users > CMTK questions (DICOM stacker, motion, etc)
Showing 1-1 of 1 posts
Results per page:
Apr 14, 2011  10:04 AM | Torsten Rohlfing - Google LLC
CMTK questions (DICOM stacker, motion, etc)
Hi Nolan:

I am going to take the liberty to cross-post your questions and my
replies on the CMTK users forum. Would be great if you could direct
further questions there.

On 04/13/2011 01:34 PM, Nolan Nichols wrote:
1. DICOM Image Staker 
- Can the output be set to a 4D NIfTI rather than single volume per file.

No, it cannot. CMTK internally handles only 3D data structures.
- Does this support Multi-Frame DICOM conversion to NIfTI?  From what I can tell it does not.

Yes and no. Multi-slice DICOM does not need stacking, as it already is a
volume, so dcm2image is the wrong tool. To convert it, you should
simply be able to use the convertx tool, something like so:

convertx I0001.dcm I0001.nii

2. Motion Correction/Registration
- What are your Best Practice recommends for parameter tweaking
during Motion Correction and Registration? (we have lots of squirmy

Motion correction is a rigid, strictly single-modality problem with high
accuracy requirement. My suggestion would be to use the "registrationx"
tool like so:

registrationx --auto-multi-levels 3 --dofs 6 --cubic --msd -o output.xform fixed.nii moving.nii

- Does CMTK have any skull stripping functions? We are using BET and are looking for something better.

CMTK currently does not have any skull stripping. We are also using BET and looking for something better ;)

What works quite well for us is to use T1/T2/PD-weighted three-channel
data, i.e., SPGR plus dual-echo FSE. Then we use BET to strip each
channel sperately and combine the three resulting masks via alignment to
SPGR (using registrationx and reformatx) plus voting (using imagemath).

3. DTI Related
- Do you have any parameter recommendations for registration that are specific to DTI?


- Does CMTK have any features for DTI, like eddy current correction? 


We use FSL pretty heavily, but if are there any areas that you see CMTK
being particularly strong let me know so we can test them out.

FSL's nonrigid registration blows. Their rigid/affine is very nice though.

CMTK has some stuff that FSL does not - e.g., super-resolution
reconstruction (film, jidb, volume_injection, and volume_reconstruction
tools). I also think our bias field correction (mrbias tool) is quite
effective, and it does not require you to strip the image first, as does
FSL's FAST (because it does the tissue classification simultaneously).