Questions on FC analysis | The R-fMRI Network

Submitted by merrf on Sun, 02/05/2012 - 17:54

Dear Dr. YAN Chao-Gan I've successfully performed seed-based (hippocampus, PAG, and so on) functional connectivity by using DPARSF with default settings. Now I'm trying to compare FC differences between patients and control subjects. I have 2 questions. Q1) When I perform unpaired t-test of second-level analysis of zFC* maps (control 46, patients 43) using spm8, it seems that results of only 'positive' correlation maps but not 'negative' correlation maps does appear in spm results. Is it right? If so, I'm very satisfied with the results because I do not want to analyze negative correlation maps. Q2) I've found several significant clusters that were cluster-level corrected for multiple comparisons using FWE. I want to correlate these clusters with severeal clinical parameters (eg, disease duration, seizure frequency in given years, neuro-cognitive variables..). I've also extensively searched many published papers in which such correlation analyses were used, however no step-by-step methodology was not described in detail. How can I do this?? Thank you in advance for your help. Kim

Re

Hi! 1. No. You can't see the negative ones just because SPM performs ONE-TAILED analysis. When you input contrast as -1 1, then you will see the other direction. If you have a hypothesis that you only interested in positive ones, they you can leave the negative ones. But you can not say there is no negative ones, because you need to see the other direction -1 1. 2. Try REST->Statistical Analysis->REST Correlation Analysis. This will correlate each voxel's measure and the clinical parameters. Or if you want have a scatter plot, you can use REST->Utilities->Extract ROI Signals to extract the measures of the significant ROI, and they do scatter plot in SPSS or MATLAB. Best, Chao-Gan

Re:

Dear Chao-Gan Thank you very much for the quick answer. Apology for my incomplete description regarding Q1. As I understand, outcome of each FC analysis on REST slice viewer or xjview shows not only positive (red-yellow) but also negative (blue-green) correlation maps (individual level). Right? When I do one-sample t-test in 40 control subjects using REST statistical analysis tool, both positive and negative correlation maps appeared on slice viewer. However, when I do same one-sample t-test using same dataset (40 zFC*.nii) and spm8 second-level analysis, it seems to me that there are no negative correlation maps. How is it possible? Does spm calculate or show only positive correlation maps? Another query here. And sorry for bothering you again. I've found that, in many published articles, the reserchers used 'functinal connectivity strength' between predefined 2 seeds and correlated these values with multitude of clinical or cognitive measures. How can I get these FC strength values? Are these values same as what you mentioned (extract ROI signals from each subjects)? Thanks again.

Re

Hi! "As I understand, outcome of each FC analysis on REST slice viewer or xjview shows not only positive (red-yellow) but also negative (blue-green) correlation maps (individual level). Right?" Right. "When I do one-sample t-test in 40 control subjects using REST statistical analysis tool, both positive and negative correlation maps appeared on slice viewer." Right. "However, when I do same one-sample t-test using same dataset (40 zFC*.nii) and spm8 second-level analysis, it seems to me that there are no negative correlation maps. How is it possible? Does spm calculate or show only positive correlation maps?" Actually, SPM will calculate both negative and positive correlations (See SPM_00001T.img). However, it will only perform multiple comparison on the positive, and will only show the positive ones in the figure. "the reserchers used 'functinal connectivity strength' between predefined 2 seeds and correlated these values with multitude of clinical or cognitive measures." In this case, just do ROI-WISE functional connectivity with the two predefined seeds, then you will get the correlation r and fisher r-to-z transformed z scores. Use SPSS or MATLAB perform a correlation analysis on the z scores and clinical measures would be fine.

Forums

Discuss