CoolFace
Datasetpublic

pucpr-br/nestedclinbr

NestedClinBr Corpus NestedClinBr is a new corpus containing nested and discontinuous entities in Brazilian Portuguese clinical narratives. The main goal of NestedClinBr is to provide a human-annotated corpus that can be used for learning and evaluating different machine learning models to extract valuable medical information in the Portuguese language, in special nested and discontinuous entities, an important but less explored task. In the context of clinical NLP, the… See the full description on the dataset page: https://huggingface.co/datasets/pucpr-br/nestedclinbr.

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face
2likes86downloads
Dataset Card

NestedClinBr Corpus

<img src="./img/nestedlicnbr_example.png" alt="Example" width="80%">

NestedClinBr is a new corpus containing nested and discontinuous entities in Brazilian Portuguese clinical narratives.

The main goal of NestedClinBr is to provide a human-annotated corpus that can be used for learning and evaluating different machine learning models to extract valuable medical information in the Portuguese language, in special nested and discontinuous entities, an important but less explored task.

In the context of clinical NLP, the recognition of entities is commonly used for the identification of diseases, body parts, medications, and other relevant information, facilitating, for example, the detection of risk factors and medical decision-making. As NestedClinBr, although small, can contribute to the healthcare domain, it will be freely available to the research community.

Entities

Percentage of entities:

<img src="./img/nestedclinbr_entidades.png" alt="Entidades" width="40%">

Event types with their respective description and examples:

Entity typeDefinitionUMLS groupExamples (free translation)
ProblemMentions that differ from normal expected conditions, including the location (body part), characterization, and severity, when available in the text.DisordersInjury, chest pain, SAH, severe dyspnea on exertion.
TreatmentMentions relating to any procedure or intervention used to treat problems, including the dosage, in the case of drugs, and the location (body part), when available in the text.Chemicals & Drugs, Devices, ProceduresPacemaker, angioplasty, Enalapril 10 mg, mitral valve repair.
TestUsed to detect and evaluate problems (such as diagnostic procedures and physical examination), also including the location (body part), when available in the text.Phenomena, PhysiologyHDL, potassium, cardiac catheterization, myocardial scintigraphy.
AnatomyRefers to body location, region, organ or organ component.AnatomyHeart valves, left hemithorax, mitral.

IAA - Inter-Annotator Agreement

The final IAA value for entity types was 94.08%, a high F1-measure that represents a substantial agreement between annotators.

<img src="./img/nestedclinbr_iaa.png" alt="Inter-Annotator Agreement" width="40%">

Corpus Statistics

Number of documents, sentences, and tokens present in NestedClinBr:

TypeTrainingTestTotal
Documents10026126
Sentences2,2736932,966
Tokens17,1545,31022,464

Statistics of the NestedClinBr corpus:

ItemTrainingTestTotal
Problem---
Nested315 (72.92%)117 (27.08%)432
Discontinuous80 (65.04%)43 (34.96%)123
Total1,110 (77.19%)328 (22.81%)1,438
Entity avg. length2.62.5-
Treatment---
Nested50 (76.92%)15 (23.08%)65
Discontinuous4 (100%)0 (0%)4
Total761 (78.05%)214 (21.95%)975
Entity avg. length2.12.1-
Test---
Nested15 (71.43%)6 (28.57%)21
Discontinuous0 (0%)0 (0%)0
Total772 (75.98%)244 (24.02%)1,016
Entity avg. length1.21.3-
Anatomy---
Nested395 (75.38%)129 (24.62%)524
Discontinuous5 (83.33%)1 (16.66%)6
Total543 (73.58%)195 (26.42%)738
Entity avg. length1.41.3-
Overall---
Nested778 (74.45%)267 (25.55%)1,045
Discontinuous89 (66.92%)44 (33.08%)133
Total3,186 (76.46%)981 (23.54%)4,167
Percentage of entities vs ’O’ (balancing)30.8%30.1%-
Entity avg. length1.91.9-
Max. tokens per sentence192146-

How to download the corpus

Please visit the data/brat-files section of this repository to access the original corpus (NestedClinBr) in BRAT format.

To visualize and edit the annotations, consider installing the BRAT annotation tool. This tool allows you to explore nested and discontinuous entities in a user-friendly web interface.

Contributors

  • —Claudia Moro
  • —Elisa Terumi Rubel Schneider
  • —Emerson Cabrera Paraiso
  • —Paloma Martínez
  • —Yohan Bonescki Gumiel

Github: https://github.com/HAILab-PUCPR/NestedClinBr

How to cite

* soon *