nnUNet Integration in MONAI ecosystem with MONet Bundle

[SimoneBendazzoli93]

Hi everyone!

Over the past year, I have been developing the MONet Bundle to better integrate nnUNet within the MONAI ecosystem. The bundle enables:

• Direct nnUNet training integration within MAIA Core
• Support for MONAI Deploy, NVFlare (federated learning), and MONAI Label (active learning)

I have started contributing to the official MONAI repositories to bring this work into the main distributions (#8328 and PR #543 in monai-deploy-app-sdk).

I also recently presented the bundle at MICCAI 2025, demonstrating its NVFlare capabilities: MONet-FL: Extending nnU-Net with MONAI for Clinical Federated Learning.

Additionally, the following trained models, using the MONet bundle, are made available:

• Brain tumor segmentation (BraTS 2025)
• Whole-body PET-CT lymphoma segmentation, adopting the MONet Bundle for Federated Learning with NVFlare

Both models can be included in the MONAI Model Zoo and can serve as examples for using the MONet Bundle.

Additionally, I have prepared some Jupyter Notebook tutorials to help users get started with the MONet Bundle. These tutorials could be integrated into the MONAI Tutorial repository.

Next Steps / Proposal:
If the MONAI community finds this interesting, I would like to bring this work to the MONAI Core Working Group to coordinate and plan the full integration of the MONet Bundle into MONAI.

[aylward]

This is outstanding! I have been following the MAIA effort, and it is a great example of a platform to support scientific research. It would be great to integrate the nnUNet/MONet bundle effort with the other nnUNet efforts in MONAI (Deploy and Core).

@rfloca (Human-AI WG / Label) @hshuaib90 (Deploy WG) and added you to Dev WG.

[rfloca]

@aylward Thanks for tagging me.
@SimoneBendazzoli93 Thank you for your efforts and bringing this up. A good integration of nnUnet with MONAI Label is certainly of interest (and was also briefly discussed in the past internally). Also a better roundtrip experience Label <-> Deploy is a thing that a lot of WG members feel is very important.
Maybe we could kill two birds with one stone and have a trilateral meeting to discuss that further (You idea with nnUnet/MONet and if/how this could be a driving use case for the better integration between Label and Deploy). @hshuaib90 What do you think?