api: Include labels in DeID service model card

Extend the base model card schema to allow passing a mapping of labels,
i.e. concept IDs (e.g. CUIs) to names. This is particularly useful
when using the MedCAT model DeID service, which is typically configured
to serve models with a relatively small CDB that can be exposed in its
entirety as part of the model card. This information can be interesting
to the user (or client application), providing insight into the concepts
that the model is capable of identifying and by extension enabling them
to decide and control which of them should be redacted. The 'labels'
field is always included in the API's responses when the '/info'
endpoint is called, but is only populated with the mapping in a dict
format when the CMS instance is configured to use the MedCAT model DeID
service.

Signed-off-by: Phoevos Kalemkeris <phoevos.kalemkeris@ucl.ac.uk>

Author: phoevos

app/domain.py

@@ -138,6 +138,7 @@ class ModelCard(BaseModel):
     model_type: ModelType = Field(description="The type of the served model")
     model_description: Optional[str] = Field(description="The description about the served model")
     model_card: Optional[dict] = Field(default=None, description="The metadata of the served model")
+    labels: Optional[Dict[str, str]] = Field(default=None, description="The mapping of CUIs to names")
 
 
 class Entity(BaseModel):

app/model_services/medcat_model_deid.py

@@ -69,6 +69,7 @@ def info(self) -> ModelCard:
             model_type=ModelType.ANONCAT,
             api_version=self.api_version,
             model_card=model_card,
+            labels=self.model.cdb.cui2preferred_name,
         )
 
     def annotate(self, text: str) -> List[Annotation]:

tests/app/api/test_serving_common.py

@@ -345,6 +345,7 @@ def test_train_unsupervised_with_hf_hub_dataset(model_service, client):
         "model_description": "huggingface_ner_model_description",
         "model_type": ModelType.MEDCAT_SNOMED,
         "model_card": None,
+        "labels": None,
     })
     model_service.info.return_value = model_card
     model_service.train_unsupervised.return_value = (True, "experiment_id", "run_id")

tests/app/api/test_serving_hf_ner.py

@@ -42,6 +42,7 @@ def test_train_unsupervised_with_hf_hub_dataset(model_service, client):
         "model_description": "huggingface_ner_model_description",
         "model_type": ModelType.HUGGINGFACE_NER,
         "model_card": None,
+        "labels": None,
     })
     model_service.info.return_value = model_card
     model_service.train_unsupervised.return_value = (True, "experiment_id", "run_id")

tests/app/api/test_serving_trf.py

@@ -37,6 +37,7 @@ def test_readyz(model_service, client):
         "model_description": "deid_model_description",
         "model_type": ModelType.TRANSFORMERS_DEID,
         "model_card": None,
+        "labels": None,
     })
     model_service.info.return_value = model_card
 
@@ -49,6 +50,7 @@ def test_info(model_service, client):
         "model_description": "deid_model_description",
         "model_type": ModelType.TRANSFORMERS_DEID,
         "model_card": None,
+        "labels": None,
     })
     model_service.info.return_value = model_card