Stand-By Time
Monday, June 27, 2016: 12:45 PM - 2:45 PM
Poster Number:
2214
Submission Type:
Abstract Submission
On Display:
Monday, June 27 & Tuesday, June 28
Authors:
Minqi Chong1, Chitresh Bhushan1, Anand Joshi1, Justin Haldar1, R. Nathan Spreng2, Richard Leahy1
Institutions:
1Univ. of Southern California, Los Angeles, United States, 2Cornell University, Ithaca, United States
Introduction:
Cortical parcellation based on resting fMRI is an important tool for investigating the functional organization and connectivity of the cerebral cortex in-vivo [1]. However individual anatomical and functional variation introduces ambiguity in group-wise or atlas-based parcellation approaches based on anatomically-drive image alignment [2]. We previously described a method [3] for performing individualized functional parcellation that uses a group approach to estimate the underlying functional connectivity between these parcels. We use a maximum a posterior (MAP) approach with a Multi-Variate Gaussian (MVG) likelihood, group sparsity constraint and a Markov Random Field (MRF) spatial prior to model the parcellation. We evaluated our approach using HCP resting fMRI data with task fMRI as ground truth. We compared with shrinkage estimation [4] and spectral clustering [5]. Our approach demonstrated an improved homogeneity in parcellation and higher consistency with task-based activation maps.
Methods:
Our method uses MAP estimation to model variability in individual parcellations while incorporating a group level connectivity consistency. We assume that a functional parcellation of cerebral cortex exists with the same number of regions for each subject [6]. These regions form the nodes of a graph. The connectivity between these nodes, represented by the nonzero elements of partial correlation matrices, is the same and sparse for all subjects, although the strength of non-zero connections may vary. For each subject we model the parcel boundaries using a Potts MRF model. Correlations between parcels are governed by a MVG likelihood, computed on the average time series, with the inverse covariance (partial correlations) regularized using a group sparsity penalty that encourages each subject to have the same set of nonzero elements. We alternate between optimization with respect to the labeling of each subject using a spectral clustering algorithm [7], and estimation of their group partial correlation matrices using an ADMM approach [8] as illustrated in Fig. 1.
We evaluated performance using the minimally pre-processed resting fMRI data from 40 unrelated subjects available from the Human Connectome Project [9]. We used regions extracted from task-based activation maps to assess performance of the MAP approach in comparison to (a) a group parcellation method (Atlas) based on anatomical alignment, and individual parcellation by (b) shrinkage estimation [3] and (c) spectral clustering [4]. For the same 40 subjects, we used task activation maps for each subject to compute concordance with the resting fMRI parcellations [10].
Results:
Fig. 2 illustrates results for faces-vs-shapes task fMRI activation (color map) overlaid with resting fMRI parcellation boundaries (black lines). Note that MAP parcellation is more consistent with the task activation map in visual cortex and para-hippocampal gyrus. In fig. 3, we quantify the consistency between resting fMRI parcellation and task fMRI activation labels, generated from 5 motor task pairs across population and levels of parcellation M={60,80,100,120}, and compared across different approaches. The MAP approach achieves a significantly (p<0.01) higher concordance as compared to all other methods for all parcellation levels. Fig. 4 shows the parcellation results for 60 parcels for three subjects.
Conclusions:
Our MAP approach combines individual functional parcellation with a group sparsity constraint to obtain resting parcellations that consistently produce improved consistency with task-based activation maps compared to other methods. The MAP approach could be used as the basis for studying individual network variability across subjects, or for defining functional regions in individuals for subsequent analysis of event-related activity.
Imaging Methods:
BOLD fMRI
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 2
Segmentation and Parcellation 1
Task-Independent and Resting-State Analysis
Keywords:
FUNCTIONAL MRI
Segmentation
1|2Indicates the priority used for review
Would you accept an oral presentation if your abstract is selected for an oral session?
Yes
I would be willing to discuss my abstract with members of the press should my abstract be marked newsworthy:
Yes
Please indicate below if your study was a "resting state" or "task-activation” study.
Resting state
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Healthy subjects
Internal Review Board (IRB) or Animal Use and Care Committee (AUCC) Approval. Please indicate approval below. Please note: Failure to have IRB or AUCC approval, if applicable will lead to automatic rejection of abstract.
Yes, I have IRB or AUCC approval
Please indicate which methods were used in your research:
Functional MRI
For human MRI, what field strength scanner do you use?
3.0T
Which processing packages did you use for your study?
FSL
Free Surfer
Other, Please list
-
BrainSuite
Provide references in author date format
[1] Eickhoff, S. B., Thirion, B., Varoquaux, G. and Bzdok, D. (2015), Connectivity-based parcellation: Critique and implications. Hum. Brain Mapp., 36: 4771–4792.
[2] Thirion, B., Varoquaux, G., Dohmatob, E., & Poline, J.-B. (2014). Which fMRI clustering gives good brain parcellations? Frontiers in Neuroscience, 8, 167.
[3] Chong, M., Joshi, A., Haldar, J., DuPre, E., Luh, W., Shattuck, D., Spreng, N. & Leahy, R. (2015) A Group Approach to Functional Cortical Parcellation from Resting-State fMRI. Annual Conference of OHBM, Honolulu, Hawaii
[4] Shou, H., Eloyan, A., Nebel, M. B., Mejia, A., Pekar, J. J., Mostofsky, S., ... & Crainiceanu, C. M. (2014). Shrinkage prediction of seed-voxel brain connectivity using resting state fMRI. NeuroImage, 102, 938-944.
[5] Craddock, R. C., James, G. A., Holtzheimer, P. E., Hu, X. P., & Mayberg, H. S. (2012). A whole brain fMRI atlas generated via spatially constrained spectral clustering. Human brain mapping, 33(8), 1914-1928.
[6] Sporns, O., Tononi, G., & Kötter, R. (2005). The human connectome: a structural description of the human brain. PLoS Comput Biol, 1(4), e42.
[7] Shi, J., & Malik, J. (2000). Normalized cuts and image segmentation. Pattern Analysis and Machine Intelligence, IEEE Transactions on, 22(8), 888-905.
[8] Ng, B., Varoquaux, G., Poline, J. B., & Thirion, B. (2012). A novel sparse graphical approach for multimodal brain connectivity inference. In Medical Image Computing and Computer-Assisted Intervention–MICCAI 2012 (pp. 707-714). Springer Berlin Heidelberg.
[9] Van Essen, D. C., Smith, S. M., Barch, D. M., Behrens, T. E., Yacoub, E., Ugurbil, K., & WU-Minn HCP Consortium. (2013). The WU-Minn human connectome project: an overview. Neuroimage, 80, 62-79.
[10] Barch, D. M., Burgess, G. C., Harms, M. P., Petersen, S. E., Schlaggar, B. L., Corbetta, M., ... & Nolan, D. (2013). Function in the human connectome: task-fMRI and individual differences in behavior. Neuroimage, 80, 169-189.