Hi Tulu - thanks for sending out the data. Yes, there was only one problem
with the analysis - that was that your input data is of type
8-bit-unsigned-char, which the filtering in FEAT doesn't like, as
intensities always get rescaled to have grand mean of 10000. To fix this
is easy - inside fsl/tcl/feat.tcl just replace the line
set thecommand "${FSLDIR}/bin/ip ${funcdata} filtered_func_data $fmri(brain_thresh)"
with the following lines:
set DATATYPE ""
if { [ exec $FSLDIR/bin/avwval $funcdata datatype ] != 4 } {
set DATATYPE "_32R"
}
set thecommand "${FSLDIR}/bin/ip$DATATYPE ${funcdata} filtered_func_data $fmri(brain_thresh)"
And then the results look very nice, and you can put back in a more normal
threshold and still find interesting activation. FLIRT registration to the
(MR) standard-space image even works!
Thanks, Steve.
On Thu, 17 Oct 2002, tulu wrote:
> Hi Smith and FSL experts,
> I setup the model and use the parameter according to your suggestion. I
> set the thresholding as uncorrected voxel P in 0.1 ( it is very large).
> Then I run the FEAT. No voxel is above the threshold. It is very strange
> that if I use the SPM to perform the same study, I can set the p value
> in 0.05 or below and get a reasonable result. Is there anything wrong ?
> Besides, if I have the variance for each image, such as the cognitive
> score in controls and patients, how to input the variance ?
> Thank for your help
> JLH
>
> -----Original Message-----
> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On
> Behalf Of Stephen Smith
> Sent: Thursday, October 17, 2002 6:27 PM
> To: [log in to unmask]
> Subject: Re: [FSL] parameter about the group analysis by FEAT
>
> Hi - so it seems like you have only one image per subject, and I will
> assume that they are all registered together (if not - you need to do
> this). It also seems like you DON'T have variance images for the
> subjects.
>
> Therefore there is not too much advantage in using FLAME (the advanced
> estimation for group-stats in FEAT) over ordinary-least-squares (ie what
> you get if you turned off FILM for first-level anllayses or turn off
> FLAME
> for higher-level analyses). The only time where you MIGHT get an
> improvement with using FLAME is if your two groups have very different
> variances. (Note for other people - if you do all your analysis at
> first-
> and second-level using FEAT then of course, FEAT will pass up parameter
> estimates AND variances to the second-level, so FLAME is definitely of
> value).
>
> So - in order to input your data into FEAT you probably may as well run
> a
> "first-level" analysis, according to Joe's suggestions. Start by
> concatenating the two 4D files into one by:
> avwmerge -t allsubjects.hdr controls.hdr patients.hdr
>
> And then do a first-level analysis with a single EV of 1's and -1's. In
> your case you want TR=1 skip=0 OFF=43 ON=7 phase=0 conv=NO
> temporalfiltering=NO temporalderivative=NO
>
> contrast 1 (patients-controls) = [1]
> contrast 2 (controls-patients) = [-1]
>
> hope this works ok!
>
> Steve.
>
>
>
> On Thu, 17 Oct 2002, tulu wrote:
>
> > Dear FSL experts,
> > Thanks Joe and Smith for your help. I only have the normalized raw
> data
> > from controls and disease. So I think I don't have first level time
> > series PET data. Dr. Joe, you understand correctly. However I still
> have
> > the problem in the stats portion. What is the basic shape I should
> > choice ? I choice the square and I did not know how to setup the skip,
> > off, on, phase and stop after. My data is a 4d analyze format which
> 1-43
> > subject is controls and 44-50 is diseases. Besides, if I need to enter
> > the covariance data, how should I do ?
> > Thanks your help
> > JLH
> >
> > -----Original Message-----
> > From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On
> > Behalf Of Joe Devlin
> > Sent: Wednesday, October 16, 2002 11:11 PM
> > To: [log in to unmask]
> > Subject: Re: [FSL] parameter about the group analysis by FEAT
> >
> > >Thank for your reply. I only have the experience of " compare
> > >populations: 1scan/subject" ( that is a two sample t test) in SPM.
> Now
> > I
> > >have the 2 groups subjects in 4d analyze format. I like to try the
> > >"simple" analysis (compared population) and I don't know how to do
> it.
> > >Please give me any suggestion. Thanks
> >
> > If I understand correctly, basically you want to compute a two-tailed
> > t-test comparing
> > patients to controls across the voxels in the brain. I don't see any
> a
> > priori reason why
> > you couldn't do this in Feat but I don't think you want any of the
> fancy
> > options.
> >
> > It sounds like you want something like:
> > - Put all scans (1 per subject) into a single 4d file containing first
> > the
> > normals and then the patients.
> >
> > - You need all of your scans registered into standard space (which you
> > can
> > do with McFlirt
> > if you haven't already done this). Probably, though, it would be best
> > to
> > do this before using
> > Feat because Feat assumes that a single time series comes from a
> single
> > individual. In this
> > case you will enter one time series from many subjects.
> >
> > - Misc: DOn't change anything here.
> > - Data: TR = 1, delete volumes = 0, & you don't want high pass
> > filtering.
> > - Pre-stats: Turn off temporal filtering here and set the spatial
> > smoothing
> > to whatever you want (typically 8-16mm in PET).
> > Definitely don't want motion correction or slice timing correction.
> > May
> > not want brain extraction depending on how much
> > skull shows up in your PET images. Probably do want some form of
> > intensity normalisation but this is slightly tricky (see
> > previous email).
> > - Stats: Turn off pre-whitening. Select full model setup and create a
> > single ev. It should be a set of 1s -- one for each
> > normal. Patients will be implicitly modeled. Turn off
> convolution,
> > temporal filtering and temporal derivatives. The final
> > model should look like a column of 1s -- 1 per control followed by
> > zeros
> > -- 1 per patient. Presumably you want two
> > contrasts: Controls > patients (1) and Patients > controls (-1).
> > - Post stats: Pick a statistical threshold for accepting significant
> > results.
> > - Registration: None. Best to do this before creating the 4d data set
> > for
> > analysis because of the assumptions Feat
> > makes -- namely that you're analysing fMRI data rather than pet.
> >
> > Good luck.
> > Joe
> >
> >
> > >JLH
> > >
> > >-----Original Message-----
> > >From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On
> > >Behalf Of Stephen Smith
> > >Sent: Wednesday, October 16, 2002 5:18 PM
> > >To: [log in to unmask]
> > >Subject: Re: [FSL] parameter about the group analysis by FEAT
> > >
> > >Hi - I'm a little confused by this - are you saying that you have
> > >already
> > >carried out the first-level analyses from each subject and want to
> pass
> > >the relevant stats up to second level? Or are you wanting to carry
> out
> > >the
> > >first-level AND the second-level anlaysis inside FEAT?
> > >
> > >Thanks, Steve.
> > >
> > >
> > >
> > >
> > >On Wed, 16 Oct 2002, tulu wrote:
> > >
> > > > Hello Smith and FSL experts,
> > > > I try to setup the study. I have one group of controls (N=43) in
> 4d
> > > > analyze format and another groups of disease (N=7). I use the
> first
> > > > level analysis and number of analysis is 2 ( 2 groups of patients
> > ?).
> > > > The TR=3, High pass filter cutoff=100. Turn off the FILM
> > prewhitening
> > > > and turn on the intensity norm. Then the FSL ask to setup model
> > >contrast
> > > > details after I push the go. Is there any thing should change ?
> > Thanks
> > > > JLH
> > > >
> > > > -----Original Message-----
> > > > From: FSL - FMRIB's Software Library [mailto:[log in to unmask]]
> On
> > > > Behalf Of Stephen Smith
> > > > Sent: Tuesday, October 15, 2002 10:30 PM
> > > > To: [log in to unmask]
> > > > Subject: Re: [FSL] parameter about the group analysis by FEAT
> > > >
> > > > Well - for the first-level analyses the things that spring to mind
> > >are:
> > > >
> > > > Set TR to whatever it effectively was.
> > > >
> > > > Turn ON intensity norm?
> > > >
> > > > Highpass - if you set TR "right" then choose A+B length etc?
> > > >
> > > > Stats - turn off FILM prewhitening and also turn off convolutions
> > and
> > > > derivatives?
> > > >
> > > > Let us know how you get on...thanks, Steve.
> > > >
> > > >
> > > >
> > > > On Tue, 15 Oct 2002, tulu wrote:
> > > >
> > > > > Hello, Smith and FSL expert,
> > > > > Thank for your reply. Since I am not familiar with the fmri
> > setting,
> > > > can
> > > > > you tell me the detail setting about this particular case ? Such
> > as
> > > > the
> > > > > TR, high pass filter cutoff and the setting in stats ? I will
> > share
> > > > the
> > > > > result if it can work. Thanks
> > > > > JLH
> > > > >
> > > > > -----Original Message-----
> > > > > From: FSL - FMRIB's Software Library [mailto:[log in to unmask]]
> > On
> > > > > Behalf Of Stephen Smith
> > > > > Sent: Tuesday, October 15, 2002 7:00 PM
> > > > > To: [log in to unmask]
> > > > > Subject: Re: [FSL] parameter about the group analysis by FEAT
> > > > >
> > > > > A scary can of worms - does FEAT work on PET/SPECT? Well - we
> > >haven't
> > > > > tested it ONCE on PET data, so seems risky to make any
> > > > promises.....but
> > > > > in
> > > > > theory.....should be ok.....
> > > > >
> > > > > There are bound to be practical issues that come up - will be
> very
> > > > > interested to see how you get on. The new FEAT5 group stats
> stuff
> > > > > (FLAME)
> > > > > should add value to group stats with PET just like with FMRI,
> but
> > as
> > >I
> > > > > said, we've not tested it at all. Your particular case
> (N=7/N=43)
> > > > could
> > > > > well be a good example of when FLAME is better than simple OLS
> > > > > (ordinary-least-squares, ie dumb "random effects").
> > > > >
> > > > > Thanks, Steve.
> > > > >
> > > > >
> > > > > On Tue, 15 Oct 2002, Jung Lung Hsu wrote:
> > > > >
> > > > > > Dear FSL experts,
> > > > > > I have a simple question about the FSL. I have 2 groups of
> > >patients
> > > > by
> > > > > PET
> > > > > > study. One is disease (N = 7)and another is control (N = 43).
> I
> > >like
> > > > > to
> > > > > > know the difference between the disease group and control. Is
> it
> > > > > possible
> > > > > > to use the FEAT to do group analysis ? If so, what is the
> detail
> > > > > parameter
> > > > > > I should enter in the FEAT program ? Thanks
> > > > > > JLH
> > > > > >
> > > > >
> > > > > Stephen M. Smith
> > > > > Head of Image Analysis, FMRIB
> > > > >
> > > > > Oxford University Centre for Functional MRI of the Brain
> > > > > John Radcliffe 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
> > > > Head of Image Analysis, FMRIB
> > > >
> > > > Oxford University Centre for Functional MRI of the Brain
> > > > John Radcliffe 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
> > > Head of Image Analysis, FMRIB
> > >
> > > Oxford University Centre for Functional MRI of the Brain
> > > John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> > > +44 (0) 1865 222726 (fax 222717)
> > >
> > > [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
> >
> > Joe
> >
>
> Stephen M. Smith
> Head of Image Analysis, FMRIB
>
> Oxford University Centre for Functional MRI of the Brain
> John Radcliffe 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
Head of Image Analysis, FMRIB
Oxford University Centre for Functional MRI of the Brain
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
