!nnUNetv2_plan_and_preprocess -d 1: RuntimeError: Background label not declared (remember that this should be label 0!)

[jimsyjohnson]

Hello nnU-Net community,

I am encountering a persistent issue while trying to run nnUNetv2_plan_and_preprocess on my custom 3D medical image segmentation dataset in Google Colab.

The command I am running is:
!nnUNetv2_plan_and_preprocess -d 1 --verify_dataset_integrity`

Initially, I received the error:
RuntimeError: Background label not declared (remember that this should be label 0!)

My setup:

• Environment: Google Colab
• Dataset: Custom 3D medical image segmentation dataset (Pancreas Cancer)
• Modalities: 1 modality (CT)
• Labels: 3 labels declared: 0 (background), 1 (pancreas), 2 (tumor)

Troubleshooting steps I have already taken:

1. Verified dataset.json: I have confirmed that the dataset.json file is located at /content/drive/MyDrive/my_colab_project/nnUNet_raw/Dataset001_PancreasCancer/dataset.json and its content correctly includes "labels": {"0": "background", "1": "pancreas", "2": "tumor"}. The format matches the documentation.
2. Checked Label Values in NIfTI files: I have programmatically checked the unique pixel/voxel values in ALL my training label files (labelsTr/*.nii.gz) and confirmed that the value 0 is present in every file. I also verified that labels 1 and 2 are present in sample files.
3. Tried verify_dataset_integrity flag: The error occurs both with and without the --verify_dataset_integrity flag.
4. Tried --clean flag: Added --clean to clear potential cache, but the error persists.
5. Tried subset dataset: Created a small subset (Dataset ID 999) with only a few image/label pairs and ran the command on it; the error still occurred.
6. Tried integer label data type: Created a subset where label NIfTI files were explicitly converted to integer data type (int16) before preprocessing; the error still occurred.

I am puzzled as to why nnU-Net is throwing the "Background label not declared" or the subsequent ValueError when all checks indicate the background label (0) is correctly declared and present in the data.

Any insights or suggestions from the community would be greatly appreciated.

Thank you!