Description
This Named Entity Recognition model is intended for detecting relevant entities from Spanish medical texts and trained by using MedicalNerApproach annotator that allows to train generic NER models based on Neural Networks. The model detects Negation Trigger (NEG), Negation Scope (NSCO), Uncertainty Trigger (UNC) and Uncertainty Scope (USCO).
Predicted Entities
NEG, UNC, USCO, NSCO
How to use
document_assembler = DocumentAssembler()\
.setInputCol("text")\
.setOutputCol("document")
sentenceDetectorDL = SentenceDetectorDLModel.pretrained("sentence_detector_dl", "xx")\
.setInputCols(["document"])\
.setOutputCol("sentence")
tokenizer = Tokenizer()\
.setInputCols(["sentence"])\
.setOutputCol("token")
word_embeddings = RoBertaEmbeddings.pretrained("roberta_base_biomedical", "es")\
.setInputCols(["sentence","token"])\
.setOutputCol("embeddings")
ner = MedicalNerModel.pretrained('ner_negation_uncertainty', "es", "clinical/models") \
.setInputCols(["sentence", "token", "embeddings"]) \
.setOutputCol("ner")
ner_converter = NerConverter()\
.setInputCols(["sentence", "token", "ner"])\
.setOutputCol("ner_chunk")
pipeline = Pipeline(stages=[
document_assembler,
sentenceDetectorDL,
tokenizer,
word_embeddings,
ner,
ner_converter])
data = spark.createDataFrame([["""Con diagnóstico probable de cirrosis hepática (no conocida previamente) y peritonitis espontanea primaria con tratamiento durante 8 dias con ceftriaxona en el primer ingreso (no se realizó paracentesis control por escasez de liquido). Lesión tumoral en hélix izquierdo de 0,5 cms. de diámetro susceptible de ca basocelular perlado."""]]).toDF("text")
result = pipeline.fit(data).transform(data)
val documenter = new DocumentAssembler()
.setInputCol("text")
.setOutputCol("document")
val sentenceDetector = SentenceDetectorDLModel.pretrained()
.setInputCols("document")
.setOutputCol("sentence")
val tokenizer = new Tokenizer()
.setInputCols("sentence")
.setOutputCol("token")
val word_embeddings = RoBertaEmbeddings.pretrained("roberta_base_biomedical", "es")
.setInputCols(Array("sentence","token"))
.setOutputCol("embeddings")
val ner_model = MedicalNerModel.pretrained("ner_negation_uncertainty", "es", "clinical/models")
.setInputCols(Array("sentence", "token", "embeddings"))
.setOutputCol("ner")
val ner_converter = new NerConverter()
.setInputCols(Array("sentence", "token", "ner"))
.setOutputCol("ner_chunk")
val pipeline = new Pipeline().setStages(Array(documenter, sentenceDetector, tokenizer, word_embeddings, ner_model, ner_converter))
val data = Seq(Array("""Con diagnóstico probable de cirrosis hepática (no conocida previamente) y peritonitis espontanea primaria con tratamiento durante 8 dias con ceftriaxona en el primer ingreso (no se realizó paracentesis control por escasez de liquido). Lesión tumoral en hélix izquierdo de 0,5 cms. de diámetro susceptible de ca basocelular perlado.""")).toDS().toDF("text")
val result = pipeline.fit(data).transform(data)
import nlu
nlu.load("es.med_ner.negtaion_uncertainty").predict("""Con diagnóstico probable de cirrosis hepática (no conocida previamente) y peritonitis espontanea primaria con tratamiento durante 8 dias con ceftriaxona en el primer ingreso (no se realizó paracentesis control por escasez de liquido). Lesión tumoral en hélix izquierdo de 0,5 cms. de diámetro susceptible de ca basocelular perlado.""")
Results
+------------------------------------------------------+---------+
|chunk |ner_label|
+------------------------------------------------------+---------+
|probable de |UNC |
|cirrosis hepática |USCO |
|no |NEG |
|conocida previamente |NSCO |
|no |NEG |
|se realizó paracentesis control por escasez de liquido|NSCO |
|susceptible de |UNC |
|ca basocelular perlado |USCO |
+------------------------------------------------------+---------+
Model Information
| Model Name: | ner_negation_uncertainty |
| Compatibility: | Healthcare NLP 4.0.2+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [sentence, token, embeddings] |
| Output Labels: | [ner] |
| Language: | es |
| Size: | 16.2 MB |
References
The model is prepared using the reference paper: “NLP applied to occupational health: MEDDOPROF shared task at IberLEF 2021 on automatic recognition, classification and normalization of professions and occupations from medical texts”, Salvador Lima-Lopez, Eulalia Farr ́e-Maduell, Antonio Miranda-Escalada, Vicent Briva-Iglesias and Martin Krallinger. Procesamiento del Lenguaje Natural, Revista nº 67, septiembre de 2021, pp. 243-256.
Benchmarking
label precision recall f1-score support
B-NEG 0.93 0.97 0.95 1409
I-NEG 0.80 0.90 0.85 119
B-UNC 0.82 0.85 0.83 395
I-UNC 0.77 0.78 0.77 166
B-USCO 0.76 0.79 0.77 394
I-USCO 0.61 0.81 0.69 1468
B-NSCO 0.92 0.92 0.92 1308
I-NSCO 0.87 0.89 0.88 3806
micro-avg 0.82 0.89 0.85 9065
macro-avg 0.81 0.86 0.83 9065
weighted-avg 0.83 0.89 0.86 9065