Thanks for this, Jen. I agree with all of this, just one correction - for
the pre-threshold masking (as opposed to using masks in Featquery) the
mask _does_ need to be in the space of the FMRI data, so will need
transforming first, before using.
Thanks, Steve.
On Tue, 18 May 2004, Jennifer Bramen wrote:
> Hi Jason,
>
> Your ROI should be in the standard space. You can register the MRIco
> template to the standard and apply the transformation to the ROI you
> drew using flirt. It should only take a minute to do this. Then they
> can be used for pre-threshold masking.
>
> As for statistical power, if you run the analysis uncorrected, you will
> not gain anything from using the ROI. the ROI will improve you
> statistic al power during the correction for multiple comparisons.
> Generally speaking, cluster thresholding is less stringent than the
> voxel wise correction. If you anticipate highly significant very small
> activations, for example, nuclei in the brain stem, a voxel wise
> correction works best.
>
> As for what type of ROI you would like to use, I can only offer an
> opinion. I generally prefer to use irregularly shaped ROI which follow
> the underlying anatomy. I am also generally more inclusive than
> exclusive. The statistical gain made by reducing the number of voxels
> is small compared to failing to search a brain area that is in fact
> significantly active. If you are very concerned about reducing the
> number of voxels, I would recommend a hypothesis driven limit
> (excluding ROI that you have weak hypotheses about).
>
> Also, I always view my uncorrected data. I find it is useful to see
> everything that may be going on with my experiment.
>
>
> i hope this is helpful!
>
> Best
>
> jen
>
> Jennifer Bramen
> UCLA Brain Mapping Center
> 660 Charles Young Dr. So
> Los Angeles, CA 90095
>
> On May 18, 2004, at 8:27 AM, Jason Steffener wrote:
>
> > Hello,
> > I have two groups of subjects
> > and would like to investigate task related signal chaneg in each group
> > seperately and to ultimately use an unpaired t-test to compare the two
> > groups.
> >
> > I would like to limit my search to only a few regions of interest. I
> > have
> > drawn the rois on the MRIcro standard image which has voxel sizes of
> > 1x1x1
> > with a matrix of 181x217x181. Can I use these ROIs as pre-threshold
> > masking in the higher level FEAT analysis?
> > or
> > Does my ROI image need to have the same dimensions and voxel sizes as
> > the
> > FSL standard? (2x2x2, 91 109 91)
> >
> > What should my threshold be then when I use the pre-threshold masking?
> > voxel wise, cluster wise, uncorrected? I would guess voxel wise.
> >
> > Also in regards to the resels issue. My ROIs have a large surface area
> > and
> > should therefore limit the statistical power gained by limiting my
> > search
> > volume. Is there some tradeoff I should make? Could I make nonspecific
> > "smooth blob" rois that encompass my regions of interest. An ideal ROI
> > in
> > this case would then be a sphere, but what about cubes or 3-D ellipses?
> >
> > Any pointers are greatly appreciated.
> > best regards,
> > Jason.
> >
> > Jason Steffener
> > Department of Radiology
> > New Jersey Medical School
> > Biomedical Engineering Department
> > New Jersey Institute of Technology
> > Newark, NJ USA
> > (973) 676-1000 x1545
> > http://web.njit.edu/~jrs3
> >
> >
> Jennifer Bramen
> UCLA Brain Mapping Center
> 660 Charles Young Dr. So
> Los Angeles, CA 90095
>
Stephen M. Smith DPhil
Associate Director, FMRIB and Analysis Research Coordinator
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
|