Hi all,
I am performing connectome construction on neonatal HARDI data. In order to implement this pipeline, I need to register the atlas labels (M-CRIB neonatal atlas) to the diffusion data. In order to keep high resolution of structural data,I perform the following steps:
## Register the atlas labels to the mask of the mean b=0 image
flirt -in BET_${S}_mask.nii.gz -ref T1_MCRIB_Structural_${S}_Labels.nii.gz -dof 6 -omat DKI_2_T1_${S}.mat
transformconvert DKI_2_T1_${S}.mat BET_${S}_mask.nii.gz T1_MCRIB_Structural_${S}_Labels.nii.gz flirt_import DKI_2_T1_mrtrix_${S}.txt
mrtransform -linear DKI_2_T1_mrtrix_${S}.txt -inverse T1_MCRIB_Structural_${S}_Labels.nii.gz T1_MCRIB_Structural_Labels_in_DKI_${S}.nii.gz
However, I later need to extract average measures of some diffusion metrics (e.g. DKI, NODDI) across atlas’ parcels but I find the two images do not overlay and have different headers:
There is a direct conflict between two of your statements:
In order to keep high resolution of structural data, …
… I later need to extract average measures of some diffusion metrics (e.g. DKI, NODDI) across atlas’ parcels …
If, for each atlas parcel, you want to compute the average of a DWI-derived measure, then by definition you need to construct a mask for each atlas parcel on the DWI voxel grid; the higher resolution of T1w image data may be beneficial in terms of registration to template and minimising quantisation effects during transformation of data, but ultimately at the end of the process you need those data on the lower-resolution DWI grid.
So you will need to not only apply the transformation of the atlas data from template to subject space, but also resample those data onto the DWI voxel grid.
At the point at which you do have your DWI metrics and label images on the same grid (and you can generate binary masks from the latter using e.g. mrcalc labels.mif 36 -eq), the relative strides of the two images should not matter. These strides only dictate how the intensity data upon a 3D voxel grid are distributed into a 1D “stripe” of data on the file system. What is important is that, when you read data from voxel index [ X Y Z ] independently from each of those two images, the data read from those two images correspond to the same anatomical location; and that is what should be guaranteed by defining the DWI voxel grid as the template when resampling the labels image.
6 degrees of freedom for registering to a template is exceptionally low; in my own experience I’ve been disappointed with even 12 DoF and have basically committed to using non-linear registration to template spaces unless there’s a strong justification for not doing so.
