Catch error in case of image/mask geometry mismatch #252
Conversation
radiomics/imageoperations.py
Outdated
| lsif.Execute(imageNode, maskNode) | ||
| except RuntimeError as e: | ||
| if "Both images for LabelStatisticsImageFilter don't match type or dimension!" in e.message: | ||
| logger.error('Image/Mask datatype or size mismatch. Potential solution: enable resampling') |
There was a problem hiding this comment.
I think it would be very helpful to point to the specific section of documentation on how this can be done!
|
Would be good to also update the documentation in the same PR to clarify the instructions about how to implement the suggestions made in the error messages. |
|
@JoostJM sorry, I am still missing the location in the documentation that specifies: 1) how to toggle resampling of the mask to the image raster; 2) how to change tolerance for SimpleITK geometry comparison operations. The comments contributed in the commit below provide explanation that would require modifying the code, or running an extra script. I thought you said it is possible to control those options from the yml configuration file - is this correct? Also, where in the documentation all the parameters that can be passed using the configuration file described? I think it would be nice to have those in one location. What do you think? |
|
@fedorov, it is currently only possible to control resampling in general (i.e. resampling of both image and mask) from the yml file. However, I will add to parameters to control the global tolerance used by SimpleITK and a parameter to specify wheter to try to resample the mask if that geometry does not match. |
|
Moreover, I am restructuring the documentation a bit, by adding documentation on customizing the extraction to the "usage" section, with additionally some documentation on setting up logging and how the parameter file works. Most of it is already documented, but at the start of the feature extractor module, which maybe not the best location. |
|
@fedorov, this last commit only represents the restructuring of the documentation, still working on the 2 paramters I want to add. |
Catch the error in checkMask and print out user-friendly error message. Then return None to prevent further feature extraction with this invalid image/mask combination.
Add documentation on what goes wrong in case of geometry errors and how to fix it. Also include reference to this in the error message that is logged. Add the SimpleITK stacktrace as a debug-level log message in case of geometry errors. Additionally, add more extensive documentation on how to set up logging (both to output and/or log file) in interactive and commandline usage of PyRadiomics. Remove global PyRadiomics Function `debug`, the functionality exposed by this function is already covered by `setVerbosity`.
Add section on customizing the extraction in the "Usage" section. Here, specify what and how the extraction can be customized in PyRadiomics. For this, move the documentation of the possible settings from the docstring in the featureextractor module to this new section. Also move the documentation that explains how the parameter file works. Change the name of `self.kwargs` to `self.settings` in the feature extractor module, this better reflects what that variable holds. Make the distances setting only affect the GLCM (distance to neighbor should always be 1 in GLRLM and GLSZM).
JoostJM
force-pushed
the
checkMask-update
branch
5 times, most recently
from
b3a4ccc
to
53b394c
Compare
Add two parameter to specify how to handle geometry mismatch between image and mask. `geometryTolerance` controls the default global tolerance used by SimpleITK and set in ProcessObject. `correctMask` will attempt to resample the mask to the image geometry if a mismatch occurs. Prior to resampling, the mask is checked for ROI validity. Update documentation accordingly.
The helloRadiomics notebook was failing due to an error that was thrown by the Click progressbar. Remove the Click progressbar example, as it also does not render the progressbar if the output is not a terminal (as is the case in a jupyter notebook).
Catch the error in checkMask and print out user-friendly error message. Then return None to prevent further feature extraction with this invalid image/mask combination.
Fixes #251