Thanks so much for your quick reply, Mark.
I managed to get the analysis more-or-less working. It simply required
changing "First timepoint is tag" to "control", as you suggested. This was
an obvious option that I should have tried in the first place, but oddly
enough, I know that my data is arranged with my "tag" as the first volume.
In fact, I would have thought that the required subtraction order to obtain
positive activation would depend on the inversion time (see S.-G. Kim & N.V.
Tsekos, MRM 37:425-435 (1997), p. 427-428). In any case, I now have my
activation patterns back!
I had chosen the Perfusion Subtraction over the Full Model because I was
given the impression from the documentation that the Full Model was
generally chosen to address BOLD contributions, which should be negligible
with TE=3ms. Would the Full Model still be more accurate?
I've attached a images showing:
A- the "graininess" of the Matlab subtraction
B- the blackened areas of the FSL subtraction
Any ideas as to what could be causing B? I presume that it shouldn't be an
issue if/when I can get the registration to work properly (i.e. so I can
overlay the activation onto my anatomical)...but wouldn't correct
registration depend on having sufficient signal intensity following subtraction?
Finally, I've also attached images showing my motion correction results.
Given the clear fluctuations from one volume to the next (i.e. fluctuating
between the tag and control), I question the validity of motion correction
on this data. Any thoughts?
Thanks again,
Nicole
On Fri, 15 May 2009 11:09:27 +0100, Mark Woolrich <[log in to unmask]>
wrote:
>Hi Nicole,
>
>> Perfusion Subtraction
>> rather than the Full Model, since I'm using a TE of 3ms.
>
>Wow, an echo time of 3ms seems unfeasibly low. That said, I am not
>sure how that relates to the choice of full model versus subtraction.
>A low TE just means that when you are doing the full model approach
>there is less need to have an EV to model the BOLD signal (see the
>webpage help http://www.fmrib.ox.ac.uk/fsl/feat5/perfusion.html).
>
>> My most significant issue is that there is no activation when using
>> the
>> perfusion subtraction in FSL. When I had been performing the
>> subtraction in
>> Matlab, then processing in FSL, I had clear motor activation and
>> well-formed
>> time courses. I have tried looking at the perfusion_subtract script,
>> but
>> unfortunately, I cannot follow it. I realize this is a broad issue
>> to be
>> asking about, but I really do not know where I'm going wrong.
>
>Is there plausible activation, but just at lower z-stats, or nothing
>coherent at all?
>
>Worth double checking all of the timings:
>- have you got the TR set correctly (should be the time taken for
>acquiring a tag OR control)
>- ideally the model should be shifted forward by TR/2 to adjust to the
>way in which the sinc interpolation is done (e.g. reduce your 3 col
>format custom timings by TR/2)
>- have you set the "First timepoint is tag" option correctly - it is
>worth checking negative contrasts as well
>
>> Other than this most prominent problem, I was also wondering why
>> your images
>> do not turn out grainy as mine do, and why in the FSL output, there
>> are
>> black areas around the edges of my images.
>
>Have not seen grainy images. But note that the default settings in
>Feat will do a small amount of spatial smoothing.
>
>> Finally, I question how one can perform motion correction on
>> perfusion data
>> at all, given the alternation between tag and control.
>
>
>In our experience, this works fine. There is enough common contrast
>between the tag and control to allow this to work. Some people have
>reported it working better on unBET-ed images, so try with and
>without. It may be trickier if there is background suppression, but we
>have seen that work OK as well. Best thing to do is to take a look at
>the motion corrected data and see if it looks alright.
>
>Cheers, Mark.
>
>----
>Dr Mark Woolrich
>EPSRC Advanced Research Fellow University Research Lecturer
>
>Oxford University Centre for Functional MRI of the Brain (FMRIB),
>John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK.
>
>Tel: (+44)1865-222782 Homepage: http://www.fmrib.ox.ac.uk/~woolrich
>
>
>
>
>On 14 May 2009, at 18:14, Nicole Pelot wrote:
>
>> Hello,
>>
>> I have multiple perfusion datasets. Up to this point, I have been
>> performing
>> the subtraction in Matlab prior to feeding the data to FSL. However,
>> the
>> Matlab subtraction produces very grainy images, preventing both motion
>> correction and registration from working properly since - to the
>> best of my
>> understanding - these are both based on image intensity.
>>
>> Given this problem, I am trying to use FSL's perfusion subtraction.
>> I have
>> read the relevant documentation, and am opting for the Perfusion
>> Subtraction
>> rather than the Full Model, since I'm using a TE of 3ms. Hence, I have
>> formatted my data into a single timecourse, with datapoints
>> alternating
>> between tagged and control. I've attached my NIFTI file.
>>
>> My most significant issue is that there is no activation when using
>> the
>> perfusion subtraction in FSL. When I had been performing the
>> subtraction in
>> Matlab, then processing in FSL, I had clear motor activation and
>> well-formed
>> time courses. I have tried looking at the perfusion_subtract script,
>> but
>> unfortunately, I cannot follow it. I realize this is a broad issue
>> to be
>> asking about, but I really do not know where I'm going wrong.
>>
>> Other than this most prominent problem, I was also wondering why
>> your images
>> do not turn out grainy as mine do, and why in the FSL output, there
>> are
>> black areas around the edges of my images.
>>
>> Finally, I question how one can perform motion correction on
>> perfusion data
>> at all, given the alternation between tag and control.
>>
>> Thanks so much,
>>
>> Nicole
>>
>> P.S.: Please let me know if any images (to show graininess,
>> blackened areas,
>> time courses, etc.) would be useful; they are too large in their
>> current state.
>>
|