Description
Named Entity Recognition annotators allow for a generic model to be trained by using a Deep Learning architecture (Char CNNs - BiLSTM - CRF - word embeddings) inspired on a former state of the art model for NER: Chiu & Nicols, Named Entity Recognition with Bidirectional LSTM,CNN.
Deidentification NER (Spanish) is a Named Entity Recognition model that annotates text to find protected health information that may need to be de-identified. It detects 17 entities, which is more than the previously released ner_deid_subentity_roberta model.
This NER model is trained with a combination of custom datasets, Spanish 2002 conLL, MeddoProf and MeddoCan datasets, and includes several data augmentation mechanisms.
This is a version that includes Roberta Clinical embeddings. You can find as well ner_deid_subentity_augmented that uses Sciwi 300d embeddings based instead of Roberta.
Predicted Entities
PATIENT, HOSPITAL, DATE, ORGANIZATION, CITY, ID, STREET, USERNAME, SEX, EMAIL, ZIP, MEDICALRECORD, PROFESSION, PHONE, COUNTRY, DOCTOR, AGE
Live Demo Open in Colab Copy S3 URI
How to use
documentAssembler = nlp.DocumentAssembler()\
.setInputCol("text")\
.setOutputCol("document")
sentenceDetector = nlp.SentenceDetectorDLModel.pretrained("sentence_detector_dl","xx")\
.setInputCols(["document"])\
.setOutputCol("sentence")
tokenizer = nlp.Tokenizer()\
.setInputCols(["sentence"])\
.setOutputCol("token")
roberta_embeddings = nlp.RoBertaEmbeddings.pretrained("roberta_base_biomedical", "es")\
.setInputCols(["sentence", "token"])\
.setOutputCol("embeddings")
clinical_ner = medical.NerModel.pretrained("ner_deid_subentity_roberta_augmented", "es", "clinical/models")\
.setInputCols(["sentence","token","embeddings"])\
.setOutputCol("ner")
nlpPipeline = nlp.Pipeline(stages=[
documentAssembler,
sentenceDetector,
tokenizer,
roberta_embeddings,
clinical_ner])
text = ['''
Antonio Miguel Martínez, varón de de 35 años de edad, de profesión auxiliar de enfermería y nacido en Cadiz, España. Aún no estaba vacunado, se infectó con Covid-19 el dia 14 de Marzo y tuvo que ir al Hospital. Fue tratado con anticuerpos monoclonales en la Clinica San Carlos.
''']
df = spark.createDataFrame([text]).toDF("text")
results = nlpPipeline.fit(df).transform(df)
val documentAssembler = new DocumentAssembler()
.setInputCol("text")
.setOutputCol("document")
val sentenceDetector = SentenceDetectorDLModel.pretrained("sentence_detector_dl_healthcare","xx")
.setInputCols(Array("document"))
.setOutputCol("sentence")
val tokenizer = new Tokenizer()
.setInputCols(Array("sentence"))
.setOutputCol("token")
val roberta_embeddings = RoBertaEmbeddings.pretrained("roberta_base_biomedical", "es")
.setInputCols(Array("sentence", "token"))
.setOutputCol("embeddings")
val clinical_ner = MedicalNerModel.pretrained("ner_deid_subentity_roberta_augmented", "es", "clinical/models")
.setInputCols(Array("sentence","token","embeddings"))
.setOutputCol("ner")
val pipeline = new Pipeline().setStages(Array(documentAssembler, sentenceDetector, tokenizer, roberta_embeddings, clinical_ner))
val text = "Antonio Miguel Martínez, varón de de 35 años de edad, de profesión auxiliar de enfermería y nacido en Cadiz, España. Aún no estaba vacunado, se infectó con Covid-19 el dia 14 de Marzo y tuvo que ir al Hospital. Fue tratado con anticuerpos monoclonales en la Clinica San Carlos."
val df = Seq(text).toDF("text")
val results = pipeline.fit(df).transform(df)
import nlu
nlu.load("es.med_ner.deid.subentity.roberta").predict("""
Antonio Miguel Martínez, varón de de 35 años de edad, de profesión auxiliar de enfermería y nacido en Cadiz, España. Aún no estaba vacunado, se infectó con Covid-19 el dia 14 de Marzo y tuvo que ir al Hospital. Fue tratado con anticuerpos monoclonales en la Clinica San Carlos.
""")
Results
+------------+------------+
| token| ner_label|
+------------+------------+
| Antonio| B-PATIENT|
| Miguel| I-PATIENT|
| Martínez| I-PATIENT|
| ,| O|
| varón| B-SEX|
| de| O|
| de| O|
| 35| B-AGE|
| años| O|
| de| O|
| edad| O|
| ,| O|
| de| O|
| profesión| O|
| auxiliar|B-PROFESSION|
| de|I-PROFESSION|
| enfermería|I-PROFESSION|
| y| O|
| nacido| O|
| en| O|
| Cadiz| B-CITY|
| ,| O|
| España| B-COUNTRY|
| .| O|
| Aún| O|
| no| O|
| estaba| O|
| vacunado| O|
| ,| O|
| se| O|
| infectó| O|
| con| O|
| Covid-19| O|
| el| O|
| dia| O|
| 14| B-DATE|
| de| I-DATE|
| Marzo| I-DATE|
| y| O|
| tuvo| O|
| que| O|
| ir| O|
| al| O|
| Hospital| O|
| Fue| O|
| tratado| O|
| con| O|
| anticuerpos| O|
|monoclonales| O|
| en| O|
| la| O|
| Clinica| B-HOSPITAL|
| San| I-HOSPITAL|
| Carlos| I-HOSPITAL|
| .| O|
+------------+------------+
Model Information
| Model Name: | ner_deid_subentity_roberta_augmented |
| Compatibility: | Healthcare NLP 3.3.4+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [sentence, token, embeddings] |
| Output Labels: | [ner] |
| Language: | es |
| Size: | 16.3 MB |
References
- Internal JSL annotated corpus
- Spanish conLL
- MeddoProf
- MeddoCan
Benchmarking
label tp fp fn total precision recall f1
PATIENT 1874.0 165.0 186.0 2060.0 0.9191 0.9097 0.9144
HOSPITAL 241.0 19.0 54.0 295.0 0.9269 0.8169 0.8685
DATE 954.0 17.0 15.0 969.0 0.9825 0.9845 0.9835
ORGANIZATION 2521.0 483.0 468.0 2989.0 0.8392 0.8434 0.8413
CITY 1464.0 369.0 289.0 1753.0 0.7987 0.8351 0.8165
ID 35.0 1.0 0.0 35.0 0.9722 1.0 0.9859
STREET 194.0 8.0 6.0 200.0 0.9604 0.97 0.9652
USERNAME 7.0 0.0 4.0 11.0 1.0 0.6364 0.7778
SEX 618.0 9.0 9.0 627.0 0.9856 0.9856 0.9856
EMAIL 134.0 0.0 0.0 134.0 1.0 1.0 1.0
ZIP 138.0 0.0 1.0 139.0 1.0 0.9928 0.9964
MEDICALRECORD 29.0 10.0 0.0 29.0 0.7436 1.0 0.8529
PROFESSION 231.0 20.0 30.0 261.0 0.9203 0.8851 0.9023
PHONE 44.0 0.0 6.0 50.0 1.0 0.88 0.9362
COUNTRY 458.0 96.0 103.0 561.0 0.8267 0.8164 0.8215
DOCTOR 432.0 38.0 48.0 480.0 0.9191 0.9 0.9095
AGE 509.0 9.0 10.0 519.0 0.9826 0.9807 0.9817
macro - - - - - - 0.9141
micro - - - - - - 0.8891