Awesome

Biomedical and clinical language model for Spanish 🏥

Official repository of the Barcelona Supercomputing Center (BSC) biomedical and clinical pretrained language models for Spanish.

Models 🤖

We released our best models under the Apache License 2.0 to encourage the development of Spanish NLP applications in the biomedical and clinical domains.

Pretrained models:

We trained two models and make them available on HuggingFace' models hub at the following links:

• bsc-bio-es: pretrained language model suitable for biomedical and clinical documents
• bsc-bio-ehr-es: pretrained language model suitable for EHR documents and clinical noteshttps://huggingface.co/PlanTL-GOB-ES/longformer-base-4096-biomedical-clinical-es

Additionally, a Longformer version has been trained to be able to work with tasks that require larger contexts, such as clinical coding. The corpora used is the same as below. Model can also be found on Hugginface:

• longformer-base-4096-biomedical-clinical-es: Longformer version of the roberta-base-biomedical-clinical-es model.

Fine-tuned NER models:

We fine-tuned our models for the Named Entity Recognition (NER) task and make the best models available on HuggingFace' models hub at the following links:

• bsc-bio-ehr-es-cantemist: NER model for tumor morphology on clinical cases
• bsc-bio-ehr-es-pharmaconer: NER model for substances, compounds and proteins on clinical cases

Corpora 📃

The training corpora are composed of several biomedical corpora in Spanish, collected from publicly available corpora and crawlers, and a real-world clinical corpus collected from more than 278K clinical documents and notes. To obtain a high-quality training corpus while retaining the idiosyncrasies of the clinical language, a cleaning pipeline has been applied only to the biomedical corpora, keeping the clinical corpus uncleaned. Essentially, the cleaning operations used are:

• data parsing in different formats
  • sentence splitting
  • language detection
  • filtering of ill-formed sentences
  • deduplication of repetitive contents
  • keep the original document boundaries

Then, the biomedical corpora are concatenated and further global deduplication among the biomedical corpora have been applied. The results are a clinical corpus of about 91M tokens and a biomedical corpus of about 963M tokens. Eventually, the clinical corpus is concatenated to the cleaned biomedical corpus resulting in a medium-size biomedical-clinical corpus for Spanish composed of more than 1B tokens. The table below shows some basic statistics of the individual cleaned corpora:

Evaluation ✅

The models have been fine-tuned on the the Named Entity Recognition (NER) tasks using three clinical NER datasets, such as, PharmaCoNER, CANTEMIST and ICTUSnet. We addressed the NER task as a token classification problem using a standard linear layer along with the BIO tagging schema. We compared our models with the general-domain Spanish roberta-base-bne, the general-domain multilingual model that supports Spanish mBERT, the domain-specific English model BioBERT, and three domain-specific models based on continual pre-training, mBERT-Galén, XLM-R-Galén and BETO-Galén. The table below shows the F1 scores obtained:

NER usage example ⚗️

The following example, show how to predict entities the bsc-bio-ehr-es model fine-tuned on the PharmaCoNER dataset:

from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

tokenizer = AutoTokenizer.from_pretrained("PlanTL-GOB-ES/bsc-bio-ehr-es-pharmaconer")

model = AutoModelForTokenClassification.from_pretrained("PlanTL-GOB-ES/bsc-bio-ehr-es-pharmaconer")

pipe =  pipeline(task='token-classification', model='PlanTL-GOB-ES/bsc-bio-ehr-es-pharmaconer')

sentence = 'Por el hallazgo de múltiples fracturas por estrés, se procedió a estudio en nuestras consultas, realizándose análisis con función renal, calcio sérico y urinario, calcio iónico, magnesio y PTH, que fueron normales.'
results = pipe(sentence)

print(results)

# Expected output with the predicted entities:

[
  {'word': 'Ġcalcio', 'score': 0.9963880181312561, 'entity': 'B-NORMALIZABLES', 'index': 24, 'start': 137, 'end': 143}, {'word': 'Ġcalcio', 'score': 0.9965023398399353, 'entity': 'B-NORMALIZABLES', 'index': 29, 'start': 163, 'end': 169}, {'word': 'Ġmagnesio', 'score': 0.996299147605896, 'entity': 'B-NORMALIZABLES', 'index': 32, 'start': 178, 'end': 186}, {'word': 'ĠPTH', 'score': 0.9950509667396545, 'entity': 'B-PROTEINAS', 'index': 34, 'start': 189, 'end': 192}
]

How to fine-tune ️▶️

1. First, clone the project:

   git clone https://github.com/PlanTL-GOB-ES/lm-biomedical-clinical-es/blob/master/README.md

2. Set up a Python virtual environment and activate it

   python3.8 -m venv venv; source venv/bin/activate

3. Install all the required python packages in the virtual environment running the following line from the project main folder: pip install -r requirements.txt.

4. Finally, run the following command on the project main folder: bash ner.sh [MODEL_NAME] [DATASET_NAME] [SEED], where

• [MODEL_NAME]: HuggingFace' model name of the pretrained model you want to use.
• [DATASET_NAME]: HuggingFace' dataset name of the NER dataset to use. Can be chosen from "PlanTL-GOB-ES/cantemist-ner" and "PlanTL-GOB-ES/pharmaconer", depending on the dataset you want to use.
• [SEED]: the seed you want to use. This allows to reproduce the same results.

The ner.sh script fine-tune a pretrained language model for the NER task applying a linear classification head. By default, the fine-tuning run for 10 epochs with an evaluation on the development set at every epoch. The model achieving the best performance on the development set is selected as the final model and evaluated on the test set. The best trained model is store in a output path of the type ./output/model-$model_name/dataset-$dataset_name/seed-$seed along with the checkpoints folders and the tensorboard data (inside the tb directory).

For example, to fine-tune the bsc-bio-ehr-es language model on the PharmaCoNER dataset, run the command:

bash ner.sh PlanTL-GOB-ES/bsc-bio-ehr-es PlanTL-GOB-ES/pharmaconer 42

Cite 📣

If you use these models, please cite our work:

@inproceedings{carrino-etal-2022-pretrained,
    title = "Pretrained Biomedical Language Models for Clinical {NLP} in {S}panish",
    author = "Carrino, Casimiro Pio  and
      Llop, Joan  and
      P{\`a}mies, Marc  and
      Guti{\'e}rrez-Fandi{\~n}o, Asier  and
      Armengol-Estap{\'e}, Jordi  and
      Silveira-Ocampo, Joaqu{\'\i}n  and
      Valencia, Alfonso  and
      Gonzalez-Agirre, Aitor  and
      Villegas, Marta",
    booktitle = "Proceedings of the 21st Workshop on Biomedical Language Processing",
    month = may,
    year = "2022",
    address = "Dublin, Ireland",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.bionlp-1.19",
    doi = "10.18653/v1/2022.bionlp-1.19",
    pages = "193--199",
    abstract = "This work presents the first large-scale biomedical Spanish language models trained from scratch, using large biomedical corpora consisting of a total of 1.1B tokens and an EHR corpus of 95M tokens. We compared them against general-domain and other domain-specific models for Spanish on three clinical NER tasks. As main results, our models are superior across the NER tasks, rendering them more convenient for clinical NLP applications. Furthermore, our findings indicate that when enough data is available, pre-training from scratch is better than continual pre-training when tested on clinical tasks, raising an exciting research question about which approach is optimal. Our models and fine-tuning scripts are publicly available at HuggingFace and GitHub.",
}

License

This project is licensed under the Apache License, Version 2.0

Copyright (c) 2021 Secretaría de Estado de Digitalización e Inteligencia Artificial (SEDIA)

Funding

This work was funded by the Spanish State Secretariat for Digitalization and Artificial Intelligence (SEDIA) within the framework of the Plan-TL.

Disclaimer

The models published in this repository are intended for a generalist purpose and are available to third parties. These models may have bias and/or any other undesirable distortions.

When third parties, deploy or provide systems and/or services to other parties using any of these models (or using systems based on these models) or become users of the models, they should note that it is their responsibility to mitigate the risks arising from their use and, in any event, to comply with applicable regulations, including regulations regarding the use of artificial intelligence.

In no event shall the owner of the models (SEDIA – State Secretariat for digitalization and artificial intelligence) nor the creator (BSC – Barcelona Supercomputing Center) be liable for any results arising from the use made by third parties of these models.

Los modelos publicados en este repositorio tienen una finalidad generalista y están a disposición de terceros. Estos modelos pueden tener sesgos y/u otro tipo de distorsiones indeseables.

Cuando terceros desplieguen o proporcionen sistemas y/o servicios a otras partes usando alguno de estos modelos (o utilizando sistemas basados en estos modelos) o se conviertan en usuarios de los modelos, deben tener en cuenta que es su responsabilidad mitigar los riesgos derivados de su uso y, en todo caso, cumplir con la normativa aplicable, incluyendo la normativa en materia de uso de inteligencia artificial.

En ningún caso el propietario de los modelos (SEDIA – Secretaría de Estado de Digitalización e Inteligencia Artificial) ni el creador (BSC – Barcelona Supercomputing Center) serán responsables de los resultados derivados del uso que hagan terceros de estos modelos.