Issue with Prediction Using Single Numpy Array: RuntimeError in nnUNetPredictor

[RewaaHummedi]

Description: I encountered an issue while attempting to perform prediction using a single Numpy array with nnUNetPredictor. Here are the details of the problem along with the steps to reproduce it:

Code Snippet:

Output:

Error:

Issue Details:

• Problem: The error suggests a mismatch between the expected input dimensions (3 channels) and the actual input dimensions (1 channel).
• Context: The code attempts to predict using a single numpy array derived from a .tif image, processed through SimpleITKIO.
• Expected Outcome: Successful prediction using nnUNetPredictor with a single image numpy array.

Request for Assistance:

• Clarification: Could you please explain why this error occurs in the context of nnUNetPredictor?
• Resolution: What steps can I take to ensure compatibility and successful prediction using a single numpy array?

Thank you for your guidance on resolving this issue.

[Lars-Kraemer]

Hey @RewaaHummedi,

when using the nnUNetPredictor you should load the data in the same way as it was done during training. If you are using SimpleITKIO take a look here:

https://github.com/MIC-DKFZ/nnUNet/blob/master/nnunetv2/imageio/simpleitk_reader_writer.py#L30

at first glance, the following line is missing from your script: npy_image = npy_image[None]

If this is not working please ensure your data has the same format as your train data and share your dataset.json and the plans file.

Best, Lars