 |
 |
Dear MikeBasically I have already implemented the method you suggested in my own dataset. I got modified T score for each resample but I want to perform some "conventional GRF-style multiple comparison correction" for each modified T score maps. Could you provide me some information of how to perform multiple comparison correction using FSL command line tools ?In order to compare the performance (type I error rate) to modified T score that you suggested in previous reply, I also calculate traditional Z-score which I post in the original question.
BestPaul
Date: Mon, 27 Jan 2014 18:47:49 -0500
From: [log in to unmask]
Subject: Re: [FSL] How to calculate voxel/cluster wised corrected P value with FSL command line tools ?
To: [log in to unmask]Hi Paul,Hopefully someone better than me with stats will chime in on this, but it sounds a little fishy to me. I'm not sure such an approach is valid, since all your individual z-score calculations from boot-strapping like that are not really independent.You could take a look at: http://homepages.abdn.ac.uk/j.crawford/pages/dept/SingleCaseMethodology.htm, which has some nice discussion about comparing single subject data to small control groups. However, it's for conventional data, not image data. Bascially, though, they give a slightly modified t-Test approach.Once you have a t-stat image and degrees of freedom, you could maybe just do conventional GRF-style inference. I'm not sure permutation approaches would work well, since you only have the one subject in group 1.Best,MikeOn Mon, Jan 27, 2014 at 5:58 PM, Paul Chou <[log in to unmask]> wrote:
Dear all FSL experts
I have a statistical question about how to calculate corrected p value (voxel-wised and cluster-wised corrected p value) using FSL command line tool. The following thing is what I want to do and wish some of you could help me to figure out this problem.
Rationale : I want to calculate the type I error rate of “one vs many statistical test (one patient vs many healthy controls)” of gray matter volume. I use Z-score to achieve this goal.
Step 1: My dataset consist of 50 healthy subjects. I random select 26 subjects from this dataset and perform 100 times resample. For each resample, the one normal control were treated as patient and the other 25 subjects were served as normal controls. I calculate the Z score for every resample using the following formula [(patient GMV - healthy control mean GMV) / standard deviation of healthy controls GMV].
Step 2: After Z-score calculation, I need to set the significant level (p value) of this test to identify which brain area have significant difference between the patient and the normal controls (something like P < 0.05 = Z > 1.73). The area with significant statistical difference was related to the type I error rate. I think I need to consider the multiple comparison correction issue since the origin of this analysis is voxel-wised whole brain analysis , but I don’t know how to calculate the voxel-wised and cluster-wised FWE corrected P value using the FSL command line tool. After quick searching the FSL mail list, I think “fsl_glm” (which calculate the residual images) , “smoothest” (which calculate the data smoothness) may be needed for my demand, is it correct ? If yes, how to use these tools to calculate the voxel / cluster wised FWE corrected P value?
Sorry for this long question, I try to describe it more clearly and hope to hear feedbacks from you
Thanks
Best
Paul
--Michael G. Dwyer, Ph.D.Assistant Professor of Neurology
Director of Technical Imaging Development
Buffalo Neuroimaging Analysis Center