ImagingDataCommons / CloudSegmentator

Medical imaging segmentation workflows for FireCloud (Terra) and Seven Bridges Cancer Genomics Cloud
Apache License 2.0
3 stars 2 forks source link

Populate units of radiomics features based on input modality #40

Open fedorov opened 8 months ago

fedorov commented 8 months ago

I.e., "no units" for MR and Hounsfeld units for CT. This should probably be done in the processing task. This is a follow up on #36.

vkt1414 commented 8 months ago

In my opinion, this may not apply to this repo as it stands now. We never worked with any other modality than CT on Terra. I think we just need to be mindful of the units if we ever extract first-order features from modalities with arbitrary units while developing a config map.

fedorov commented 7 months ago

We never worked on it, but there is nothing to stop user from selecting DICOM series that has modality different from CT and launch the workflow. It is also not impossible that TotalSegmentator would produce something for a non-CT input. Maybe for now we can add a check, and if the input series is not CT, abort processing? We have no control over what user will throw at the workflow.

vkt1414 commented 7 months ago

I see your point but the radiomics notebook/functions are highly customized to process the segmentation files from the Totalsegmentator inference step of any of our workflows (oneVM, twoVM, and threeVM). I cannot imagine how one can get to radiomics step if they start with data from other modalities.

fedorov commented 7 months ago

As agreed, for now we should abort the workflow is the input modality is not CT.

For the future, if we want to have a modality-aware workflow, may need to consider, but I am afraid in the general case, there is no reliable way to know the units: https://github.com/ImagingDataCommons/CloudSegmentator/pull/34#issuecomment-1881111780.