search

CoBrALab / antsRegistration-MAGeT

A reimplementation of MAGeTbrain using only ANTs tools.
Other
16 stars 6 forks source link

Conceptual Question: Should smoothing sigmas be based on voxel space or real space? #54

Closed gdevenyi closed 7 years ago

gdevenyi commented 8 years ago

antsRegistration, if you don't specify anything, assumes that smoothing sigmas are in voxel space.

For the case of 1x1x1 data, this means that antsRegistration runs the same as ANTS.

As soon as we move to non-isotropic or non-1x1x1 data, our sigma concept breaks.

This will affect registration timing no doubt

gdevenyi commented 8 years ago

Probably want to do:

--transform SyN[0.5,3,0] --metric CC[$fixedfile,$movingfile,1,4] --convergence [100x100x100x0,1e-6,10] --shrink-factors 8x4x2x1 --smoothing-sigmas 4x2x1x0mm
--transform SyN[0.5,3,0] --metric CC[$fixedfile,$movingfile,1,4] --convergence [100x20,1e-6,10] --shrink-factors 2x1 --smoothing-sigmas 1x0

Do a classic "real space" set of SyN steps and then move into voxel space for completion.

gdevenyi commented 7 years ago

Bigger problem was actually the shink factors, based on resolution.

These are now input independent in latest testing code.