Niftyreg interpolation choice not respected

[rijobro]

As per KCL-BMEIS/niftyreg#59, my impression is that cubic interpolation is always used during reg_aladin. The bug is therefore present in our SIRF wrappers around it.

This means that resulting images may contain negatives, which could be problematic (e.g., HKEM expects priors with non-negative voxel values).

The short-term fix for this (whether using NiftyReg directly or using the SIRF wrapper) is to register images, extract the transformation matrix and then resample using that matrix. Using niftyreg directly would look something like:

reg_aladin -ref ${source} -flo ${moved} -rigOnly -interp 0 -aff tm_aladin.txt
reg_resample -ref ${source} -flo ${moved} -res ${output} -inter 0 -trans tm_aladin.txt

[rijobro]

@HarryMarquis once this is sorted, it should simplify your workflow that we were setting up.