Optimal registration is important when dealing with MPRAGE and PD images.
Visual inspection at every pre-processing stage is crucial.
High resolution data can be very unforgiving of small errors.
The MPRAGE and PD are acquired back to back. And image pre-processing steps begin with the division the MPRAGE image by the PD image.
While there should be little motion between the scans, at sub-millimetre scales, even the slightest head motion can be problematic.
An affine registration can be done moving the PD to the space of the MPRAGE before division.
Unconstrained co-registration between the images can be sub-optimal (It is important to visually inspect the data before and after any processing steps).
Optimal Registration Strategy involves creating a brain mask and constraining the registration over the brain only.
Taking a closer look:
With whole-image affine registration
With brain-constrained affine registration
If you cannot spot the differences between the two, maybe looking at the registered PD images does the trick.
How does the registration affect the result of the division image (the image used for tissue segmentation)?
Division without any type of image registration is clearly sub-optimal.
The division image after constrained registration is better than the whole-image affine. The quality I’m talking about is ofcourse after visual inspection and a noticable reduction in division artefacts.
In regions with a lot of vasculature, clear aliasing of the bright vascular signal and voxel shifts can be seen when the division was done using the unconstrained PD image.
Notes: All alignment was done using ANTs’s antsRegistration using Mattes Mutual Information. Mask was created using FSL’s BET after quick biascorrection using AFNI’s 3dUnifize.