ANTsX / ANTs

Advanced Normalization Tools (ANTs)
Apache License 2.0
1.22k stars 381 forks source link

About the segmentation results from antsCorticalThickness #1801

Open JinDJsuper opened 1 month ago

JinDJsuper commented 1 month ago

Operating system and version

Ubuntu 20.04

CPU architecture

Please select

ANTs code version

ants-2.5.0

ANTs installation type

Please select

Summary of the problem

hi When using antsCorticalThickness, I found that the segmentation results had regions in the cortical areas that were connected, leading to inaccurate cortical thickness (CT) measurements. Is there any way to resolve this issue?

My species is a dog.

thanks

image image image 68_BrainSegmentationPosteriors2.nii.gz 68_BrainSegmentationPosteriors3.nii.gz 68_CorticalThickness.nii.gz

Commands to reproduce the problem.

antsCorticalThickness.sh -d 3 -a ${input} -e ${sub_template} -m ${sub_template_mask} -p ${prior_dir}/template_%d.nii.gz -y 0 -w 0.25 -o ${ct_output_dir}/${output_name}_

Output of the command with verbose output.

.

Data to reproduce the problem

.

ntustison commented 1 month ago

antsCorticalThickness was developed in the context of human data and it might not generalize cleanly to other species. Given your application, you'll probably need to decompose the pipeline into its components (Atropos + KellyKapowski) and debug each one. But you'd have to figure out the best parameters yourself.

JinDJsuper commented 1 month ago

antsCorticalThickness was developed in the context of human data and it might not generalize cleanly to other species. Given your application, you'll probably need to decompose the pipeline into its components (Atropos + KellyKapowski) and debug each one. But you'd have to figure out the best parameters yourself.

ok,i got it . thank for your reply

cookpa commented 1 month ago

There are canine atlases that might be of help, eg

https://www.nature.com/articles/s41598-020-61665-0

The thickness algorithm is somewhat resistant to the "closed sulcus problem", where you have placed your crosshairs, you'll see that thickness is roughly half of the distance across the whole joined area. Of course, the more you can resolve the correct tissue labels, the better it will be.

I agree though that you will probably need to customize the scripts to do what you need.