Using a scaled down version of a volume to segment, then using this segment on original large size volumes

Support
1

Dear all,
I have large datasets (10 to 30GB per image) which are thus difficult to segment on original files. One solution I imagined is
1- to use a 10X downscaled version of my images,
2- paint “seeds” in the 14 regions I want segmented,
3- shrink the sizes of each segment by a couple of voxels,
4- then apply the seg.nrrd file to the original large size image after and
5- apply a grow from seed function (and go make myself a coffee…).
Everything works well up to step 5. Segments are matching to the new high resolution image but are somewhat fixed to low resolution settings.

image
image1920×1080 142 KB

Thanks for your help

2

Apologies for posting something I eventually found a solution for.
I needed to edit the Specify Geometry option to oversampling the seg.nrrd file and increase it 10X
I am now running grow from seed the way it is supposed to on high resolution images.
Kind regards (and if moderators feel this is not a useful topic, feel free to delete)

3

Thanks for posting this - yes, we definitely like to keep this kind of post in the system so that others can learn in the future.

1 Like