Dynamics of functional modes represent behavior during task fMRI
Poster No:
1774
Submission Type:
Abstract Submission
Authors:
Ikko Kimura1, Janine Bijsterbosch2, Matthew Glasser2, Takuya Hayashi1
Institutions:
1Laboratory for Brain Connectomics Imaging, RIKEN Center for Biosystems Dynamics Research, Kobe, Hyogo, 2Department of Radiology, Washington University in St Louis, St Louis, MO
First Author:
Ikko Kimura
Laboratory for Brain Connectomics Imaging, RIKEN Center for Biosystems Dynamics Research
Kobe, Hyogo
Laboratory for Brain Connectomics Imaging, RIKEN Center for Biosystems Dynamics Research
Kobe, Hyogo
Co-Author(s):
Takuya Hayashi
Laboratory for Brain Connectomics Imaging, RIKEN Center for Biosystems Dynamics Research
Kobe, Hyogo
Laboratory for Brain Connectomics Imaging, RIKEN Center for Biosystems Dynamics Research
Kobe, Hyogo
Introduction:
Several methods have been proposed to decompose fMRI data into a small number of networks1. Among them, PRObabilistic FUnctional MOdes (PROFUMO)2 utilizes bayesian modeling and better characterizes the individual traits of the resting-state networks than conventional approaches with independent component analysis (ICA)3. While the previous studies used PROFUMO for resolving mode at rest2–5, it is not well known whether it is applicable to evaluate modes during tasks. Here, we examined whether the modes (1) can be reproducibly estimated from task fMRI data and (2) are sensitive to the conditional differences or the individual variability in task performance during the scan.
Methods:
We obtained fMRI data during working memory tasks (tfMRI) and at rest (rsfMRI) of 50 unrelated participants from Human Connectome Project Young Adult database. Both data were denoised with multi-run FIX, re-applied manually reclassified noise components, surface registered with multi-modal surface matching algorithm6, de-drifted, and resampled to 32k standard mesh surfaces. PROFUMO was applied separately to tfMRI and rsfMRI data (dimension: 50) to extract the spatial maps and time signal of each mode. We chose five major modes for further analysis (i.e., dorsal attention [DA], left fronto-parietal [left FP], right fronto-parietal control [right FP], default mode network [DMN] and posteromedial cortex, inferior parietal lobule [PMPL]) reported in the previous study2. Similar functional networks were also extracted using group spatial ICA and dual regression (ICA-DR)7 for comparison.
The similarity of the spatial maps derived from tfMRI to those from rsfMRI data was assessed by cosine similarity. A general linear model (GLM) was applied to evaluate the degree of differences in the time signal of the modes across conditions (i.e., 2-back tasks [2bk], 0-back tasks [0bk], and at rest). Dynamic functional connectivity (dyFC)8 between modes was calculated using a sliding window method (timepoint window: 15, sliding steps: 1) to test whether interactions between modes changed across conditions. We also tested whether the accuracy of the WM task was correlated with the metrics derived from PROFUMO with Pearson's correlation coefficient.
The similarity of the spatial maps derived from tfMRI to those from rsfMRI data was assessed by cosine similarity. A general linear model (GLM) was applied to evaluate the degree of differences in the time signal of the modes across conditions (i.e., 2-back tasks [2bk], 0-back tasks [0bk], and at rest). Dynamic functional connectivity (dyFC)8 between modes was calculated using a sliding window method (timepoint window: 15, sliding steps: 1) to test whether interactions between modes changed across conditions. We also tested whether the accuracy of the WM task was correlated with the metrics derived from PROFUMO with Pearson's correlation coefficient.
Results:
Figure 1 shows the spatial maps derived from PROFUMO (Fig. 1A) and ICA-DR (Fig. 1B), highlighting that more widespread brain regions were positively involved in PROFUMO than in ICA-DR (red arrows). Modes from tfMRI showed high spatial cosine similarity with those from rsfMRI (0.78–0.89).
Figure 2 depicts the group-averaged time signal of each mode in tfMRI (Fig. 2A) and the estimated contrasts with GLM analysis (Fig. 2B). While task-related signal increase (2bk vs. 0bk) was clearly found in both PROFUMO and ICA-DA (right FP and DA), the resting-related signal change (rest vs. task) was relatively small in PROFUMO compared to ICA-DR (DA, left FP, DMN, PMPL; red arrows in Fig. 2A). In PROFUMO, the task-related contrast of DMN (2bk vs. 0bk) showed a negative correlation with the task accuracy (rho = -0.43, FWE-corrected P = 0.023) but not in ICA-DR.
Figure 2C illustrates the Z statistics (thresholded with FWE-corrected P < 0.05) comparing the dyFC between 2bk and the others. The dyFCs from PROFUMO revealed conditional differences in a larger number of modes than those from ICA-DR. Moreover, the task accuracy was positively correlated with the temporal variability of dyFC between right FP and DMN (rho = 0.41, FWE-corrected P = 0.031) and between right FP and PMPL (rho = 0.47, FWE-corrected P = 0.0064) in PROFUMO while not in ICA-DR.
Figure 2 depicts the group-averaged time signal of each mode in tfMRI (Fig. 2A) and the estimated contrasts with GLM analysis (Fig. 2B). While task-related signal increase (2bk vs. 0bk) was clearly found in both PROFUMO and ICA-DA (right FP and DA), the resting-related signal change (rest vs. task) was relatively small in PROFUMO compared to ICA-DR (DA, left FP, DMN, PMPL; red arrows in Fig. 2A). In PROFUMO, the task-related contrast of DMN (2bk vs. 0bk) showed a negative correlation with the task accuracy (rho = -0.43, FWE-corrected P = 0.023) but not in ICA-DR.
Figure 2C illustrates the Z statistics (thresholded with FWE-corrected P < 0.05) comparing the dyFC between 2bk and the others. The dyFCs from PROFUMO revealed conditional differences in a larger number of modes than those from ICA-DR. Moreover, the task accuracy was positively correlated with the temporal variability of dyFC between right FP and DMN (rho = 0.41, FWE-corrected P = 0.031) and between right FP and PMPL (rho = 0.47, FWE-corrected P = 0.0064) in PROFUMO while not in ICA-DR.
·Figure 1
·Figure 2
Conclusions:
PROFUMO can reliably extract the spatial maps of the major modes from tfMRI as from rsfMRI. While both PROFUMO and ICA-DR depicted a distinct pattern of time signal across conditions, PROFUMO was sensitive to the differences in the interactions of modes across conditions and to the individual variability in task performance.
Modeling and Analysis Methods:
Bayesian Modeling 2
fMRI Connectivity and Network Modeling 1
Keywords:
FUNCTIONAL MRI
Modeling
Other - task fMRI; functional connectivity; individual variability; working memory; dynamic functional connectivity; bayesian modeling; bayesian inference; PROFUMO
1|2Indicates the priority used for review
Provide references using author date format
2. Harrison, S. J. et al. (2015), Large-scale probabilistic functional modes from resting state fMRI. Neuroimage 109, 217–231.
3. Harrison, S. J. et al. (2020), Modelling subject variability in the spatial and temporal characteristics of functional modes. Neuroimage 222, 117226.
4. Bijsterbosch, J. D. et al. (2018), The relationship between spatial configuration and functional connectivity of brain regions. Elife 7, e32992.
5. Bijsterbosch, J. D., Beckmann, C. F., Woolrich, M. W., Smith, S. M. & Harrison, S. J. (2019), The relationship between spatial configuration and functional connectivity of brain regions revisited. Elife 8.
6. Robinson, E. C. et al. (2018), Multimodal surface matching with higher-order smoothness constraints. Neuroimage 167, 453–465.
7. Nickerson, L. D., Smith, S. M., Öngür, D. & Beckmann, C. F. (2017), Using Dual Regression to Investigate Network Shape and Amplitude in Functional Connectivity Analyses. Front. Neurosci. 11, 115.
8. Hutchison, R. M. et al. (2013), Dynamic functional connectivity: promise, issues, and interpretations. Neuroimage 80, 360–378.