Finetune nnUNet with new classes

[GauthierDot]

Hi all,

We are using nnUNet framework with great results in our research about dento-maxillo-facial segmentation, and recently released an extension for 3D Slicer (https://github.com/gaudot/SlicerDentalSegmentator). Thank you again for this amazing tool.

I have a small question. Say we have a nnUNet model_1 trained on Dataset_1 labelled with classes 1, 2, 3, 4, 5 and we have a Dataset_2 (with new images) labelled with classes 6, 7, 8. We would like to finetune our model_1 with Dataset_2 to have a model able to segment classes 1 to 8.

From the doc I read it looks like nnUNet does not offer this option at the moment. Could you confirm it to me ? Is there a workaround we could try ?

Thank you very much,
Gauthier