Switching between models for automatic segmentation from command line

[lukasvanderstricht]

Hi everyone

I am training some modules via MONAILabel and 3D Slicer and was currently I am working with two models that I have trained (and are in the same app). For the workflow I want to achieve, I would want to automate the process as much as possible, and hence I am looking at ways to turn the workflow into a Python script.

Here is in broad lines what I want to do:

1. Run automatic segmentation with my first model, let's call it deepedit1
2. Calculate some volumetric properties of the resulting segments
3. Run automatic segmentation (on the same volume), with my second model, let's call it deepedit2
4. Calculate some volumetric properties of the resulting segments

Here is what I have done already:
I am able to calculate the the volumetric properties with the segment statistics module in Python, so steps 2 and 4 are no problem. I also know how to trigger the automatic segmentation, namely with the command slicer.modules.MONAILabelWIDGET.onClickSegmentation().

Now here is my problem
I start my MONAILabel server by loading in my two models, something like this:
monailabel start_server --app radiology/ --studies data/train_images/ --conf models deepedit1,deepedit2
This ensures that in the Slicer GUI, under the tab Automatic segmentation, I can choose which model I want to use for the automatic segmentation.
I however cannot figure out how I could specify the model used for automatic segmentation using just Python commands. I tried giving the model as a parameter for the onClickSegmentation() function, but that doesn't work.

Alternatively:
I also tried using the MONAILabel Logic module to circumvent this problem. More specifically, I tried something along the following lines:

logic = slicer.util.getModuleLogic('MONAILabel')
result_file,params = logic.infer("deepedit1_seg","/home/data/train_images/DICOMX")

where "DICOMX" is the name of the file that is currently loaded into Slicer.
When I do this, however, the labels seem to be computed, yet they do not show up in my viewers and also not in my data tree in Slicer. If I run the following command:
slicer.util.loadSegmentation(result_file)
the segmentation is visible and loaded into Slicer, but not on the segmentation node where the my segmentation would be if I just clicked to run automatic segmentation in the GUI. On top of that, the segments don't have the correct name.

Does anyone have any idea how I could solve this problem?

Thanks in advance!

Kind regards

Lukas

[diazandr3s]

Hi @lukasvanderstricht,

Have you tried running MONAI Label main file as a standard Python script? Here you can find an example: https://github.com/Project-MONAI/MONAILabel/blob/main/sample-apps/radiology/main.py#L263-L346

Just specify the studies, models and command you want to run (i.e. train, infer): https://github.com/Project-MONAI/MONAILabel/blob/main/sample-apps/radiology/main.py#L289-L291

Hope this helps,

[lukasvanderstricht]

Hi @diazandr3s

Thanks a lot for your reply! I would however want to run this interference from within a Python file that automates a whole intricate workflow and that is run from the Python console in 3D Slicer. So I think this solution is hard to tie in with other actions I want to automate.

I will keep looking for solutions and keep you updated!

Again thanks for the help!

[lukasvanderstricht]

Hi again,

I found a solution to my problem! Using the following command, you can choose the loaded model which you want to use for automatic segmentation:
slicer.modules.MONAILabelWidget.ui.segmentationModelSelector.setCurrentText()

As argument you give a string with the name of the preferred model.
For example,
slicer.modules.MONAILabelWidget.ui.segmentationModelSelector.setCurrentText("deepedit_seg")
switches the selected model to the deepedit model.

Thanks for the help!

Lukas