Add support for running inference remotely

[che85]

It seems complicated to be to distribute the trained nnUNet models to every machine and I think it would be useful to set up servers that host the models.

Less maintenance, server-side updates can be done quickly without the need of redistributing specific models.

SlicerMONAIAuto3DSeg (https://github.com/lassoan/SlicerMONAIAuto3DSeg/tree/main) could be used as an example here.

[Thibault-Pelletier]

Hi @che85,

Thanks for the request.
As SlicerNNUNet dependency install can be a pain, and GPU/CPU ressources may not be available on the end machine, I understand the appeal of having a dedicated server and being able to connect to it through Slicer.

I'll have a look at SlicerMONAIAuto3DSeg implementation.

[lassoan]

There are many similarities between monai and nnunet based segmentation extensions (BreastUltrasoundAnalysis, CADSWholeBodyCTSeg, HDBrainExtraction, LungCTAnalyzer, MEMOS, MONAIAuto3DSeg, MuscleMap, PolycysticKidneySeg, RVXLiverSegmentation, ScapularMorphology, SegmentHumanBody, SlicerMOOSE, TotalSegmentator, VoxTell) in how they install dependencies and run inference (foreground/background, in-process/separate process/remote server).

It would be nice if they would all use a common classes so that we don't need to implement and maintain all these similar implementations in parallel.

[che85]

@lassoan Do any of these modules (except MONAIAuto3DSeg) remote inference implemented? I agree that we could identify a common ground here instead of reimplementing things.