Yes, indeed: I have seen that the <design.mat> and <design.con> files are part of the second step and that I can use Glm to create these files.
I am, however, sorry to have to say that I'm still in the dark as to HOW I should use Glm (or Glm_gui) to create the <design.mat> and <design.con> files.
So I already ran:
melodic -i list_of_six_4D_fMRI_datasets.txt --report -v --tr=5 --Oall -o output.ica -a concat
and my goal is to use the dual-regression script to regress the group-level maps into the individual runs original individual runs, and from there can go on to cleanup the individual runs.
I would appreciate it immensely if I can get some instructions on how to do this. I would prefer to remain working on my linux box and know that I can call the Glm gui: fsl/current/bin/Glm
Thanks!
Koene
On Fri, 9 Mar 2012 08:06:14 +0000, Stephen Smith <[log in to unmask]> wrote:
>Hi - for the design.mat and design.con files, this is the second step on the dualreg web page - for example, use the Glm (Glm_gui on Mac) GUI to create these.
>Cheers.
>
>
>On 8 Mar 2012, at 19:30, Koene Van Dijk wrote:
>
>> Hi,
>>
>> I've tried to follow Matt's suggestion: first concatenate my 6 runs that I have of the same subject and feed that as a single 4D file into melodic. I used fslmerge -t (etc) and then ran melodic on that single file but got strange results: I got --using the automated dimension estimation-- 231 components instead of the nice 25 that I got when I fed in the 6 runs separately using a text file with the "concat" or "tica" options.
>>
>> Alternatively, I would not mind if I can just keep working in my stream where I feed in 6 runs separately in with the "concat" flag, and go for the dual regression approach as Steve suggested (below).
>>
>> Steve wrote: "use dual-regression to regress the group-level maps into the individual subject's original datasets"
>>
>> I have not yet used the dual regression script. I read this:
>> ---------------------
>> 1. Run Melodic on your group data in Concat-ICA mode ("Multi-session temporal concatenation"). Let's assume that the group-level output directory from this is called groupICA.gica
>>
>> 2. Use Glm (or any other method) to create your multi-subject design matrix and contrast files.
>>
>> 3. Run dual_regression. Just type the script name to get the usage - should be mostly self-explanatory
>>
>> (source: http://www.fmrib.ox.ac.uk/analysis/dualreg/)
>> ---------------------
>>
>> Step 1 is done. I believe I get step 3 if I only have <design.mat> and <design.con>.
>>
>> So my question now is:
>> Where do I get, or how do I make, the <design.mat> and <design.con> files?
>>
>> (after that I will use the "dr_stage1_subject000*.txt files as the -d input to fsl_regfilt" as I understood from the message by Steve Mayhew)
>>
>> Thanks again for any suggestions!
>>
>> -Koene
>>
>> PS: Steve: I assumed that when I fed several runs into melodic using a text file and I did not specify the "approach", the program would take tica by default. I've tried both tica and concat now, and they give quite similar results -- but I will continue with the "concat" version as suggested.
>>
>>
>>
>>
>>
>> On Thu, 8 Mar 2012 14:25:43 +0000, Stephen Smith <[log in to unmask]> wrote:
>>
>>> Hi - that's nearly what I was going to start by saying: Koene's first call to melodic looks weird - you pass in a text file of multiple inputs, but don't specify whether to use the concat or tensor group-level approach - so I'm not even sure what melodic will do in this case. I would recommend just using the -a concat option, which is nearly what Matt suggested, but not quite.
>>>
>>> THEN, you use dual-regression regress the group-level maps into the individual subject's original datasets, and from there can go on to (eg) cleanup those individual datasets on the basis of the outputs of that. http://www.fmrib.ox.ac.uk/analysis/dualreg/
>>>
>>> Cheers.
>>>
>>>
>>> On 7 Mar 2012, at 14:49, Matt Glasser wrote:
>>>
>>>> It might be better to concatenate the data first in a single 4D file before
>>>> running melodic, as this isn't really "group" ICA. Then fsl_regfilt would
>>>> run properly on the output melodic_mix when you specify the components.
>>>>
>>>> Peace,
>>>>
>>>> Matt.
>>>>
>>>> -----Original Message-----
>>>> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
>>>> Of Koene Van Dijk
>>>> Sent: Tuesday, March 06, 2012 9:24 PM
>>>> To: [log in to unmask]
>>>> Subject: [FSL] UPDATE: Applying fsl_regfilt after melodic to denoise the
>>>> several 4D datasets.
>>>>
>>>> In the meantime I have learned a little bit more about my melodic output.
>>>> Let me update my questions that I submitted yesterday:
>>>>
>>>> If I run this command:
>>>> melodic -i list_of_4D_fMRI_datasets.txt --report -v --tr=5 --Oall -o
>>>> output.ica
>>>>
>>>> ...on 1 BOLD run of 90 volumes I get a file for the first component with 1
>>>> column with 90 entries:
>>>> ./output.ica/report/t1.txt
>>>>
>>>> If I run the same command but include 2 BOLD runs with each 90 volumes (i.e.
>>>> 180 volumes) I will get a total of 3 columns in the "t1.txt" file. I see
>>>> that the number of columns is always: number of bold runs + 1.
>>>>
>>>> I guess that there is one column for the time course of the respective
>>>> component for each run + one summary measure or so. I tried it with flag
>>>> "-a" set to default (=tica) and with concat which gave the same file
>>>> structure.
>>>>
>>>> Can somebody tell me which column is which?
>>>>
>>>> And what the +1 column represents?
>>>>
>>>>
>>>> Thanks,
>>>>
>>>> Koene Van Dijk
>>>>
>>>> PS: Here are the first 10 lines of such a file when I used two bold runs as
>>>> input (using tica)
>>>>
>>>> head /output.ica/report/t1.txt
>>>> 1.86719 1.90851 1.75934
>>>> 1.72553 1.74516 1.6743
>>>> 2.00228 1.65566 2.90701
>>>> 1.71611 1.65537 1.87466
>>>> 1.68615 1.71394 1.61362
>>>> 1.61661 1.36676 2.26876
>>>> 1.43963 1.50949 1.2573
>>>> 1.88293 1.77197 2.17257
>>>> 1.3938 1.35532 1.49425
>>>> 1.3309 1.58513 0.667326
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> On Mon, 5 Mar 2012 17:18:27 +0000, Koene Van Dijk <[log in to unmask]>
>>>> wrote:
>>>>
>>>>> Dear all,
>>>>>
>>>>> I am using melodic on a single subject. I have multiple runs of resting
>>>> state fMRI: 6 runs of 90 volumes each, i.e 540 volumes in total. I have
>>>> already performed slice-timing correction and motion correction (all 539
>>>> volumes were registered to the first volume of the first run).
>>>>>
>>>>> This is what I ran on my linux box using fsl 4.1.9 (64-bit):
>>>>> melodic -i list_of_five_4D_fMRI_datasets.txt --report -v --tr=5 --Oall -o
>>>> output.ica
>>>>>
>>>>> It gives 25 components with beautiful maps. I now would like to remove some
>>>> obvious noise components from the 4D data set and save the residual as new
>>>> 4D dataset for other analyses.
>>>>>
>>>>> On the following website I read about using fsl_regfilt:
>>>>> http://www.fmrib.ox.ac.uk/fslcourse/lectures/practicals/melodic/index.htm#d
>>>> enoising
>>>>>
>>>>> And I tried this to remove components 1,2 and 3:
>>>>> fsl_regfilt -i list_of_five_4D_fMRI_datasets.txt -d
>>>> ./output.ica/melodic_mix -o denoise_ICAfiltered -f "1,2,3"
>>>>>
>>>>> This does not work: "ERROR: cannot read input image".
>>>>>
>>>>> So you can see that I entered my "list_of_five_4D_fMRI_datasets.txt" that I
>>>> also used for the initial melodic command. I guess fsl_regfilt is not
>>>> designed to take a list as input files, as melodic is.
>>>>>
>>>>> My first question is:
>>>>> 1) Would anyone have a suggestion for how I can remove some components from
>>>> this dataset using fsl_regfilt?
>>>>>
>>>>>
>>>>> If I would like to regress out the components with another tool (any glm
>>>> tool that allows me to save the 6 residual 4D fMRI runs), I would need to
>>>> input the time course of the components. For example component 1, I found a
>>>> text file here:
>>>>> ./output.ica/report/t1.txt
>>>>>
>>>>> This file contains 90 rows and 7 columns. This surprised me because I was
>>>> expecting to find 90 x 6 (=540) and not 90 x 7 (=630) values to represent
>>>> the time course of component 1.
>>>>>
>>>>> So my next questions are:
>>>>>
>>>>> 2. Why are there 90 x 7 columns and not 90 x 6?
>>>>>
>>>>> 3. Which time course would you recommend I take to use as regressor of
>>>> no-interest if I would want to regress out a noise component?
>>>>>
>>>>>
>>>>> Thanks a lot for any advice!
>>>>>
>>>>> Best regards,
>>>>>
>>>>> Koene
>>>>
>>>
>>>
>>> ---------------------------------------------------------------------------
>>> Stephen M. Smith, Professor of Biomedical Engineering
>>> Associate Director, Oxford University FMRIB Centre
>>>
>>> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
>>> +44 (0) 1865 222726 (fax 222717)
>>> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>>> ---------------------------------------------------------------------------
>>>
>>>
>>>
>>>
>>
>
>
>---------------------------------------------------------------------------
>Stephen M. Smith, Professor of Biomedical Engineering
>Associate Director, Oxford University FMRIB Centre
>
>FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
>+44 (0) 1865 222726 (fax 222717)
>[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>---------------------------------------------------------------------------
>
>
>
>
|