MedCAT can be used to extract information from Electronic Health Records (EHRs) and link it to biomedical ontologies like SNOMED-CT and UMLS. Paper on arXiv.
- New Release [1. August 2021]: Upgraded MedCAT to use spaCy v3, new scispaCy models have to be downloaded - all old CDBs will work without any changes.
- New Feature and Tutorial [8. July 2021]: Integrating 🤗 Transformers with MedCAT for biomedical NER+L
- General [1. April 2021]: MedCAT is upgraded to v1, unforunately this introduces breaking changes with older models (MedCAT v0.4), as well as potential problems with all code that used the MedCAT package. MedCAT v0.4 is available on the legacy branch and will still be supported until 1. July 2021 (with respect to potential bug fixes), after it will still be available but not updated anymore.
- Paper: What’s in a Summary? Laying the Groundwork for Advances in Hospital-Course Summarization
- (more...)
A demo application is available at MedCAT. This was trained on MIMIC-III and all of SNOMED-CT.
A guide on how to use MedCAT is available in the tutorial folder. Read more about MedCAT on Towards Data Science.
- MedCATtrainer - an interface for building, improving and customising a given Named Entity Recognition and Linking (NER+L) model (MedCAT) for biomedical domain text.
- MedCATservice - implements the MedCAT NLP application as a service behind a REST API.
- iCAT - A docker container for CogStack/MedCAT/HuggingFace development in isolated environments.
- Upgrade pip
pip install --upgrade pip - Install MedCAT
- For macOS/linux:
pip install --upgrade medcat - For Windows (see PyTorch documentation):
pip install --upgrade medcat -f https://download.pytorch.org/whl/torch_stable.html
- Get the scispacy models:
pip install https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.4.0/en_core_sci_md-0.4.0.tar.gz
or
pip install https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.4.0/en_core_sci_lg-0.4.0.tar.gz
-
Downlad the Vocabulary and CDB from the Models section bellow
-
Quickstart:
from medcat.vocab import Vocab
from medcat.cdb import CDB
from medcat.cat import CAT
# Load the vocab model you downloaded
vocab = Vocab.load(vocab_path)
# Load the cdb model you downloaded
cdb = CDB.load('<path to the cdb file>')
# Create cat - each cdb comes with a config that was used
#to train it. You can change that config in any way you want, before or after creating cat.
cat = CAT(cdb=cdb, config=cdb.config, vocab=vocab)
# Test it
text = "My simple document with kidney failure"
doc_spacy = cat(text)
# Print detected entities
print(doc_spacy.ents)
# Or to get an array of entities, this will return much more information
#and usually easier to use unless you know a lot about spaCy
doc = cat.get_entities(text)
print(doc)
# To train on one example
_ = cat(text, do_train=True)
# To train on a iterator over documents
data_iterator = <your iterator>
cat.train(data_iterator)
#Once done, save the new CDB
cat.cdb.save(<save path>)from medcat.meta_cat import MetaCAT
# Assume we have a CDB and Vocab object from before
# Download the mc_status model from the models section below and unzip it
mc_status = MetaCAT.load("<path to the unziped mc_status directory>")
cat = CAT(cdb=cdb, config=cdb.config, vocab=vocab, meta_cats=[mc_status])
# Now annotate a document, it will have the meta annotation 'status'
doc = cat.get_entities(text)A basic trained model is made public for the vocabulary and CDB. It is trained for the ~ 35K concepts available in MedMentions.
Vocabulary Download - Built from MedMentions
CDB Download - Built from MedMentions
MetaCAT Status Download - Built from a sample from MIMIC-III, detects is an annotation Affirmed (Positve) or Other (Negated or Hypothetical)
(Note: This was compiled from MedMentions and does not have any data from NLM as that data is not publicaly available.)
If you have access to UMLS or SNOMED-CT and can provide some proof (a screenshot of the UMLS profile page is perfect, feel free to redact all information you do not want to share), contact us - we are happy to share the pre-built CDB and Vocab for those databases.
Entity extraction was trained on MedMentions In total it has ~ 35K entites from UMLS
The vocabulary was compiled from Wiktionary In total ~ 800K unique words
A big thank you goes to spaCy and Hugging Face - who made life a million times easier.
@ARTICLE{Kraljevic2021-ln,
title="Multi-domain clinical natural language processing with {MedCAT}: The Medical Concept Annotation Toolkit",
author="Kraljevic, Zeljko and Searle, Thomas and Shek, Anthony and Roguski, Lukasz and Noor, Kawsar and Bean, Daniel and Mascio, Aurelie and Zhu, Leilei and Folarin, Amos A and Roberts, Angus and Bendayan, Rebecca and Richardson, Mark P and Stewart, Robert and Shah, Anoop D and Wong, Wai Keong and Ibrahim, Zina and Teo, James T and Dobson, Richard J B",
journal="Artif. Intell. Med.",
volume=117,
pages="102083",
month=jul,
year=2021,
issn="0933-3657",
doi="10.1016/j.artmed.2021.102083"
}