Discrepancy between segmentation threshold and first order radiomics


Thanks to the advice of Mr. Andras Lasso, I’ve been able to segment the pericardium of the heart (using the Surface Cut option from Slicer). I’ve then applied a threshold to only select the fatty tissue (HU -150 to -50) and it is corretly identified as green pixels on the attached image.
I’ve also checked that this segmentation was correct looking at the label statistics. However, once I run the radiomics on this segment, I noticed that the first order minimum was -346 and the maximum was 73 (outside the range of my initial threshold).

Is this normal? What could cause this difference?
Thank you

Perhaps I’ve found the solution.
I’ve been playing with the options and I’ve noticed that:

  • when I resample with resampled voxel size 2,2,2 → the result is what is seen above
  • when I do not resample → the first order minimum and maximum are exactly -150 and -50

Sounds like the issue indeed. A potential fix is enabling the resegmentationRange in PyRadiomcs this excludes voxels outside the specified range and works like thresholding, but is applied after resampling.