README.md

metadata

language:
  - es
pipeline_tag: token-classification
tags:
  - biomedical
  - ner
  - clinical
  - ehr
  - cardiology
  - nlp
  - edsnlp
  - caradioccc
license: apache-2.0
library_name: edsnlp
base_model:
  - PlanTL-GOB-ES/bsc-bio-ehr-es
model-index:
  - name: Aremaki/eds-ner-cardioccc
    results:
      - task:
          type: token-classification
        dataset:
          name: CardioCCC
          type: public
        metrics:
          - type: precision
            name: Token Scores / MEDICATION / Precision
            value: 0.93
          - type: recall
            name: Token Scores / MEDICATION / Recall
            value: 0.94
          - type: f1
            name: Token Scores / MEDICATION / F1
            value: 0.93
          - type: precision
            name: Token Scores / PROCEDURE / Precision
            value: 0.85
          - type: recall
            name: Token Scores / PROCEDURE / Recall
            value: 0.85
          - type: f1
            name: Token Scores / PROCEDURE / F1
            value: 0.85
          - type: precision
            name: Token Scores / DISEASE / Precision
            value: 0.82
          - type: recall
            name: Token Scores / DISEASE / Recall
            value: 0.82
          - type: f1
            name: Token Scores / DISEASE / F1
            value: 0.82
          - type: precision
            name: Token Scores / SYMPTOM / Precision
            value: 0.8
          - type: recall
            name: Token Scores / SYMPTOM / Recall
            value: 0.81
          - type: f1
            name: Token Scores / SYMPTOM / F1
            value: 0.8

EDS-NER-CARDIOCCC

This repository contains the final NER model trained on the CardioCCC dataset. CardioCCC is a collection of cardiology clinical case reports used for domain adaptation. Clinical case reports are a textual genre in medicine that describe a patient’s medical history, symptoms, diagnosis, and treatment in detail.

The model implementation is based on EDS-NLP, a library developed by the data science team of the Greater Paris University Hospitals (AP-HP) for clinical natural language processing.

The entities that are detected are listed below.

Label	Description
MEDICATION	Names of drugs or chemical substances used in treatment, e.g., Metformina.
PROCEDURE	Medical or surgical procedures performed on a patient, e.g., biopsia, radiografía.
DISEASE	Diagnosed diseases or medical conditions, e.g., diabetes mellitus, hipertensión.
SYMPTOM	Reported signs or symptoms experienced by a patient, e.g., fiebre, dolor de cabeza.

Quickstart

1. Install the latest version of edsnlp

   pip install "edsnlp[ml]" -U
   

2. Load the model

   import edsnlp
   
   nlp = edsnlp.load("Aremaki/eds-ner-cardioccc", auto_update=True)
   doc = nlp(
       "La paciente con diabetes mellitus "
       "presentó fiebre y se le realizó "
       "una radiografía antes de tomar metformina. "
   )
   
   for ent in doc.ents:
       print(ent, ent.label_)
   

To apply the model on many documents using one or more GPUs, refer to the documentation of edsnlp.

Metrics

Token Scores	Precision	Recall	F1
MEDICATION	93.0	94.0	93.0
PROCEDURE	85.0	85.0	85.0
DISEASE	82.0	82.0	82.0
SYMPTOM	80.0	81.0	80.0

Installation to reproduce

If you'd like to reproduce eds-ner-cardioccc's training or contribute to its development, you should first clone it:

git clone https://github.com/Aremaki/eds_ner_cardioccc.git
cd eds_ner_cardioccc

Acknowledgement

We would like to thank the Life science team at the Barcelona Supercomputing Center (BSC) who designed the CardioCCC dataset and trained the base model bsc-bio-ehr-es We would like to thank the data science team of the Greater Paris University Hospitals (AP-HP) who developped the EDS-NLP library.