A Supervoxel-Based Method for Groupwise Whole Brain Parcellation with Resting-State fMRI Data

Abstract

Node definition is a very important issue in human brain network analysis and functional connectivity studies. Typically, the atlases generated from meta-analysis, random criteria, and structural criteria are utilized as nodes in related applications. However, these atlases are not originally designed for such purposes and may not be suitable. In this study, we combined normalized cut (Ncut) and a supervoxel method called simple linear iterative clustering (SLIC) to parcellate whole brain resting-state fMRI data in order to generate appropriate brain atlases. Specifically, Ncut was employed to extract features from connectivity matrices, and then SLIC was applied on the extracted features to generate parcellations. To obtain group level parcellations, two approaches named mean SLIC and two-level SLIC were proposed. The cluster number varied in a wide range in order to generate parcellations with multiple granularities. The two SLIC approaches were compared with three state-of-the-art approaches under different evaluation metrics, which include spatial contiguity, functional homogeneity, and reproducibility. Both the group-to-group reproducibility and the group-to-subject reproducibility were evaluated in our study. The experimental results showed that the proposed approaches obtained relatively good overall clustering performances in different conditions that included different weighting functions, different sparsifying schemes, and several confounding factors. Therefore, the generated atlases are appropriate to be utilized as nodes for network analysis. The generated atlases and major source codes of this study have been made publicly available at http://www.nitrc.org/projects/slic/.

Functional density and edge maps: Characterizing functional architecture in individuals and improving cross-subject registration

“…The correlation between the average functional density maps of the two independent groups was 0.997, and the correlation between the two functional edge maps was 0.987, indicating high reproducibility of the proposed functional maps at the population level. The high reproducibility of these functional maps also ensured the robustness group-level parcellations: when the total number of parcels was set to 150 (Figure 6, bottom), a Dice coefficient of 0.916 was achieved between the parcellations derived from the two independent groups, which is comparable to those of the state-of-the-art parcellation results (Shen et al, 2013; Parisot et al, 2016; Wang and Wang, 2016). …”

Regularized-Ncut: Robust and homogeneous functional parcellation of neonate and adult brain networks

“…The tuning parameter r with testing range value from 2 to 8 can be optimized with SI values obtained from parcellations of real human rs-fMRI data, as demonstrated in Supp Fig 1. Thirdly, selection of optimal number of brain parcellations remains to be an open problem for developing parcellation algorithms (e.g. Power et al, 2011;Yeo et al, 2011;Craddock et al, 2012;Thirion et al, 2014;Wang and Wang, 2016). Establishing an appropriate cost function integrating different measurements such as SI and Dunn index (Dunn, 1974) may shed light on determining optimal parcellation number.…”

“…Cauda et al, 2010Cauda et al, , 2011, hierarchical clustering (Mumford et al, 2010;Bellect et al, 2006), spectral clustering (e.g. Van den Heuvel et al, 2008;Shen et al, 2010Shen et al, , 2013Craddock et al, 2012;Cheng et al, 2014;Parisot et al, 2016;Wang et al, 2016;Shi et al, 2017), Gaussian Mixture Models (e.g. Yeo et al, 2011) and region growing (e.g.…”

Generation of Individual Whole-Brain Atlases With Resting-State fMRI Data Using Simultaneous Graph Computation and Parcellation

“…It has also been widely applied in three-dimensional (3D) image segmentation tasks (Lucchi et al, 2012 ; Menze et al, 2015 ). We have previously used SLIC to generate brain atlases (Wang and Wang, 2016 ; Wang et al, 2016 ). Both previous studies treated supervoxels as clusters in brain atlases.…”

“…In the normalized cuts (Ncut) approach (Craddock et al, 2012 ), spatial structure is introduced by the spatial constraint in weight definition. In the SLIC approach (Wang and Wang, 2016 ), spatial structures are introduced by initializing an ideal geometric pattern, integrating the spatial distance into the unified distance, and searching in a local space. As Wang and Wang ( 2016 ) have shown, incorporating suitable spatial structures in whole-brain parcellation approaches is quite necessary to guarantee the spatial contiguity of the resultant clusters.…”