reg_aladin v1.3.9 and v1.5.58 discrepancy

[mmodat]

In gitlab by @rmharbr on Apr 17, 2018, 16:34

Using the executable reg_aladin, I notice a large discrepancy between v1.3.9 & v.1.5.58 (most recent pull) in the registration of some head images. As I'm dealing with head images, I'm using the -rigOnly flag.

I'm experiencing this problem for multiple PET frames (longitudinal), as well as multimodal registration between an MR and PET image.

Strangely, v1.3.9 seems to perform the registration correctly (the output image is nicely superposed on the reference image in Slicer). With v1.5.58, although the output image moves closer to the reference image, it is still poorly aligned. The required rigid transformation is largely a translation in the z-direction, so it is easy to qualitatively say which one is better.

Is there anything I'm missing when upgrading the version of NiftyReg? I know that by default symmetric aladin is now used, but whether I use symmetric or non-symmetric, the old version seems to give better results.

[mmodat]

In gitlab by @rmharbr on Apr 17, 2018, 16:41

changed title from reg_aladin v1.3.9 and v1.5 discrepancy to reg_aladin v1.3.9 and v1.5{+.58+} discrepancy

[mmodat]

In gitlab by @rmharbr on Apr 17, 2018, 16:41

changed the description