Skip to content

Configurations

Annotator

The MiADE processor is configured by a yaml file that maps a human-readable key for each of your models to a MedCAT model ID and an MiADE Annotator. The config file must be in the same folder as the MedCAT models.

Required

  • models: The models section maps human-readable key-value pairing to the MedCAT model ID to use in MiADE
  • annotators: The annotators section maps human-readable key-value pairing to Annotator processing classes to use in MiADE

Optional

  • lookup_data_path: Specifies the lookup data to use. If None a default MiADE set will be used.
  • negation_detection: negex (default rule-based algorithm) or None (use MetaCAT models)
  • structured_list_limit: Specifies the maximum number of concepts detected in a structured paragraph section. If there are more than the set number of concepts in a structured list, then concepts detected in prose are ignored (prioritises concepts detected in structured lists over free-form text to avoid returning too many irrelevant concepts). Default 100 so this feature is essentially disabled.
  • disable: Disable any specific postprocessing pipeline components - the usage here is similar to spaCy pipelines.
  • add_numbering: Option to add a number prefix to the concept display names e.g. "01 Diabetes"

config.yaml
models:
  problems: f25ec9423958e8d6
  meds/allergies: a146c741501cf1f7
annotators:
  problems: ProblemsAnnotator
  meds/allergies: MedsAllergiesAnnotator
general:
  problems:
    lookup_data_path: ./custom_lookup_data/
    structured_list_limit: 0  # setting as 0 will ignore all concepts found in prose
    add_numbering: True
  meds/allergies:
    disable: ["vtm_converter"]
The default configurations for annotators are defined below:

Bases: BaseModel

Source code in miade/utils/annotatorconfig.py 
 5
 6
 7
 8
 9
10
11
class AnnotatorConfig(BaseModel):
    lookup_data_path: Optional[str] = None
    negation_detection: Optional[str] = "negex"
    structured_list_limit: Optional[int] = 100
    refine_paragraphs: Optional[bool] = True
    disable: List[str] = []
    add_numbering: bool = False

Lookup Table

Lookup tables are used to convert and filter concepts in the MiADE postprocessing steps for ProblemsAnnotator and MedsAllergiesAnnotator. We have packaged default lookup data (curated and used at UCLH) with MiADE for sample use.

For a more detailed explanation on the creation and format of the lookup data, check out miade-dataset.

To customise your own lookup tables, you can pass in a directory which contains your lookup data in the config.yaml lookup_data_path field. Note you currently need to have ALL of the required lookup data in your directory (this will be improved in the future).

Problems 

negated.csv
historic.csv
suspected.csv
problem_blacklist.csv

MedsAllergies 

reactions_subset.csv
allergens_subset.csv
allergy_type.csv
valid_meds.csv
vtm_to_text.csv
vtm_to_vmp.csv