Hello Christian and Ben,
Thank you very much for your prompt and kind reply. Please let me explain a little bit more the analyses I need to do. I am working on resting-state data. I have to run three different analysis applying seed-based correlation approach on the following cohorts:
1) The first one is a single group analysis of subjects with depression.
2) the second one is a group comparison between patients with PTSD and controls
3) and the last one is the patients with PTSD pre and post treatment.
I already have three ROI masks for each hemisphere, That is, six ROI seed masks as a total. I have the coordinates I am interested on, So I have created those spheric masks using fslmaths based on those coordinates. I am going to use those six ROI masks doing the three seed-based correlation analyses described above. I think if I want to run these analysis with fsl_glm and randomise (dual regression) I need to follow the following steps for each analysis:
*** I already did all the preprocessing and registration in FEAT and I ran featregapply to get the filtered_func_data.nii.gz images into MNI space
***** I have ran FAST in the MNI space images ( filtered_func_data_MNI.nii.gz) getting the white matter (WM) , and CSF masks for each subject.
*******I did a spatial regression of white matter (WM) , and CSF masks onto the EPI filtered_func_data_MNI.nii.gz in standard space, using fsl_glm
*****Now, I think I would have to add the motion regressors and the nuisance .txt files (WM and CSF) to the 2nd stage of dual regression.
*****And feed my six standard space seed ROI masks concatenated as a "group map" into dual-regression script
*******Then, I would have to run randomise on the spatial maps from the second stage of dual regression using the appropriate design and contrast matrix (one sample t-test for my first analysis, unpaired t-test for my second analysis and paired t-test for my third analysis).
So, after read the FSl archives and found that is possible to run that kind the analysis with fsl_glm (Dual regression) and randomise (as suggested by Dr. Smith and Dr. Beckmann in some of their posts) I was wondering if with my data I could do the seed-based analyses as I have described above. So, I greatly appreciated your insights and suggestions on this matter.
Thank you VERY much in advance for your advise!
Lorena
__________________
Lorena Jimenez-Castro, MD
Postdoctoral Fellow
UTHSCSA
________________________________
Subject: Re: correlation analysis
From:"Christian F. Beckmann" <[log in to unmask]>
Reply-To:FSL - FMRIB's Software Library <[log in to unmask]>
Date: Tue, 22 May 2012 06:48:28 +0200
Reply
Hi
Both of Ben's answers are pretty spot on ;) The approaches differ in terms of ease of use etc - so 'best option' depends on your circumstances, e.g. do you want to use a fixed MNI coordinate or have you got hand-drawn masks etc. Unless you go into more detail in your question it's hard to recommend anything more specific. Steve's answer relates to the case of having a single fixed ROI you want to apply across all subjects. If this is not the case in your study then no, you should not be using this.
hth
Christian
On 22 May 2012, at 03:44, Lorena Jimenez-Castro wrote:
> Hello Benjamin and thanks a lot for your answer, I appreciate it!
>
> Though, I think you've answered a slightly different question. Perhaps I didn't make my question clear. I fact what I asked was in regards to a seed-based correlation analysis. That is, I already have the ROIs and I want to use them in the seed_based analysis, so I was wondering (which were my questions)
>
> 1) what is the best option for doing a seed-based correlation analysis of fMRI data in FSL, using FEAT (similar than a PPI analysis as in Castellanos at al., 2008 cited in my previous post) or using dual regression script as have suggested professor Stephen Smith in the following post:
>
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1112&L=fsl&P=R46634&1=fsl&9=A&I=-3&J=on&X=4FDE1362374539B9A8&Y=lojicas%40yahoo.com&d=No+Match%3BMatch%3BMatches&z=4
>
>
> 2)Why one analysis could be better than the other?
>>
>> I know that running a seed-based analysis in FEAT involves several steps, but in addition to that, Is there any other reason to choose dual_regression instead of running the analysis in FEAT?
>
>
> I would greatly appreciate any input about these questions,
>
>
> Lorena
> __________________
> Lorena Jimenez-Castro, MD
> Postdoctoral Fellow
> UTHSCSA
>
>
> ______________________
> Subject: Re: correlation analysis
>
> From: Benjamin Kay <[log in to unmask]>
>
> Reply-To: FSL - FMRIB's Software Library <[log in to unmask]>
>
> Date: Mon, 21 May 2012 20:43:38 -0400
>
>
> Reply
>
> Someone else may clarify this, but I think that dual regression is a tool to use with ICA, ICA being an alternative to seed-based voxel correlation. Seed based correlation is a farily straightfoward way to find regions connected with a seed of interest. ICA doesn't require you to specify a seed region, so it's advantageous when you don't really know what region is serving as the connectivity "hub" of your resting state network. When it comes to canonical networks like the default mode network, I don't think it makes *that* much of a difference since you're going to find the network either way.
>
> On Monday, May 21, 2012 10:17:14 you wrote:
>> Dear FSL users (Especially professor Christian Beckmann),
>>
>> I sent the message below on May 8, 2012 9:54 AM. I have not received any answer yet; I would appreciate very much any input on this matter. My questions are:
>>
>> I would like to know, 1) what is the best option for doing a seed-based correlation analysis of fMRI data in FSL, using FEAT (as implemented in "Cingulate-precuneus interactions: a new locus of dysfunction in adult attention-deficit/hyperactivity disorder.
>> Biol Psychiatry. 2008 Feb 1;63(3):332-7. Epub 2007 Sep 21.Castellanos FX et al.,") or running the dual-regression script ( applying fsl_glm to get the nuisance covariate time-courses and then randomize)? and
>> 2) Why one analysis is better than the other?
>>
>> I know that running a seed-based analysis in FEAT involves several steps, but in addition to that, Is there any other reason to choose dual_regression instead of running the analysis in FEAT?
>>
>> Best
>> Lorena
>>
>>
>> __________________
>> Lorena Jimenez-Castro, MD
>> Postdoctoral Fellow
>> UTHSCSA
>>
>>
>>
>>
>>
>>
>> ---------------------------------
>> Print - Close Window
>> Subject: [FSL] Seed-based analysis
>> From: Lorena Jimenez-Castro ([log in to unmask])
>> To: [log in to unmask];
>> Date: Tuesday, May 8, 2012 9:54 AM
>>
>>
>> Hello experts,
>>
>> I would like to know, what is the best option for doing a seed-based correlation analysis of fMRI data in FSL, using FEAT or running the dual-regression script ( applying fsl_glm to get the nuisance covariate time-courses and then randomize)? and
>> Why one analysis is better than the other?
>>
>> I know that running a seed-based analysis in FEAT involves several steps, but in addition to that, Is there any other reason to choose dual_regression instead of running the analysis in FEAT?
>>
>> Thank you very much
>>
>>
>> Lorena Jimenez-Castro, MD
|