Hi Freesurfers,

(Sending again as I think the original may have been lost in the OHBM shuffle)

I'm using mri_robust_register to obtain a registration between high resolution (.7mm isotropic) whole-head T1w and T2w images from the same subject. An example command line is:

mri_robust_register --mov T2w.nii.gz --dst T1w.nii.gz --lta T2w_to_T1w.lta --warp T2w_reg.nii.gz --cost ROBENT --entradius 2 --entcorrection --satit --iscale

This works extremely well -- much better than FLIRT. (I'm using this registration to do some preprocessing before running recon-all, so unfortunately bbregister is not an option).

But, it takes *forever*. (My runtimes are about 90 minutes using OMP parallelization with 24 cores, or about 16 hours running in serial).

I'm interested in playing with the various options to see if I can get a suitable registration with a faster runtime. But it's not obvious to me how I should fiddle with the knobs to get started. For example, would increasing or decreasing `entradius` be expected to make things go faster? Also, I am acquiring the images in the same session, back to back, so I can assume that the images start off with a pretty good (but not exact) registration. Is there a way to use that assumption to speed things up?

Thanks, Michael

On Tue, Jun 27, 2017 at 10:35 AM, Michael Waskom mwaskom@nyu.edu wrote:

I'm using mri_robust_register to obtain a registration between high resolution (.7mm isotropic) whole-head T1w and T2w images from the same subject. An example command line is:

mri_robust_register --mov T2w.nii.gz --dst T1w.nii.gz --lta T2w_to_T1w.lta --warp T2w_reg.nii.gz --cost ROBENT --entradius 2 --entcorrection --satit --iscale

This works extremely well -- much better than FLIRT. (I'm using this registration to do some preprocessing before running recon-all, so unfortunately bbregister is not an option).

But, it takes *forever*. (My runtimes are about 90 minutes using OMP parallelization with 24 cores, or about 16 hours running in serial).

I'm interested in playing with the various options to see if I can get a suitable registration with a faster runtime. But it's not obvious to me how I should fiddle with the knobs to get started. For example, would increasing or decreasing `entradius` be expected to make things go faster? Also, I am acquiring the images in the same session, back to back, so I can assume that the images start off with a pretty good (but not exact) registration. Is there a way to use that assumption to speed things up?

Thanks, Michael