Install and run¶
Choose the path that matches how you intend to use MedDeID:
| Goal | Start here |
|---|---|
| De-identify a note or a batch of files | Run MedDeID locally |
| Add de-identification to a Python application | Use the Python API |
| Explore the browser interface or HTTP API | Try MedDeID in your browser |
| Operate a shared institutional service | Production deployment |
Training, evaluation, and annotation tools are separate workflows linked at the end of this page.
Run MedDeID locally¶
This is the shortest path for individual notes, scripts, and batches. It requires Python 3.10 or newer but does not require writing Python code.
Install¶
meddeid models lists the available models and their validation scope. Select
one explicitly with --model; it is downloaded once and used locally.
Choose Dutch or English¶
The current public Dutch model declares nl-BE; the English model supports
en-GB and en-US.
Use the Python API¶
from meddeid import Deidentifier
deidentifier = Deidentifier.from_pretrained(
"stighellemans/meddeid-dutch-synth"
)
result = deidentifier("Patiënt Alex Voorbeeld kwam op controle.")
print(result.deid_text)
print(result.spans)
deidentifier.close()
Process a batch of notes¶
meddeid batch documents.jsonl \
--output predictions.jsonl \
--model stighellemans/meddeid-dutch-synth
The batch command keeps document order and records how the results were produced.
Try MedDeID in your browser¶
Install and start Docker Desktop. Clone MedDeID once:
Choose the language you want to test:
When MedDeID is ready, the browser opens with the API key already filled in.
Paste a note and select De-identify. The English model also lets you switch
between en-GB and en-US in the browser.
API documentation is available at http://127.0.0.1:8000/docs.
Stop the service with ./scripts/stop-local.sh.
Developers testing source changes can add --build. For a shared service, see
production deployment.
Reproducible or offline runs¶
Inspect and record the exact model version when results must be reproducible:
To stage and manage a model bundle yourself, pass its local directory with
--model. The local inference
guide covers
revision pinning, complete local bundles, and air-gapped environments.