Package snomed mapping as a totalsegmentator package resource

[jcfr]

• Organize snomed mapping resource into a totalsegmentator namespace package
• Ensure snomed mapping is installed alongside totalsegmentator package. See https://setuptools.pypa.io/en/latest/userguide/datafiles.html

To retrieve the location after installing totalsegmentator, the following may be done:

from importlib.metadata import distribution
from pathlib import Path


def _lookup(path: str, optional: bool = False) -> Path | None:
    """Support editable installation by looking up path using distribution API."""
    files = distribution("idc_index_data").files
    if files is not None:
        for _file in files:
            if str(_file) == path:
                return Path(str(_file.locate())).resolve(strict=True)
    if optional:
        return None

    msg = f"Failed to lookup '{path}`."
    raise FileNotFoundError(msg)


TOTALSEGMENTATOR_SNOMED_MAPPING_FILEPATH: Path | None = _lookup("resources/totalsegmentator_snomed_mapping.csv")

We could also look into adding the code above directly into totalsegmentator/__init__.py

cc: @fedorov @lassoan

[jcfr]

For reference, here is the output reported when building the wheel locally:

[fedorov]

Very cool, thank you @jcfr! 🥇

@wasserth for the background, the motivation for this is to make it easier to reuse this mapping by the users of the package. Currently, there is a duplicate out-of-sync mapping in SlicerTotalSegmentator extension. Jointly with #324 we are trying to consolidate and remove duplication to simplify maintenance.

[wasserth]

Thanks for this PR. This sounds reasonable. I am not very familiar with adding data files to the package.
I read that when doing include_package_data=True, setuptools will include the files specified in MANIFEST.in in the package distribution. But in this case there is no MANIFEST.in file. So how does setuptools know that it has to add the snomed csv file?

[jcfr]

Thanks for the review 🙏

But in this case there is no MANIFEST.in file. So how does setuptools know that it has to add the snomed csv file?

This was on oversight¹, the settings in setup.py was updated.

Footnotes

1. While experimenting, I did not removed a previously created directory (TotalSegmentator.egg-info) which led me to over simplify the approach. ↩

[wasserth]

This looks better. Thank you!