between a restricted ROI x gray matter mask region for Degree Centrality

Submitted by jwhnavy on Sun, 02/23/2014 - 07:47

Hi REST experts,

I'd like to perform Degree Centrality analysis within restricted ROI (e.g., only PFC region).
To do so, I do make a correlation matrix between voxels in the restricted ROI and voxels in whole gray matter mask (or default whole brain mask) before applying threshodling (r>0.025) and counting or averaging.

1) Can I perform such Degree Centrality analysis (restricted ROI x whole brain mask) using REST v1.8?
If yes, How can I do this?

2-1) I know that there is still a debate about negative correlation in resting-state functional connectivity in terms of global singal regression. So, I used r>0 or r>0.025 as a thresholding value for Degree Centrality analysis according to Buckner et al. (2009). I understand that means there is no negative correlation values in a correlation matrix between voxels of whole brain mask. As far as I understand, all values of PositiveBinarizeSumBrainMap must be positive values because these values are the number counting voxels that have values more than r>0.025. However, I found negative values on the "PositiveBinarizeSumBrainMap". Why are there negative values on the map?

2-2) After completing z-normalization (with mean and S.D. values for each subject), there are negative values. As far as I know, statistical group analysis in seed-based functioanl connectivity use a binary mask for positive correlation network to remove negative correlation values. When I perform between-group comparison analysis with DC maps, I also use a binary mask for only positive DC values.

There are two Degree Centrality analysis methods; weighted DC (Cole et al., Neuroimage 2010; 39:3132-48) and binary DC (Buckner et al., 2009). According to Cole et al. (2010), they made a correlation matrix between voxels in whole gray matter mask and then performed Fisher's r-to-z transformation. After that, they did averaged values in the correlation matrix.
3) Does REST v1.8 perform Fisher's r-to-z transformation during process of "DC Positive Weighted Sum Brain map" before summing or averaging values? So, I think that the "DC PositiveWeightedSumBrainMap" methods is a little different from the method developed by Cole et al. (2010).

4) What do you think about applying spatial smoothing (e.g., 4~6mm FWHM) to "DC PositiveWeightedSumBrainMap" and "DC PositiveBinarizeSumBrainMap" rather than z-normalization (with mean and S.D.) before statistic group analysis?

Thank you in advance

Submitted by ZHANG_RESTadmin on Mon, 02/24/2014 - 14:36

 thanks for interesting with REST DC calculation.

1. If you want calculate DC in a specific mask, just simply input your mask file in the DC GUI in a place "input mask file". Then all claculation should be done in this mask.

2. Yes your understanding is right. There should be no neg value in PositiveBinarizeSumBrainMap. If one voxel have no significant positive connection with other voxels, it should have 1 connection with r = 1 (this connection is self-connection), however, we considered this situation and further subtract this from the total count. Therefore, in this case, the count should be exactly zero. As you asked, why there was negative values in the map, I really don't know.  One possibility is, since you use a mask (whether the whole brain or a single ROI mask), the value outside of the mask should be zero and those in side of it should be positive.  However, at the voxels near the border of hte mask, the interpolation applied to the DC image when you visualize it in some toolboxes (e.g., MRIcron) could make negative values. However, these neg values should be quite small.  To check if this happened, you can load the DC image into matlab, using "rest_readNiftiImage" function, then using matlab function "find" to find any voxel < 0.    

Another case is, when you choose z transform DC map, the DC map will be subtracted by mean and divided by std. This will also cause negatives. 

3. See my answer above.

4. Both ways are reasonable, I mean, smoothing before or after statstical anaysis. However, there is no agreement. 


Submitted by jwhnavy on Mon, 02/24/2014 - 23:11

Thank you very much for your quick and helpful answers.

1) In term of answer #1, I understand if I perform DC analysis with "input mask file" (e.g., the PFC mask), this DC analysis will results in DC values from "the correlation matrix between each PFC voxels" rather than DC values from "the correlation matrix between each PFC voxels (i.e., region-of-interest) and all other PFC voxels (i.e., voxels within whole brain mask except for the PFC)". I'd like to calculate the DC values between each PFC voxels and all other PFC voxels. How can I perform this analysis using REST v1.8? Does it just apply the PFC mask as "input mask file"?

2) Can I use "zDegreeCentrality_PositiveBinarizeSumBrainMap" or "zDegreeCentrality_PositiveWeightedSumBrainMap" image data obtained from REST v1.8 or DPARSFA to perform one-sample (or two sample) t-tests? Should I smooth those data to perform such statistic analyses?

3) Is it possible to calculate the weighted and non-thresholded DC values (i.e., the avearge value for each row value on the correlation matrix for each subject), suggested by Cole et al. (Cole et al., Neuroimage 2010;39:3132-48 or Cole et al., Biol Psychiatr 2011;70:43-50) using REST v1.8?

Thank you in advance.