Cross-lingual Medical Entity Linking Recommenders for INCEpTION

Simple integration of pre-configured xMEN pipelines as recommenders for (biomedical) entity linking in INCEpTION.

German SNOMED CT Quickstart (with Docker)

• Download pre-computed xMEN index for SNOMED CT and extract contents into xmen_index:

  • HPI Nextcloud Link (password: name of this repository)

• docker run -m=12g -p <PORT_NUMBER>:5000 -v "$(pwd)"/xmen_index/index:/index/ ghcr.io/hpi-dhc/xmen-inception-recommender:main

Replace <PORT_NUMBER> with the port you want to use for the external recommender in INCEpTION. It will be available on http://localhost:<PORT_NUMBER>/xmen_snomed

Create xMEN KB and Index from Scratch

• Adapt snomed_german.yaml as needed
• Create KB and indices (takes about 1 hour with a GPU)
  • git clone https://github.com/hpi-dhc/xmen
  • xmen dict snomed_german.yaml --code xmen/examples/dicts/umls_source.py
  • xmen index snomed_german.yaml --all --overwrite

Running the Recommender without Docker

Install dependencies

• pip install -r requirements.txt (see here for known issues during installation of xmen)

Run the recommender

• Start the Ariadne Server (on http://localhost:5000/xmen_snomed):
  • python run_snomed_german_recommender.py ./xmen_index/index