How can I improve ants brain extraction performance?

Software Support
, , ,
1

I am trying to extract the brain from T1 weighted image using antsBrainExtraction. However, the result is disappointing because part of the skull can’t be removed. I used MNI T1 weighted image 1mm as template for skull stripping and its brain mask as brain probability mask.

The command line used is the following:

antsBrainExtraction.sh -d 3 -a /path/sub_T1.nii.gz -e /path/MNI152_T1_1mm.nii.gz -m /path/MNI152_T1_1mm_brain_mask.nii.gz -o /path/sub_stripped_ants.nii.gz

Do you have any kind of suggestion to improve the result ? I attach the image example.

Thanks

doubt_antsBrainExtraction
doubt_antsBrainExtraction1366×247 132 KB

2

synthstrip is very robust. Alternatively, mindgrab provides similar performance but is faster, brainchop -m mindgrab -i input.nii.gz -o output.nii.gz. You can even try out mindgrab by dragging and dropping your NIfTI image to the brainchop web page and choosing the omnimodal skull strip segmentation model - this uses your own web browser’s graphics card for the AI model, so data is not shared with a cloud server.

1 Like
3

Just to add a couple more options:

HD-BET works quite well and is fast, but requires a GPU to run in a reasonable time.

ANTsPyNet (and a new implementation in ANTsTorch) also provides a pre-trained model for brain extraction (see the “t1threetissue” option).