Contribution of slow, brain-wide patterns of activity to on-going experience in resting-state fMRI.
Poster No:
2022
Submission Type:
Abstract Submission
Authors:
Isabel Gephart1, Javier Gonzalez-Castillo2, Megan Spurney1,3, Daniel Handwerker1, Peter Bandettini1,4
Institutions:
1Section on Functional Imaging Methods, NIMH, NIH, Bethesda, MD, 2Section on Functional Imaging Methods, NIMH, Bethesda, MD, 3Clinical and Translational Neuroscience Branch, NIMH, NIH, Bethesda, MD, 4Functional MRI Core, NIMH, NIH, Bethesda, MD
First Author:
Co-Author(s):
Megan Spurney
Section on Functional Imaging Methods, NIMH, NIH|Clinical and Translational Neuroscience Branch, NIMH, NIH
Bethesda, MD|Bethesda, MD
Section on Functional Imaging Methods, NIMH, NIH|Clinical and Translational Neuroscience Branch, NIMH, NIH
Bethesda, MD|Bethesda, MD
Peter Bandettini, Ph.D.
Section on Functional Imaging Methods, NIMH, NIH|Functional MRI Core, NIMH, NIH
Bethesda, MD|Bethesda, MD
Section on Functional Imaging Methods, NIMH, NIH|Functional MRI Core, NIMH, NIH
Bethesda, MD|Bethesda, MD
Introduction:
Prior work has shown that subjective experience during scanning can influence such metrics as functional connectivity (FC) and regional homogeneity[1–3] of resting-state fMRI (rs-fMRI). Also, recent work shows that approx. 30% of the variance in rs-fMRI is accounted for by three low frequency spatiotemporal activity patterns found using complex principal component analysis (cPCA)[4]. It remains unknown how these patterns relate to rs-fMRI on-going experience. To address this question, we use rs-fMRI scans[6] annotated with subject's reports of the content and form of their thoughts, and wakefulness levels. We use connectome predictive modeling (CPM)[7] to assess how much these aspects of experience are reflected in the rs-fMRI data before and after regression of cPCA patterns.
Methods:
Dataset: 471 rs-fMRI scans (15 mins, TR=1.4s, voxel=2.3x2.3x2.3mm3) from the MPI-Leipzig Mind-Brain-Body dataset[6].
Experience Data: At the end of each scan, subjects completed the Short New York Cognition Questionnaire (sNYCQ; Fig. 1.A). sNYCQ responses shown in Fig. 1.B. The content and form responses were input into a Sparse Box-constrained Non-negative Matrix Factorization algorithm to extract linear positive combinations of questionnaire items that jointly explain variance, resulting in two summary factors that we refer to as "Thought Patterns" (TPs) (Fig. 1.C).
Basic fMRI Pipeline: Discard 5 TRs, motion correction, distortion correction, registration to MNI, scaling (divide by the mean), nuisance regression (motion + 1st derivative, linear & quadratic trends, COMPCOR[8] regressors), and filtering (0.01 - 0.1Hz). (Fig. 1.D-green).
cPCA Pipeline: We extracted the first three cPCA patterns using 50 randomly selected scans and procedures described in Bolt et al.[4] Next, we generated voxel-wise, scan-specific regressors for these patterns as in Abbas el al.[5] Finally, regressors were used as additional nuisance factors (Fig. 1.D-red).
FC Matrices: Constructed using the 400 ROI/7 Networks Schaefer Atlas[9] extended with 8 subcortical regions from the AAL atlas[10]. FC matrices computed for both pipelines.
CPM: We used TPs 1 & 2 and Wakefulness as prediction targets. We first attempted prediction using FC matrices from the basic pipeline. Prediction accuracy was evaluated as the correlation between observed and predicted values. As CPM is non-deterministic, we attempt prediction 100 times per target. We assess statistical significance by comparing to a null distribution generated using 10,000 randomizations. Next, we attempted prediction using FC matrices from the cPCA regression pipeline. Significant differences across pipelines evaluated via paired T-tests.
Experience Data: At the end of each scan, subjects completed the Short New York Cognition Questionnaire (sNYCQ; Fig. 1.A). sNYCQ responses shown in Fig. 1.B. The content and form responses were input into a Sparse Box-constrained Non-negative Matrix Factorization algorithm to extract linear positive combinations of questionnaire items that jointly explain variance, resulting in two summary factors that we refer to as "Thought Patterns" (TPs) (Fig. 1.C).
Basic fMRI Pipeline: Discard 5 TRs, motion correction, distortion correction, registration to MNI, scaling (divide by the mean), nuisance regression (motion + 1st derivative, linear & quadratic trends, COMPCOR[8] regressors), and filtering (0.01 - 0.1Hz). (Fig. 1.D-green).
cPCA Pipeline: We extracted the first three cPCA patterns using 50 randomly selected scans and procedures described in Bolt et al.[4] Next, we generated voxel-wise, scan-specific regressors for these patterns as in Abbas el al.[5] Finally, regressors were used as additional nuisance factors (Fig. 1.D-red).
FC Matrices: Constructed using the 400 ROI/7 Networks Schaefer Atlas[9] extended with 8 subcortical regions from the AAL atlas[10]. FC matrices computed for both pipelines.
CPM: We used TPs 1 & 2 and Wakefulness as prediction targets. We first attempted prediction using FC matrices from the basic pipeline. Prediction accuracy was evaluated as the correlation between observed and predicted values. As CPM is non-deterministic, we attempt prediction 100 times per target. We assess statistical significance by comparing to a null distribution generated using 10,000 randomizations. Next, we attempted prediction using FC matrices from the cPCA regression pipeline. Significant differences across pipelines evaluated via paired T-tests.
Results:
Fig. 2.A-C show our cPCA patterns, which account for 12.3% of variance: cPC1 shows an alternating activation/deactivation of sensory regions (Fig. 2.A), cPC2 alternating activation/deactivation of the DMN (Fig. 2.B), and cPC3 alternating activation/deactivation of attention and control networks (Fig. 2.C).
Fig. 2.D shows average FC across scans for basic pipeline and Fig. 2.E for the cPCA regression pipeline. Fig 2.F shows differences across pipelines [66796 edges (93%) significantly different at pFDR<0.05].
We can predict TPs 1, 2 and Wakefulness significantly above chance using the basic pipeline (Fig. 2.G). After cPCA regression, predictability of TP 1 and Wakefulness significantly improved (p<0.05), while predictability for TP 2 remained unchanged (Fig. 2.H).
Fig. 2.D shows average FC across scans for basic pipeline and Fig. 2.E for the cPCA regression pipeline. Fig 2.F shows differences across pipelines [66796 edges (93%) significantly different at pFDR<0.05].
We can predict TPs 1, 2 and Wakefulness significantly above chance using the basic pipeline (Fig. 2.G). After cPCA regression, predictability of TP 1 and Wakefulness significantly improved (p<0.05), while predictability for TP 2 remained unchanged (Fig. 2.H).
Conclusions:
We found cPCA patterns similar to those in Bolt et al.[4] although they differ in ordering and variance explained; likely due to disparities in applied smoothing (none here vs. 5mm in Bolt et al.[4]). Regressing cPCA patterns significantly altered FC across the brain, in agreement with similar analyses for quasi-periodic patterns[5]. Yet, cPCA regression had marginal or no effect on predictability of rs-fMRI on-going experience; suggesting that these activity patterns are not directly linked to rs-fMRI cognition.
Higher Cognitive Functions:
Higher Cognitive Functions Other
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Connectivity (eg. functional, effective, structural)
fMRI Connectivity and Network Modeling
Task-Independent and Resting-State Analysis 1
Keywords:
FUNCTIONAL MRI
Machine Learning
Meta-Cognition
Modeling
Open Data
Statistical Methods
Other - Connectome Predictive Modeling
1|2Indicates the priority used for review
Provide references using author date format
2. Gorgolewski, K. J. et al. A Correspondence between Individual Differences in the Brain’s Intrinsic Functional Architecture and the Content and Form of Self-Generated Thoughts. Plos One 9, e97176 (2014).
3. Gonzalez-Castillo, J. et al. Imaging the spontaneous flow of thought: Distinct periods of cognition contribute to dynamic functional connectivity during rest. Neuroimage 202, 116129 (2019).
4. Bolt, T. et al. A parsimonious description of global functional brain organization in three spatiotemporal patterns. Nat. Neurosci. 25, 1093–1103 (2022).
5. Abbas, A. et al. Quasi-periodic patterns contribute to functional connectivity in the brain. NeuroImage 191, 193–204 (2019).
6. Mendes, N. et al. A functional connectome phenotyping dataset including cognitive state and personality measures. Sci Data 6, 180307 (2019).
7. Shen, X. et al. Using connectome-based predictive modeling to predict individual behavior from brain connectivity. Nature Protocols 12, (2017).
8. Behzadi, Y., Restom, K., Liau, J. & Liu, T. T. A component based noise correction method (CompCor) for BOLD and perfusion based fMRI. Neuroimage 37, 90–101 (2007).
9. Schaefer, A. et al. Local-Global Parcellation of the Human Cerebral Cortex from Intrinsic Functional Connectivity MRI. Cereb Cortex 28, 3095–3114 (2017).
10. Tzourio-Mazoyer, N. et al. Automated Anatomical Labeling of Activations in SPM Using a Macroscopic Anatomical Parcellation of the MNI MRI Single-Subject Brain. Neuroimage 15, 273–289 (2002).