 |
 |
Hi, I'm glad that that is what happened, as this makes the most sense. You haven't quite interpreted what I said about FLIRT correctly though. For FIRST I recommend that you do not do registration beforehand as the registration inside FIRST is optimised for subcortical structures, and this needs to match what happened with the training data (where such a subcortical-optimised registration was used). Outside of FIRST (regardless of whether you run FIRST or not) we recommend using other registrations (e.g. with FLIRT) and not the FIRST registration (as the FIRST one sacrifices cortical alignment to get better subcortical alignment, and so is only really useful inside of FIRST). I hope this answers your question, and that you haven't thrown your FLIRT registrations away, as you should still use them outside of FIRST. All the best, Mark On 14 Sep 2012, at 14:01, Victor DelBene <[log in to unmask]> wrote: > Hi Mark, > > A little while after posting I realized that using FLIRT and then the run_first_all was basically applying an affine registration a second time. I completed all the steps again, but in the run_first_all script, I dropped the -a portion and only used FLIRT for the registration. The ROIs were much more anatomically reasonable after this. So you are right that they should line up (and they do now), but it looks like doing two affine registrations unintentionally was the problem. > > Now that I am reading your response though, I am under the impression that I should avoid FLIRT initially, and just use the affine registration in the run_first_all script. Am I interpreting this correctly? What is the benefit of this over using the FLIRT function in the GUI? When would using FLIRT be necessary? > > Thanks Mark! > > Victor > > On Fri, Sep 14, 2012 at 2:50 PM, Mark Jenkinson <[log in to unmask]> wrote: > Dear Victor, > > This sounds quite concerning. > The segmentations should align well with your input image (i.e. con001_brain). > I assume that this image is the image you use as an input to both FLIRT and FIRST. > In fact there is no reason to run FLIRT prior to FIRST, as run_first_all will run the registration for you - and slightly differently, so we recommend that you do not do your own unless necessary. However, this is unlikely to be the big problem. > > So, to be sure, if you load con001_brain and con001_brain_all_fast_firstseg into FSLView then they do not line up correctly (and the subcortical segmentations are in the cortical areas or outside of the brain) - is that right? > > All the best, > Mark > > > > On 14 Sep 2012, at 10:11, Victor DelBene <[log in to unmask]> wrote: > > > Dear FSL experts, > > > > I am having some difficulty with the registration step of FSL-FIRST. > > > > What I have done: > > > > After converting all raw data files to .nii.gz, I used fslreorient2std. Following that, I used BET to extract the brain from the skull, and then the FLIRT linear registration in the GUI. These are T1 structural images (2X2X2) and I used an Affine (12 parameter model) with the reference image being MNI152_T1_2mm_brain. > > > > After this, I then used the following: > > > > run_first_all -i con001_brain -b -s L_Hipp,L_Amyg,R_Hipp,R_Amyg \ -o con001 -a con001_brain_to_std_sub.mat > > > > After this was completed, I was able to open up the con001_all_fast_firstseg.nii.gz file in FSL View. Using the CMA labels (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FIRST/UserGuide#Labels) I was able to confirm that the correct regions were segmented. > > > > cat con001.logs/*.e* provided no output. In the documentation it states "If everything worked well you will see no output from this, otherwise it will show the errors." I received no outputs. > > > > I then opened the T1, skull stripped image in FSL View. I then added the con001_all_fast__firstseg file. The segmented regions are shown to be in the upper cortex - motor and preforntal areas mainly, but the L amygdala and L Hippocampus are not even matching up the with the brain. All images, including the segmented regions, are in MNI152 space. > > > > Before I run stats on any volumes for the ROIs, I wanted to get feedback on this issue. At this point I am not confident that the voxel and volume outputs from fslstats con001_all_fast_firstseg -1 16.5 -u 17.5 -v would be reliable. > > > > Is there any reason why there sub-cortical regions would be placed in the cortex and outside the brain when I overlay the segmented data on the T1 image? > > > > Any help wold be immensely appreciated! > > > > Thanks, > > > > Victor > > >