Use standard terminology to define content of a segment

Specifying meaning of segment by label values or names is very limited and fragile. It is practically impossible to use consistent label values when the same data sets are used for multiple projects, usage of label names are too error-prone (due to spelling mistakes or differences in spelling), and meaning of labels names are often not well defined.

I would recommend identifying segments using a universally applicable standard terminology.

Since DICOM has to be supported anyway and DICOM terminology specification is so generic yet fairly minimal, it could be a good choice in MONAILabel (and in general in MONAI).

A short summary of how DICOM specifies a segment:

You need to specify a category and a type at a minimum. For example a tumor is defined as Morphologically Altered Structure/Mass. Or, liver is defined as Tissue/Liver. You can have a few more optional values, the complete list is this:

These values are not just stored as user-defined strings, because then you would need to define the exact meaning of that name, translations to different languages, etc., which would be a lot of work. Instead, existing standard terminologies (collection of terms with well-defined meaning) can be leveraged, such as SNOMED CT for clinical terms, or TA2 terminologia anatomica for anatomical terms. So the idea is to just use these existing terms by specifying the name of the terminology and the code value according to that terminology. The end result is that you store each value as 3 strings:

• coding scheme designator: defines the terminology, for example SCT (SNOMED CT)
• code value: code value in the chosen terminology, for example 4147007
• code meaning: this is optional, for convenience only, a displayable human-readable string that can be displayed, for example Mass

Slicer already uses this approach, user can choose from standard terms by a few clicks, and the chosen term is saved in to the segmentation file.

https://github.com/Slicer/Slicer/releases/download/docs-resources/terminology_selector_advanced.png

No files, tests, or entry points are named. Start by locating how MONAILabel currently defines and stores segment labels, then compare that flow with the proposed DICOM category, type, and terminology codes. Done would require an agreed design for standard terminology support and a defined migration or compatibility plan.