Detect PHI for Deidentification in Romanian (BERT)

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 (Romanian) is a Named Entity Recognition model that annotates text to find protected health information that may need to be de-identified. It is trained with bert_base_cased embeddings and can detect 17 entities.

This NER model is trained with a combination of custom datasets with several data augmentation mechanisms.

Predicted Entities

AGE, CITY, COUNTRY, DATE, DOCTOR, EMAIL, FAX, HOSPITAL, IDNUM, LOCATION-OTHER, MEDICALRECORD, ORGANIZATION, PATIENT, PHONE, PROFESSION, STREET, ZIP

Live Demo Open in Colab Download Copy S3 URI

How to use

documentAssembler = DocumentAssembler()\
	.setInputCol("text")\
	.setOutputCol("document")

sentenceDetector = SentenceDetector()\
	.setInputCols(["document"])\
	.setOutputCol("sentence")

tokenizer = Tokenizer()\
	.setInputCols(["sentence"])\
	.setOutputCol("token")

embeddings = BertEmbeddings.pretrained("bert_base_cased", "ro")\
	.setInputCols(["sentence","token"])\
	.setOutputCol("word_embeddings")

clinical_ner = MedicalNerModel.pretrained("ner_deid_subentity_bert", "ro", "clinical/models")\
	.setInputCols(["sentence","token","word_embeddings"])\
	.setOutputCol("ner")

ner_converter = NerConverter()\
	.setInputCols(["sentence", "token", "ner"])\
	.setOutputCol("ner_chunk")

nlpPipeline = Pipeline(stages=[
	documentAssembler, 
	sentenceDetector, 
	tokenizer, 
	embeddings, 
	clinical_ner, 
	ner_converter])

text = """
Spitalul Pentru Ochi de Deal, Drumul Oprea Nr. 972 Vaslui, 737405 România
Tel: +40(235)413773
Data setului de analize: 25 May 2022 15:36:00
Nume si Prenume : BUREAN MARIA, Varsta: 77
Medic : Agota Evelyn Tımar
C.N.P : 2450502264401"""

data = spark.createDataFrame([[text]]).toDF("text")

results = nlpPipeline.fit(data).transform(data)

val documentAssembler = new DocumentAssembler()
	.setInputCol("text")
	.setOutputCol("document")

val sentenceDetector = new SentenceDetector()
	.setInputCols(Array("document"))
	.setOutputCol("sentence")

val tokenizer = new Tokenizer()
	.setInputCols(Array("sentence"))
	.setOutputCol("token")

val embeddings = BertEmbeddings.pretrained("bert_base_cased", "ro")
	.setInputCols(Array("sentence","token"))
	.setOutputCol("word_embeddings")

val clinical_ner = MedicalNerModel.pretrained("ner_deid_subentity_bert", "ro", "clinical/models")
	.setInputCols(Array("sentence","token","word_embeddings"))
	.setOutputCol("ner")

val ner_converter = new NerConverter()
	.setInputCols(Array("sentence", "token", "ner"))
	.setOutputCol("ner_chunk")
	
val pipeline = new Pipeline().setStages(Array(
	documentAssembler, 
	sentenceDetector, 
	tokenizer, 
	embeddings, 
	clinical_ner, 
	ner_converter))

val text = """Spitalul Pentru Ochi de Deal, Drumul Oprea Nr. 972 Vaslui, 737405 România
Tel: +40(235)413773
Data setului de analize: 25 May 2022 15:36:00
Nume si Prenume : BUREAN MARIA, Varsta: 77
Medic : Agota Evelyn Tımar
C.N.P : 2450502264401"""

val data = Seq(text).toDS.toDF("text")

val results = pipeline.fit(data).transform(data)

import nlu
nlu.load("ro.med_ner.deid.subentity.bert").predict("""
Spitalul Pentru Ochi de Deal, Drumul Oprea Nr. 972 Vaslui, 737405 România
Tel: +40(235)413773
Data setului de analize: 25 May 2022 15:36:00
Nume si Prenume : BUREAN MARIA, Varsta: 77
Medic : Agota Evelyn Tımar
C.N.P : 2450502264401""")

Results

+----------------------------+---------+
|chunk                       |ner_label|
+----------------------------+---------+
|Spitalul Pentru Ochi de Deal|HOSPITAL |
|Drumul Oprea Nr             |STREET   |
|Vaslui                      |CITY     |
|737405                      |ZIP      |
|+40(235)413773              |PHONE    |
|25 May 2022                 |DATE     |
|BUREAN MARIA                |PATIENT  |
|77                          |AGE      |
|Agota Evelyn Tımar          |DOCTOR   |
|2450502264401               |IDNUM    |
+----------------------------+---------+

Model Information

Model Name:	ner_deid_subentity_bert
Compatibility:	Healthcare NLP 4.0.0+
License:	Licensed
Edition:	Official
Input Labels:	[sentence, token, embeddings]
Output Labels:	[ner]
Language:	ro
Size:	16.5 MB

References

Custom John Snow Labs datasets
Data augmentation techniques

Benchmarking

label  		precision    recall  f1-score   support
AGE       		0.98      0.95      0.96      1186
CITY       		0.94      0.87      0.90       299
COUNTRY       	0.90      0.73      0.81       108
DATE       		0.98      0.95      0.96      4518
DOCTOR       	0.91      0.94      0.93      1979
EMAIL       	1.00      0.62      0.77         8
FAX       		0.98      0.95      0.96        56
HOSPITAL       	0.92      0.85      0.88       881
IDNUM       	0.98      0.99      0.98       235
LOCATION-OTHER	1.00      0.85      0.92        13
MEDICALRECORD	0.99      1.00      1.00       444
ORGANIZATION	0.86      0.76      0.81        75
PATIENT       	0.91      0.87      0.89       937
PHONE       	0.96      0.98      0.97       302
PROFESSION      0.85      0.82      0.83       161
STREET       	0.96      0.94      0.95       173
ZIP       		0.99      0.98      0.99       138
micro-avg       0.95      0.93      0.94     11513
macro-avg       0.95      0.89      0.91     11513
weighted-avg    0.95      0.93      0.94     11513

PREVIOUSDetect PHI for Deidentification in Romanian (BERT)

NEXTDetect PHI for Deidentification in Romanian (Word2Vec)