Michael as Jesper mentioned-- if you don't use nearest-neighbor
interpolation-- it will basically "average" the labels so a voxel that
was a "7" now becomes a 6.45 or something like that... and goofs up
the label assignments.
By using nearest-neighbor it keeps things as integers and they get the
label based on the ... nearest-neighbor :-) (i.e. not some % of the
26 or so neighboring voxels averaged together)
dg
On Tue, May 3, 2011 at 5:50 PM, Jesper Andersson <[log in to unmask]> wrote:
> Dear Michael,
>
>> Hi there,
>>
>> This post is in reference to "FA & MD value extraction within thalamus" (Wed, 9 Feb 2011).
>>
>> I would like to quantify FA (from TBSS) and grey matter volume (from FSL-VBM) in the 7 subregions of the thalamus, as determined by the Oxford thalamic connectivity atlas, in each of 90 subjects. To this end, I've tried to move the thalamic atlas into each subject's space. This is what I have done so far:
>>
>> 1. Moved the masked T1 image into MNI space
>> flirt -in T1_masked -ref MNI -omat T1_to_MNI -out T1_masked_MNI
>>
>> 2. Inverted the resultant transformation matrix
>> convert_xfm -omat MNI_to_T1 -inverse T1_to_MNI
>>
>> 3. Moved the thalamic atlas into T1 space
>> flirt -in atlas -ref T1_masked applyxfm -init MNI_to_T1 -out atlas_T1
>>
>> As you have probably expected, step 3 gave rise to labels "mixing up." What is the best way to solve this problem? That is, what is the best way to move the thalamic atlas into each subject's space? Could I threshold the thalamic atlas to create 7 masks, one each for the 7 subregions of the thalamus, move each of these masks into a subject's space, and then add them together?
>
> that is the method I have been using in similar situations. It is less fiddly than it sounds if you write a little script for it. Remember also to either use nearest-neighbour interpolation or to threshold the images after the interpolation.
>
> Good Luck Jesper
>
--
David A Gutman, M.D. Ph.D.
Center for Comprehensive Informatics
Emory University School of Medicine
|