Probabilities contain NaN while using my nifti input

[josedolz]

Probabilities contain NaN while using my nifti input

[josedolz]

Hi @divinediu

We have found that not having ROI images causes NaN values during the training due to the sampling strategy. Basically, if there is no ROI we sample from all the image, which may produce the NaN values for empty (black) sub patches. There are some easy fixes, but we didn't have time to modify it. A simple solution is to generate a ROI by yourself and include it in the training folder. A simple thresholding should be enough for this purpose.

Jose

[divinediu]

Thanks a lot for your quick response, I am not sure how to generate a quick ROI. Could you refer me to something to look up, please?

[josedolz]

I included also some code for that

LiviaNET/src/generateROI.py

Line 1 in 7a53f6f

"""

[divinediu]

I cannot thank you enough! I will try this out! :)