[Feature request] nnUNet V2 integration

[rcremese]

I just want to mention that this is a great work and that I love the 3D Slicer plugin !

I would be interested to contribute to the project with models trained with the nnUNetV2 pipeline. As I found exemples using the nnUNetV1 inference pipeline using the NNUnetRunner implemented in the library, I wondered if it's possible to add a nnUNetV2Runner, which CLI slightly differ.

I can work on this if it's considered as a feature to integrate.

[LennyN95]

Hi @rcremese thank you for your interest!

This is an excellent idea, we already have some models on the platform using nnUNetV2 but, I agree with you, having a convenience runner as we have for nnUNetV1 would be great.

I recently added MHub-IO collections, a new feature to better organize and to use MHub-IO modules on-demand. It seems to me that this could become the very first nnUNetV2 collection - what do you think?

• An example collection can be found here.
• The direct benefit I see, is that we can install nnUNetV2 as part of the collection's setup script (whereas it has to be installed manually when using the old NNUnetRunner).
• The setup.sh script should have access to all ENV variables declared in the Dockerfile, thus parametrization is possible (to some reasonable extend).
• Independence from our core development (pro's and con's, I think the better separation and encapsulation is beneficial, however, breaking changes must be manually updated downstream).

I would like to hear your opinion. I'll move forward then setting up the repository under the MHubAI Github Org and adding it as official collection.

It'd be amazing to have you taking on or support me with the development of the new NNUnet2Runner MHub-IO module!

Best,
Leo.

[LennyN95]

On a separate note, would you be interested in helping me updating the 3D Slicer extension? If yes, we can sync on this in a separate thread :)