I have 2D images of the brain where the gray scale value (min = 0, max = 100) is a quantitative measure of the amount of gray matter. I would like to normalize these to MNI space, so that I might perform some group-level analyses on about 20 subjects. The original resolution of the data is 0.5 x 0.5 x 5mm (32 slices covering the whole brain). I also have standard 3D T1-weight images on these same subjects with the typical 1 x 1 x 1mm spatial resolution.
I have tried following the user guide for fnirt for normalizing images to MNI space using the following commands:
bet T1 T1_brain -R
flirt -ref MNI152_T1_2mm_brain -in T1_brain -omat my_affine_transf.mat
fnirt --in=T1 --aff=my_affine_transf.mat --cout=my_nonlinear_transf --config=T1_2_MNI152_2mm
flirt -ref T1_brain -in GMimage -dof 6 -omat GMimage2T1.mat
applywarp --ref=MNI152_T1_2mm --in=GMimage --warp=my_nonlinear_transf --premat=GMimage2T1.mat --out=warped_GMimage.nii
The problem I am running into is downsampling the gray matter images from 0.5 mm in-plane resolution to 2 mm in-plane resolution and upsampling the through-plane resolution from 5mm to 2mm seems to be introducing a lot of noise into these quantitative maps such that I am not sure whether the results I get from performing a between-group analysis are the result of this increased noise/blurriness or represent real, statistically significant differences. For many of the clusters, I am seeing between-group differences in regions where I essentially have no data in the original images (gray scale value < 1).
My question then is: is there a way for me to normalize these 2D images to MNI space, such that I can preserve original spatial resolution, rather than having to reslice the images into the 2 x 2 x 2mm resolution of the MNI template? From reading through the fnirt user guide section about the applywarp command, it looks like it may be possible. However, it looks like if I want to specify the output resolution using either --warpres or --knotspace, then I can only get a 4D file of spline coefficients out rather than an image. Am I understanding this correctly, and if so, my question then becomes if it is possible to use either of these parameters (or others) to specify an output resolution and still get an image file out?
Thanks in advance,