memory consumption and parameters

[antsx]

memory consumption and parameters

[stnava]

the major area where you will improve performance will be here:

,Random,0.05 => ,Regular, 0.25

you may also want a -x option, i.e. some kind of mask to focus the registration

another helpful thing is

-r [ fixed, moving , 1 ]

to initialize .... starter parameters are here:

https://github.com/stnava/ANTs/blob/master/Scripts/newAntsExample.sh

finally, you might try ANT

antsAffineInitializer

if need be ....

[satra]

for the -r flag is there expected to be an inverse transformation and also is that transform from the -r flag included in the composite? i'm trying to clean up the way nipype reads and collects these transforms.

[stnava]

you should also add

 -z, --collapse-output-transforms (1)/0

so that regardless of the number of .mat type transforms, they are all
collapsed into one ... although i believe this is default behavior.

brian

On Tue, May 7, 2013 at 9:25 PM, Satrajit Ghosh [email protected]:

for the -r flag is there expected to be an inverse transformation and
also is that transform from the -r flag included in the composite? i'm
trying to clean up the way nipype reads and collects these transforms.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/17#issuecomment-17581361
.

[ntustison]

Yes, -z 1 is the default behavior.

Also, to answer Satra's questions --- "no" to the first and "yes" to the second.

[satra]

thanks guys.

running it now with the command below taken from the ANTSexample that brian pointed me to (here -z 0 is being used). should i change the nipype default to work with -z 1 (@hjmjohnson : what do you think?)

i'm presuming that -z 1 will not generate the various transform0/1/2... files

in this case, let's say i wanted to get the affine component of the composite - is there a way to extract/generate that from the composite?

appendix:
(in nipype we like to be verbose)

'/software/ANTS/versions/130123/antsRegistration --collapse-linear-transforms-to-fixed-image-header 0 --collapse-output-transforms 0 --dimensionality 3 --initial-moving-transform [ /gablab/p/CBMPARC/satra/human_2_anatomy.nii, /gablab/p/CBMPARC/satra/human_1_anatomy.nii, 1 ] --interpolation Linear --output [ transform, output_warped_image.nii.gz ] --transform Translation[ 0.1 ] --metric Mattes[ /gablab/p/CBMPARC/satra/human_2_anatomy.nii, /gablab/p/CBMPARC/satra/human_1_anatomy.nii, 1, 32 ,Regular,0.1 ] --convergence [ 10000x0x0, 1e-08, 20 ] --smoothing-sigmas 4x2x1vox --shrink-factors 6x4x2 --use-estimate-learning-rate-once 1 --use-histogram-matching 0 --transform Rigid[ 0.1 ] --metric Mattes[ /gablab/p/CBMPARC/satra/human_2_anatomy.nii, /gablab/p/CBMPARC/satra/human_1_anatomy.nii, 1, 32 ,Regular,0.1 ] --convergence [ 10000x0x0, 1e-08, 20 ] --smoothing-sigmas 4x2x1vox --shrink-factors 6x4x2 --use-estimate-learning-rate-once 1 --use-histogram-matching 0 --transform Affine[ 0.1 ] --metric Mattes[ /gablab/p/CBMPARC/satra/human_2_anatomy.nii, /gablab/p/CBMPARC/satra/human_1_anatomy.nii, 1, 32 ,Regular,0.1 ] --convergence [ 10000x0x0, 1e-08, 20 ] --smoothing-sigmas 4x2x1vox --shrink-factors 6x4x2 --use-estimate-learning-rate-once 1 --use-histogram-matching 0 --transform SyN[ 0.3, 3.0, 0.0 ] --metric Mattes[ /gablab/p/CBMPARC/satra/human_2_anatomy.nii, /gablab/p/CBMPARC/satra/human_1_anatomy.nii, 1, 32 ] --convergence [ 10x0x0, 1e-08, 20 ] --smoothing-sigmas 2x1x0vox --shrink-factors 4x2x1 --use-estimate-learning-rate-once 1 --use-histogram-matching 1 --winsorize-image-intensities [ 0.0, 1.0 ] --write-composite-transform 1'

[satra]

and this time it did not create transform0DerivedInitialMovingTranslation.mat. is there a reason why it would create it once but not again?

[ntustison]

The -z 1 option collapses all linear transforms together into a single affine transform file.

[satra]

i'm closing this for now - the memory consumption might still be an issue - given that brian's starter parameters aren't meant to be really used. will reopen if this problem persists.