Model Card for Cardioner Medroberta.Nl Multilabel

This a medroberta.nl base model finetuned for span classification. This specific model is the average of the best checkpoints per fold over a ten-fold cross-validation. For this model we used the IOB-tagged. Using the IOB-tagging schema facilitates the aggregation of predictions over sequences.

For the chunking we used paragraph based chunking, and we assumed the maximum context length of the base model, i.e. 512 tokens.

Expected input and output

The input should be a string with Dutch cardio clinical text.

CardioNER_MedRoBERTa.nl_multilabel is a muticlass span classification model. The classes that can be predicted are disease, medication, procedure and symptom.

Extracting span classification from CardioNER_MedRoBERTa.nl_multilabel

The following script converts a string of <512 tokens to a list of span predictions.

from transformers import pipeline

le_pipe = pipeline('ner', 
                    model=model, 
                    tokenizer=model, aggregation_strategy="simple", 
                    device=-1)

named_ents = le_pipe(SOME_TEXT)

To process a string of arbitrary length you can split the string into sentences or paragraphs using e.g. pysbd or spacy(sentencizer) and iteratively parse the list of with the span-classification pipe.

Data description

CardioCCC; Manually annotated parallel-language corpus for the clinical cardiology domain

On a 10-fold cross-validations the multilabel metrics are:

Acknowledgement

This is part of the DT4H project.

Doi and reference

For more details about training/eval and other scripts, see CardioNER github repo. and for more information on the background, see Datatools4Heart Huggingface/Website