Dear Steve,
Sorry to bother you again.
I used fslhd to generate the header from the raw 4D DTI images and found one discrepancy. In the four dimension (pixdim4), which is the spacing between different directional images, some of our data have a value of 4.7mm, while others have 0.0781mm.
When I visualize two images obtain from 4.7mm z-spacing and 0.0781mm z-spacing by using AFNI (it can lock the two images), I found a much higher (almost 0.002 to 0.006) radial diffusivity value in the same location of the skeletons of the 0.0781 z-spacing images.
Therefore, my question is "does the z-spacing differences change the DTI model fitting of L1, L2, L3 values?"
Further, since the FA images are derived from the L1, L2, L3 values as well, does it mean that my FA estimation is inaccurate too? Mmm.. I am starting to have a sick feeling in my stomach now....
Thanks.
Yi-Shin
---
Yi-Shin Sheu
Research Assistant
Developmental Biopsychiatry Research Program
McLean Hospital / Harvard Medical School
115 Mill Street
Belmont, MA 02478
Tel: 617-855-2942
Fax: 617-855-3712
On Tue, Mar 31, 2009 at 12:11 AM, Steve Smith
<[log in to unmask]> wrote:
Hi - this all sounds fine, assuming your design.mat and design.con are correct. You also might like to load all_RD_skeletonised into FSLView and turn on "timeseries" and check that you have valid, varying, data on the skeleton. Sounds like you probably do.
The beauty of TFCE is that it is designed to be sensitive to both focal, strong, effects, as well as diffuse, less strong effects - so it seems like possibly you have a very diffuse (i.e. covering lots of the skeleton), low effect that correlates with your model. There is nothing invalid about thism as randomise is still valid for this scenario. You might want to show both the p<.05 corrected map (lots of voxels surviving) and also a more conservatve thresholding (e.g. p<.005 etc) for also showing where the effects are stronger.
Cheers.
On 30 Mar 2009, at 20:00, Yi-Shin Sheu wrote:
Dear FSL experts,
I am looking at a set of DTI images in terms of FA, MD, L1 and RD (L2+L3)/2
images.
I didn't get any significant TBSS result for FA, MD, L1, however, when I ran
TBSS on RD images, I got a corrected p of 0.002 for the same contrast
(patient > control) and same design matrix. When I check the tfce correct p
stats, and threshold them at p<0.05 (in fslview, I put down 0.95), it is pretty
much the "entire" RD skeleton was significant, which is really awkward.
I went back and visually check the raw RD images, normalized RD images, and
RD skeleton, everything was normal. However, I still feel something is wrong,
so here is my tbss_non_FA pipeline.
1. copy the L2 images and L3 images for each subject, and use fslmaths to
derive the average of these two.
2. creat a RD folder contaims all the Rd images, and make sure the naming is
the same as images in "origdata"
3. Run tbss_non_FA RD
4. Run randomise -i all_RD_skeletonmised -o tbss_RD -m
mean_FA_skeleton_mask -d design.mat -t design.con -n 500 --T2 -V
Did I miss any step? I read some of the FSL threads that I might need to
scale the RD images. By the way, the intensity in my RD skeleton image
ranges from 0 to 0.001.
Kindly advise.
Yi-Shin
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
---------------------------------------------------------------------------
