Inverse transformation to transform brain images onto atlas space

[kenjp1223]

My undestanding is that in the default approach explained in the tutorial, the atlas space is aligned to the brain space to calculate a transformation matrix.
Is there a way to apply an inverse transformation so that I can visualize how the brain aligns with the atlas?

[SaibotMagd]

Hi I also tried to do this for a long long time. Short answer: clearmap uses the elastic toolbox which isn't doing a diffeomorphistic registration (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2276735/ ). In theory it is possible to calculate the inverse afterwards and it worked sometimes but never good enough, and needs a lot of ressources (my registrations last 1 day, to calculate the inverse its at least 1 week). For visualisation purposes I found its the easiest way to do just the backward registration by hand (switch fixed and moving image) or use a toolkit which can create diffeomorphistic registration (e.g. Ants

And then use a nice performance viewer like Imaris or the 3d-viewer for Fiji (itk-snap is great but only for small images < 1GB) to overlay the registration result onto your subject tissue.

But if you really want to mark the regions onto the subject image in full resolution (e.g. if you don't want to mark the region borders by hand) you have to upscale the registered atlas annotation afterwards to the same size as the subject tissue. I never found a way to do this for high resolution images, so I build an easy tool for myself: https://github.com/SaibotMagd/3D-annotation-image-upscaler-for-huge-images as proof-of-concept. You can create regional masks out of the results and place them onto your subject tissue to create stunning images in unlimited resolution and size.