 |
 |
Hi Michelle,
I haven't seen your attachments (large ones are not allowed on the FSL list, so you need to put the image somewhere and send a link to it) but I can comment on other things.
What you are describing and the command you are using looks fine. As long as you use the -b option in run_first_all then it should be OK to pass in brain extracted images.
Have you looked specifically at the registrations? You can find info on this here:though loading individual examples into FSLeyes (or FSLView) rather than relying on slicesdir can be better. Just check the *_to_std_sub.nii.gz image and see if it aligns reasonably well with the MNI or not (in the subcortical areas - the cortex may be a little off).
As for your questions:1) You can run the boundary correction separately. This is described here:and you might find that the thresh option with -t give you most control.
2) the bias field corrected by FAST is about corruption to the intensities due to RF (or B1) inhomogeneities, and these are totally different from geometrical distortion and signal loss caused by B0 inhomogeneities (where the latter is corrected via fieldmaps). So they are totally separate and the impact of B0 inhomogeneities on most structural scans is negligible, so it is only the bias field due to the RF/B1 inhomogeneities that we worry about.
All the best,Mark
Hi Mark,
The brains are already skull-stripped before running FAST and FIRST. They are skull-stripped using AFNI because that has worked better for this particular dataset relative to BET options. I see according to the documentation that FAST should be done with brain-extracted, but FIRST should be done without brain-extracted. But what in the case of wanting to use the bias-corrected images in FIRST? The first time I ran FIRST, when it worked quite well, I also used the brain-extracted structural scan. So it doesn't make sense to me that it would just be about brain-extraction. However, our structural scans always have extra skull included after brain-extraction, no matter what program and options are used. So perhaps FAST thinks that extra skull is brain, and that is then making FIRST confused during registration?
The FAST output looks pretty good and it seems like the bias-correction is doing a good job. Here is the command, and attached are screenshots of the output.
fast -g -b -B mprage_deoblique_brain.nii.gz
Here is my command for FIRST. I am using the "restored" image as input; is this correct?
run_first_all -i FSL_FIRST_SEG_biascorrected/mprage_deoblique_brain_restore -o FSL_FIRST_SEG_biascorrected/${ n}_biascorrected_structural -b -s L_Hipp,L_Amyg,R_Hipp,R_Amyg -v
Two related questions:1 - is there any way to modify the parameters of the FIRST segmentation? When comparing to Freesurfer, the segmentations for each region are smaller and upon visual inspection, may have more conservative boundary correction than desired. Is there a way to run FIRST with different options that would influence this?
2- How is the bas correction done by FAST different than bias correction that occurs when you add a field map to your registration? Are these similar, or two separate kinds of bias correction?
Thanks!Michelle
On Wed, Aug 16, 2017 at 6:38 PM, Mark Jenkinson <[log in to unmask]> wrote:
Hi Michelle,
What are you doing about brain extraction?Can you tell me exactly what steps you are taking and what your FIRST command line options are?I wonder if your problem is due to a mixture of brain-extracted and non-brain-extracted images.Also, does it look like the bias-correction is doing a good job?
All the best,Mark
Hi Mark,
I ran FAST and used the bias-corrected images ('restored' image) in FIRST resulted in approximately 15% of subjects failing registration in FIRST, whereas the rate of failure without using FAST- bias-corrected images was ~1%. I had expected that FIRST would be improved after removing bias fields from the T1. Do you have any idea why this might happen?
M
On Mon, Jul 3, 2017 at 3:34 AM, Mark Jenkinson <[log in to unmask]> wrote:
Hi Michelle,
It is not always necessary to run FAST to bias-correct an image before FIRST as we find that FIRST is not very sensitive to bias field. However, the best practice would be to run FAST initially to get a bias-corrected image and then run FIRST.
All the best,Mark
Hello,
I recently attended the FSL Course in Vancouver and I have a follow up question on FIRST/FAST. I would like to use FIRST to get subcortical segmentation from subject's anatomical scans. Is it also recommended that I use FAST before using FIRST, such that i could use the anatomical scan that has been bias-corrected in FAST to use as input into FIRST? Or, are they two entirely separate entities for segmentation and that wouldn't make sense?
Thank you,Michelle
--
Michelle VanTieghemPhD student in PsychologyDevelopmental Affective Neuroscience LabColumbia University
--
Michelle VanTieghemPhD student in PsychologyDevelopmental Affective Neuroscience LabColumbia University
--
Michelle VanTieghemPhD student in PsychologyDevelopmental Affective Neuroscience LabColumbia University