Hi,
I'm not sure why you are doing a 6 DOF transformation to MNI space. This is not robust and I would expect it to fail quite often. Instead I would do a 12 DOF transformation to the MNI space and then you can transform your mask back into the original space, without worrying about changes of volume (since they will be undone when you go from MNI to native space).
In response to your specific questions:
1 - It doesn't matter if your intensities lie outside the min/max range in FSLView - these values are only used to display purposes and do not affect any analyses. They are purely a guide on how to display the image and can be changed without any consequences.
2 - I think this is a symptom of doing a 6 DOF transformation, rather than a 12 DOF one.
3 - I'm not sure what you mean by this. The ventricles should not be "distorted", but is unclear of exactly what kind of distortion is occurring.
I'm not sure what you mean by "correct" for intensity changes. You can use different interpolation methods (e.g. spline) which give different output intensities. However, I'm not what the problem is here or how it interacts with what you are trying to do.
I hope that some of this is helpful.
All the best,
Mark
On 4 Dec 2012, at 19:28, Julia Dieter <[log in to unmask]> wrote:
> Dear FSL-experts,
>
> I’m relatively new to FSL. I have some questions regarding FLIRT and would be very thankful for any help. I tried to find solutions elsewhere and in the FSL-list archives before posting but didn’t find the right directions. Please excuse if I overlooked something.
>
> We want to extract FA values from a specific brain region. Therefore we aim to create an individual mask for each subject based on a structural scan (we work with RRMS-patients) and then apply the mask to the FA-map. In order to make up for different head rotations in the scanner, we want to register the b0 to the MNI standard and apply the matrix to the FA-map, before mask creation.
>
> Since we don’t have an MPRAGE or MDEFT for all subjects, we decided on going with the B0-image of the dti-scans for mask creation. We have 3 identical dti-sequences and processed them as follows: Conversion to Nifti, ECC, averaging of 3 dti-sequences, BET. Of the resulting file, the B0-volumes 1-10 were extracted and averaged to give the final individual B0-image. We use fsl version 4.1.9.
>
> For our analysis, we now want to register the individual mean B0-image to the MNI-template via FLIRT, and we merely want rotations and transformations (i.e. 6 DOF).
>
> Registration of B0-image to MNI was done using the FLIRT-GUI (Input-image-> ref image; ref-image=MNI152_T1_2mm_brain; DOF=Rigid body 6 parameter model; Input-image=individual_B0). FLIRT-output matrix was applied to the FA-map with the following command:
>
> flirt -dof 6 -in <name of individual FA-map> -ref <name of FLIRT output image in MNI-space.nii.gz> -applyxfm -init <name of .mat-file created in FLIRT> -out <name of output-file>
>
> And here is what gets us stuck:
>
> 1) In the flirted B0-image (viewed by FSLview), the voxel-grey-values (when clicking into the image onto a particular voxel) are sometimes (especially in the CSF) higher than the maximum value in the min/max-scale at the menu bar. Is that normal (e.g. due to robust min/max) or is this because FLIRT went wrong? Is this just based on something like an altered storage of intensity information in the image header?
>
> 2) For one subject, FLIRT with 6 DOF did not work: The flirted image was completely wrongly oriented. The only difference of that subject’s original image to the original images of the other participants I could find was the brain’s position in the FOV, which might have been a bit more in the upper right of the image. FSLinfo did not show any irregularities. In separate trials, I then repeated FLIRT (6 DOF) for that subject with input-weighting the reference (MNI), OR search-angles x, y, z=0-90 OR “(normal) mutual” as cost function. For all 3 options, the FLIRT-output for the critical subject looked similar and ok (although a bit differently shaped to most of the other participants, but not much). Can either of these options be applied to the subject? Which option would be best? Is it still comparable to the other subjects or has the respective procedure then be applied to all participants?
>
> 3) Is it normal that the ventricles are distorted in the flirted B0-image, even when only doing rotation and translations?
>
> Also, is it possible to correct the FLIRT-based change of intensity-values (changed by rotation and translation) via information of FLIRT-output matrix?
>
> I would appreciate any kind of help. Thank you very much!
>
> Best regards
>
> Julia
>
>
>
|