the functional nature of cognition by investigating dynamic and domainspecific brain networks. This brain state during cognition can be uncovered by assessing connectivity in either static or dynamic brain networks. Most neuroimaging studies represent static brain networks over the entire duration of an fMRI experiment [24][25][26] . These networks represent the functional connectivity and interactions between various cortical and subcortical regions of the brain over this task-related period. However, activity in static networks does not clearly show the changes that occur over short durations during the fMRI scan; hence, there is a need to develop dynamic reconfiguration-based methods to identify the modular architectures of evolving networks 15,27,28 . Minute-by-minute changes in neuronal activities or blood flow changes in the order . The z-scored Pearson correlation coefficients for 32 ROIs were calculated and a statistical threshold of p < 0.05, false discovery rate (FDR)-correction was applied to obtain the estimated correlation map. A single adjacency matrix was estimated and a community detection algorithm was applied to identify communities for the whole brain flexibility over the entire scan duration. (B) Dynamic reconfiguration. The Blood oxygen level dependent (BOLD) time series signal was extracted from each of the 32 regions of interests (ROIs). An overlapping sliding window (length W) was used and time ordered functional connectivity matrices were estimated over these windows for all the ROIs using z-scored Pearson correlation. The estimated correlation maps W n (n is number of windows) were statistically thresholded with p < 0.05 (FDR correction). A community detection algorithm was applied to all these matrices to detect communities across the windows. www.nature.com/scientificreports/ session FPN connected with the VN. The cerebellum in the task session 1 showed connectivity with the DMN, LN and SN whereas in task session 2 FPN connected with the SMR, SN and dorsolateral network (DA). An intrinsic functional network architecture was seen in both the task sessions along with the resting-state session indicating creative cognition during task and at rest. The region-specific interactions showed slight variations in functional connectivity in the three sessions. The regions of the DMN showed stronger intra-DMN connections along with the bilateral posterior superior temporal gyrus in all the three sessions. During task session 1, the regions of the DMN showed stronger connectivity with the bilateral prefrontal cortex, bilateral posterior parietal cortex, and the right inferior frontal gyrus. The connectivity between the regions of the DMN-bilateral posterior parietal cortex was also seen during task session 2. On the other hand, during the resting-state session observed regions of the DMN having stronger connections with the posterior cerebellum. The FPN showed intra-FPN connectivity across all sessions. During task session 1, the FPN showed stronger connectivity with the regions of the DMN, bilateral prefrontal cortex, right inferior frontal gyrus, left posterior superior temporal gyrus and the posterior cerebellum whereas during task session 2, the FPN showed stronger connections with the left posterior superior temporal gyrus, left inferior frontal gyrus, left lateral parietal cortex and posterior cerebellum. During the resting-state session, the FPN showed connectivity with only the anterior cerebellum. The results of the static configuration were indicative of the role of DMN and FPN and their interaction with other brain networks associated with creative cognition during task session 1 and 2. The resting-state session showed lesser inter-regional connectivity between the DMN, FPN and other networks associated with creative cognition. The cerebellum across all sessions showed intra-cerebellar connectivity and the bilateral prefrontal cortex. During the task session 1, the cerebellum showed stronger connectivity with the visual cortex (occipital and lateral) and bilateral inferior frontal gyrus whereas during the task session 2, the cerebellum showed stronger connectivity with the visual cortex (occipital and lateral) only. During the resting-state session, the cerebellum showed stronger connectivity with the bilateral posterior parietal cortex, visual cortex (lateral, medial and occipital) and left prefrontal cortex. The results of the cerebellar connectivity with the cerebral regions during task sessions 1 and 2 are indicative of the role of cerebro-cerebellar connectivity and the role of the cerebellum in creative cognition. The functional connections are represented by the respective correlation matrices and the networks formed have been depicted in row 5 of Fig. 3 for all the three sessions.
Task and rest dynamic functional connectivity results. All the results of the dynamic functional connectivity were reported with a threshold of p < 0.05 FDR-corrected. The total number of windows (105 for task and 94 for rest) were divided into four halves and the average of each half was taken and a new window between the time points was constructed as shown in Fig. 3 (row 1, 2 and 3). Similar to the static configuration, the functional connections between brain regions associated with creative cognition during creative task sessions (1 and 2) and resting-state session were compared with the dynamic functional connectivity analysis depicted in row 4 of Fig. 3.
In region-specific interactions, during the first windowed average during task session 1 (i.e. time 1-51 s), the regions of the DMN showed intra-DMN connectivity and with the visual cortex (lateral, medial and occipital), left prefrontal cortex, right posterior parietal cortex, bilateral inferior frontal gyrus and left posterior superior temporal gyrus. The regions of the FPN during this window showed stronger intra-FPN connectivity and with the bilateral parietal cortex, posterior cingulate cortex, lateral regions of the visual cortex, bilateral ipsilateral sulcus, left supramarginal gyrus and superior sensorimotor cortex. The regions of the cerebellum showed stronger connections with the medial and occipital regions of the visual cortex. During the task session 2 (i.e. time 1-51 s), the region of the FPN showed strong connections with the posterior cingulate cortex, lateral parietal cortex, left supramarginal gyrus, right posterior cingulate cortex, right inferior frontal gyrus and bilateral superior temporal gyrus. The cerebellum showed stronger intra-cerebellar connections, left parietal cortex, bilateral posterior parietal cortex and right superior temporal gyrus. For the resting-state session, during the first averaged window (i.e. time 1-33 s), the DMN showed connectivity with the posterior cingulate cortex and occipital and lateral visual cortex. The regions of the FPN showed intra-FPN connections and with superior sensorimotor cortex, right insula, right supramarginal gyrus, bilateral ipsilateral sulcus and anterior cerebellum. The regions of the cerebellum showed strong connectivity with the right prefrontal cortex. The functional connections are represented by the respective correlation matrices and the networks formed have been depicted in row 1 of Fig. 3 for all the three sessions.
For the second windowed average during task session 1 (i.e. time 53-103 s), the DMN showed intra-DMN connections along with superior sensorimotor cortex, lateral visual cortex, right anterior insula, bilateral prefrontal cortex, right supramarginal gyrus, bilateral ipsilateral sulcus, right prefrontal cortex, right posterior parietal cortex, right inferior frontal gyrus, right superior temporal gyrus and anterior cerebellum.

Community detection analysis and flexibility results.
To examine the evolution of the brain networks associated with creative cognition during the convergent thinking creative task sessions (1 and 2) and resting-state session, a fast-greedy community detection algorithm was applied to the static and dynamic connectivity matrices. All the results of the community detection and network flexibility were reported with a threshold of p < 0.05 (FDR-corrected). The focus of this study was the brain networks associated with creative cognition representing the DMN, FPN and the CN and the regions associated with these networks. The interactions within these regions was also an important focus of this study. Changes in the dynamic functional connectivity of the 8 major functional networks involving various brain regions for various averaged time windows are shown in Fig. 4.
Only two communities were detected in the static configuration associated with creative cognition over the entire scan duration for the whole creative task session as depicted in Fig. 1. After dividing the entire scan duration for task into two sessions i.e. task session 1 and task session 2, we found two communities for task sessions 1 and 2 and for rest as well which is depicted in  . Network-specific dynamic re-configuration of creative brain using sliding window approach (window size = 44 s) for creative task session 1, creative task session 2 and resting-state session. The 32 BOLD time series signals representing 32 ROIs were averaged to form 8 distinct networks. The first column in each session indicates the single windowed correlation map between these times. The bottom row indicates the average of all the first four rows across all the sessions. Second column in every session indicates the graph representation and community detection formed using the fast-greedy algorithm community detection. Each colour indicates community formation for every session, averaged and every windowed correlation map. The edge color here represents the connection within the same community (same edge color of that of the community) or from one community to the other community (edge color of one community indicates the connection from that community to the other). Node Label List: DMN default mode network, FP frontoparietal network, SMR somatomotor network, DA dorsal attention network, SN salience network, VN visual network, LN language network, CN cerebellar network.  Fig. 4 for all the three sessions. The DMN, FPN have been considered as important construct networks for the processes involved during creative cognition. Apart from all these there was variation in the network connections of the salience network which formed communities with the VN and SMR. Further, there was differentiation in the dynamic re-configuration observed with a 66 s window approach and the network interaction have been depicted in the supplementary information ( Supplementary Fig. S3).

Scientific Reports
Flexibility is an important parameter used to quantify the community evolution of the dynamic brain network in this task-based fMRI study. The variation in flexibility across the dynamic temporal windows is depicted in Fig. 5A. Flexibility in the dynamic network oscillates across the entire time window. Maximum flexibility was observed during the creative thinking task and this increase in flexibility across windows was not modulated by the creative task. We did not find variation in the network flexibility between the CAT and math tasks or across the entire fMRI session. www.nature.com/scientificreports/ p < 0.001]. Flexibility of the dynamic brain network associated with both creative task and rest sessions using the 66 s window approach has been depicted in the Supplementary Information (Supplementary Fig. S4).

Discussion
Network analysis methods in neuroscience help to uncover the coverts processes of cognition. This is achieved by examining the activities of various brain networks involved during various cognitive processes. Creative cognition is a complex cognitive process that involves the engagement of various networks to produce unusual and unique ideas. In this fMRI study, we used a data-driven approach to characterize the neural processes involved during a convergent thinking task that involves the reconfiguration of the large and complex creative networks. Using a network neuroscience approach, we provide evidence of dynamic functional connectivity organized into different communities during creative thinking tasks. The results indicate the activation of the FPN and DMN for creative ideation, as described by previous creativity studies. We also examined the dynamic reconfiguration and interaction and evolution of these brain networks associated with creative cognition. Our results provide support for the evolution and engagement of these brain networks associated with creative cognition which are relevant to distributed brain networks. Creativity can be categorized into various processes including, but are not limited to, idea generation and retrieval and the combination of remote semantic associations. Creative thinking is a bottom-up process that involves the engagement of the DMN. The combination of semantic associations and their retrieval also follows a bottom-up approach that involves engagement of the DMN and the FPN. Our results extend the prior knowledge of the network neuroscience of creative cognition by revealing the engagement of these two networks, which have been shown by previous studies to be responsible for creative cognition 38,[91][92][93] . Previous creativity studies have identified the medial pre-frontal cortex and posterior cingulate cortex as the core regions of the DMN involved in creative thought generation 35,94,95 . The dynamic association between the DMN and the FPN indicate that the strength of the connections between them is greater during creative cognition. Jung et al. 39 proposed that interaction between the DMN and the FPN occurs during the production of novel and useful ideas in creative thinking, which is also one of the findings of the present study, in which we tracked the transitions in the network reconfiguration associated with the adaptive nature of the DMN and FPN and examined their role in creative cognition.
Very few neuroimaging studies have examined the role of the subcortical regions such as the cerebellum in cognitive tasks. The cerebellum is considered to play a role in locomotion and co-ordination 96,97 . Some studies have investigated its role in cognitive functions [98][99][100] . Other studies have suggested that there is connectivity between the cerebellum and the brain regions responsible for creative cognition [101][102][103][104][105] . In this task-based fMRI study, we highlight the role of the cerebellum in creativity using a dynamic connectivity approach based on community detection. Our results reveal interplay between the cerebellum and the DMN during a creative thinking task in what has been referred to as a "hub system" of creative cognition 106 . This interplay between the DMN and the cerebellum explains the role of dynamic brain networks in this task-based fMRI study. Our findings extend the understanding of the role and evolution of the DMN and the cerebellum in convergent thinking tasks.
We also examined network flexibility-a network parameter that has previously been linked with changes in brain networks 17 . Brain network flexibility can be used to track the specific and repetitive changes in networks during a task. The present study also focused on the temporal dynamics of brain networks and tracked the transition of different brain states. Little or no change in flexibility was found across the whole fMRI session during a creative thinking task. In the major networks for creative thinking, the DMN and FPN, greater flexibility was found in the FPN than in DMN during a creative thinking task indicating that the FPN is more likely than the DMN to change its allegiances during combination tasks or a combination of remote semantic association task. The domain-general role of networks, resulting in an increase of flexibility in the FPN, has been reported previously for various tasks 69,107,108 . Our results also suggest the reconfiguration and reorganization of dynamic brain networks. In this specific task-related fMRI study, the communities associated with the frontal network showed strong associations with the posterior regions such as the CN.
To conclude, the present study extends the dynamic role of functional connectivity between various creative regions previously defined by Beaty et al. 35 . Our results corroborated the role of the DMN, which is the hub of creative processes and important for idea generation during a creative thinking task 109 . Further, our findings support the role of the cerebellum during a creative thinking task in a large-scale brain network associated with creative cognition, which was previously found in neuroimaging studies 96,98,99 . We also observed the role of interactions between the DMN and various other networks during the creative thinking task. Using the dynamic network neuroscience approach to reveal the dynamic functional connectivity between various brain networks associated with creative cognition during convergent thinking sheds light on the evolution of brain networks associated with creative cognition.

Methods
Participants. Nineteen healthy, young, right-handed adults (8 females; 11 males; mean age = 23 ± 2.28 years; age range 20-29 years) participated in this event-related fMRI study. The participants were asked to provide written consent for their participation. The study was approved by the Institutional Review Board (IRB) of Academia Sinica and National Chiao Tung University, Taiwan. The study was carried out in accordance with the relevant ethical guidelines and regulations. The participants had no history of neurological or psychiatric disorders. They were screened for normal or corrected to normal vision. A written informed consent was given by all participants prior to their participation.  111 , including the subsets of immediate and delayed memory assessment to measure shortterm and long-term memory performance, the vocabulary test to assess language knowledge, digit-symbol and letter-number sequencing tests to assess the psychomotor speed, digital span-forward and digital span-backward tests to measure the capacity of working memory, and the arithmetic test to measure quantitative reasoning and ability. The California Verbal Learning Test (CVLT) was used to measure episodic verbal learning and memory performance. The creative quotient was quantified using creativity assessment questionnaire (CAQ) 112 with the participants showing a mean score of 10.78 ± 8.97. The participants performed digit span-forward and digit span-backward test to assess short term memory 113 and showed a mean score of 15.42 ± 0.69 and 11.78 ± 2.61 respectively. A letter-number sequencing test was conducted to measure working memory capacity 113 where the participants showed a mean score of 16.68 ± 3.55.
Creative fMRI task. The participants performed a modified version of the Chinese version of the remote associates task (CAT), which represents convergent thinking as defined by Huang et al. 74 . In this task, the participants were presented with three Chinese "stimulus words" and were asked to find the "target word, " which was semantically related to the three stimulus words. Before the participants entered the MRI scanner, they took part in practice sessions to understand the kind of test they would be asked to complete. The fMRI experiment consisted of two runs of the CAT, with each run consisting of 14 trials, yielding 28 trials. The experimental paradigm consisted of a 2 s fixation at the start of the experiment followed by a CAT question of 20 s duration. A math problem was provided of 8 s duration after the CAT question, to distract the participants from remembering the previous CAT question. Finally, a jitter of 4 s, 6 s or 8 s (mean = 6 s) was randomly provided for every subject. The duration of each run was 36 s and the experiment consisted of 14 trials of 504 s duration. The participants could press the button only once when they had an answer. An MRI-compatible response box was used to record the button press responses. The responses were verified to meet the CAT demands.
Imaging data acquisition and pre-processing. Whole-brain imaging data were recorded at the National Yang-Ming University, Taipei, Taiwan using a Siemens 3.0T Magnetron Trio MRI scanner (Siemens, Germany). Using a single-shot T2* weighted echo planar image (EPI) sequence, 252 functional scans were obtained in each run. The image parameters were as follows: TE/TR = 30 ms/2000 ms, flip angle = 90°, Thirty-three contiguous axial slices were acquired with a slice thickness of 4 mm, a 64 × 64 matrix, and a 3.44 mm × 3.44 mm in-plane resolution, FOV = 220 mm. Structural images were acquired with a 3-D ultrafast magnetization prepared rapid acquisition with gradient echo (MPRAGE) imaging sequence with the following parameters: FOV = 256 mm, resolution 256 × 256, slice thickness 1 mm, 192 slices, TR 3500 ms, TE 3.5 ms, TI 1100 ms, and flip angle 7° with no gap between slices. During the fMRI scanning, foam padding and earplugs were used to limit head movement and reduce scanner noise for the subjects. All pre-processing and denoising of the acquired fMRI data were done using CONN (http://www.nitrc .org/ proje cts/conn) toolbox in MATLAB 2018b (https ://matla b.mathw orks.com). All the pre-processing steps were performed with the pre-processing pipeline suggested by CONN for volume-based analysis 114 . First, the functional data was realigned and unwrapped for subject motion estimation and correction. Next, the data was slicetime corrected using an ascending interleaved slice order. After realignment, the fMRI data was controlled for all the head motion artifacts. Artifact reduction tools (ART)-based scrubbing was then used to detect and repair the bad volumes using two measures: (1) framewise displacement is greater than 0.9 mm in all directions and (2) global mean intensity threshold is greater than 5 standard deviations from the mean intensity for the entire scan. The artifact reduced functional data was co-registered using an affine transformation to the structural data using the inter-modality co-registration procedure 115,116 in SPM 12 (Wellcome Department of Imaging Neuroscience, London, U.K.). Then the structural MPRAGE data was normalized into the standardized Montreal Neurological Institute (MNI) space and segmented into white matter (WM), grey matter (GM) and cerebrospinal fluid (CSF) tissue classes using the SPM 12 unified-segmentation and normalization procedure 117 . The indirect normalization in CONN applied this unified-segmentation and normalization procedure to the structural data using the T1 volumes as a reference image and applies similar estimated non-linear transformation to the functional scans. These data are then resampled to the 180 × 216 × 180 mm bounding box with a 2 mm and 1 mm isotropic voxel for functional and structural data respectively using a 4th order spline interpolation. The normalized scans were then smoothened by spatial convolution with an 8-mm full-width half-maximum Gaussian kernel. Apart from the ART-based noise correction, six head motion parameters were obtained from the spatial motion correction and were added for the denoising steps.
To estimate the potential influences of head motion on functional connectivity, the mean head motion for all the participants and the maximum head motion of individuals were estimated using frame wise displacement (FD) 89 in creative task fMRI and resting-state fMRI data. A series of correlation analysis between the mean FD and the BOLD time series signal across the ROIs were performed.
Moreover, the anatomical component-based noise correction method (CompCor) 118  Seed ROI analysis. Thirty-two seed regions of interest (ROIs) were defined in the CONN toolbox, based on those originally defined by ICA analysis based on the Human Connectome Project dataset (http://www.human conne ctome proje ct.org) for 497 subjects, like Wolak et al. 122 . The use of publicly available ROI regions in this study would provide a platform of external validation on our findings by other research groups and minimizes bias when the ROIs were selected from our own dataset. The 32 seed ROIs within anatomically and functionally defined cortical and subcortical regions are listed in Table 1 and depicted in the Supplementary Information  (Supplementary Fig. S1). For each participant, we estimated individual BOLD time series for the 32 seed ROIs. BOLD signals for each of the individual seed ROIs lasted for 504 s for task run and 180 s for resting-state data.
Task and rest static functional connectivity analysis. Since the creative-task session was longer than the resting-state session, we divided the task session into two equal halves, i.e. creative task session 1 and crea- www.nature.com/scientificreports/ tive task session 2 (252 s duration each) and compared with the resting-state session (180 s duration), a similar approach used by Denkova et al. 123 was used in this study. We examined the static nature of the functional connectivity for the task sessions and the resting-state session. We obtained the BOLD time-series signals for all the participants from 32 ROIs representing 8 known functional networks defined in Table 1. The data for all participants was concatenated and the z-scores were estimated. The Pearson correlation coefficients of the z-scored data was obtained to form 32 × 32 correlation maps for all the three sessions. The correlation maps were further tested for FDR of p < 0.05 15,124,125 . Adjacency matrices were formed using graph measures.
Task and rest dynamic functional connectivity analysis. The time series BOLD signals for all the three sessions were obtained from 32 seed ROIs and were concatenated for the dynamic functional connectivity analysis. To examine dynamic functional connectivity across all the sessions, we used the most used sliding window approach to segment the BOLD signals into time windows. The selection of window length used was 44 s and 66 s with a step of 1TR was based on previous literature which also utilized a window length between 30 and 60 s 123,126-132 and also suggesting that, the variability found in this window length (30-60 s), is not found in larger window lengths 126,133 . In the 44 s window length approach, we obtained for 105 windows for the task, session 1 and session 2 each and 69 windows for the resting-state session whereas for the 66 s window length approach, we obtained 94 windows for task sessions each and 58 windows for the resting-state session. We calculated the z-scores followed by the Pearson correlation coefficients for all windows to obtain a 32 × 32 correlation map for every window.
As noted by Achard et al. 124 , not every element in a correlation matrix indicates a functional interaction between different regions. Therefore, it is necessary to compute a statistical false discovery rate (FDR) correction to rate only the significant and true values. We therefore estimated the significant p-values for every 32 × 32 window matrix for dynamic reconfiguration indicating the probability of obtaining correlation coefficient values as large as the observed values when there is null true correlation. We obtained the p-value matrix by using t-tests between the ROIs using the corrcoef function in MATLAB 134 . We further tested the significant p-values for FDR of p < 0.05 15,124,125 . The final correlation matrix contained only correlations that passed the FDR threshold. Values that did not pass the threshold were set to zero. The FDR based statistical threshold was applied to the 105 windows for task sessions and 69 windows for the resting-state session, irrespective of correlation coefficient being positive or negative. The FDR-corrected correlation matrices for each participant were calculated for the creative task and resting-state sessions. The respective adjacency matrices and community-based networks were obtained for all the windows across all sessions.
To understand the evolution of dynamic functional connectivity across all the 105 windows for both creative task sessions (1 and 2), we obtained a single windowed correlation map between the times 1-51 s, 53-103 s, 105-155 s and 157-208 s. Similarly, to understand the evolution of dynamic functional connectivity across 69 windows for resting-state session, we obtained single windowed correlation maps between the times 1-33 s, 35-67 s, 69-101 s and 103-136 s. Network specific interaction for the dynamic connectivity analysis were also obtained for all windows and sessions. A similar approach was used for the window length 66 s to obtain single windowed correlation maps between the times 1-47 s, 49-95 s, 97-143 s and 145-186 s for task sessions and single windowed correlation maps between the times 1-29 s, 31-59 s, 61-89 s and 91-114 s for resting-state session.

Network interactions and community detection.
Community detection is an approach to decompose a network into sub-networks. In this study, a community detection algorithm was used to understand if communities exist in all the three sessions. The algorithm was applied to the FDR-corrected adjacency matrices in static reconfiguration and dynamic reconfiguration in both creative task session as well as resting-state session. This was done to understand the region-specific interaction during creative task performance as well as during resting-state. We used R software (https ://www.r-proje ct.org/) and its integrated development environment R-Studio (https ://www.rstud io.com/produ cts/rstud io/) to develop an in-house code for community detection. After the identification of the communities in the dynamic brain networks, the flexibility of the networks was calculated to identify temporal variability of brain networks.
In terms of static interactions, a community detection algorithm was applied to identify communities for the whole brain over the entire scan duration to observe the region-specific changes in the 32 distinct functional brain regions. The community detection algorithm was also applied to understand the role of the 8 distinct functional networks and to observe network-specific changes across all the three sessions.
In terms of dynamic interactions, the FDR-corrected windowed correlation maps between respective times across all the three sessions were then transformed into adjacency matrices to better understand the nature of functional interactions dynamic functional connectivity. For the window length of 44 s, using graph measures the 105 (for each task session) and 69 (for resting-state session), correlation maps were converted to respective adjacency matrices. A community detection algorithm was applied to each of these adjacency matrices across all three sessions to observe the dynamic changes in communities and interaction between 32 ROIs. A similar approach was used to understand the nature of the algorithm on network-specific changes.
The modularity of a network, i.e., the probability of the number of edges that fall in a particular community 85 , was defined using equation I: