Resources showing programmatic segmentation using available models

[mattwarkentin]

Hi,

Maybe I've been looking in the wrong spots, but are there any scripts/resources showing how to programmatically perform segmentation in Python rather than the Slicer UI using one of the available models? Thanks.

[lassoan]

Here is the inference script: https://github.com/lassoan/SlicerMONAIAuto3DSeg/blob/main/MONAIAuto3DSeg/Scripts/auto3dseg_segresnet_inference.py

[mattwarkentin]

Thanks, Andras!

[diazandr3s]

Hi @mattwarkentin,

Very good question. Thanks for asking.
There is no document specifically for this. But performing segmentation programmatically in Python w/out the Slicer UI is quite easy.

As @lassoan mentioned, you just need this script: https://github.com/lassoan/SlicerMONAIAuto3DSeg/blob/main/MONAIAuto3DSeg/Scripts/auto3dseg_segresnet_inference.py

In a virtual environment with MONAI installed:

pip install monai[fire,pyyaml,nibabel,pynrrd,psutil,tensorboard,skimage,itk,tqdm]

an example command will be:

python auto3dseg_segresnet_inference.py --model_file /PATH_TO_THE_MODEL_PT/model.pt --image_file /PATH_TO_SAMPLE_IMAGE/sample.nrrd --result_file /PATH_TO_OUTPUT/

If your images are in .nii.gz format, you may need to change these lines by this only:

nib.save(nib.Nifti1Image(seg, affine = original_affine), result_file)

Remember to do the import: import nibabel as nib

Hope this helps,

[mattwarkentin]

Great, thanks for the extra info @diazandr3s. This will definitely get me on the right track.

[diazandr3s]

Model info and links are here: https://github.com/lassoan/SlicerMONAIAuto3DSeg/blob/main/MONAIAuto3DSeg/Resources/Models.json