Description
This model can be used to detect symptoms, treatments and other entities in medical text in German language.
Predicted Entities
DIAGLAB_PROCEDURE, MEDICAL_SPECIFICATION, MEDICAL_DEVICE, MEASUREMENT, BIOLOGICAL_CHEMISTRY, BODY_FLUID, TIME_INFORMATION, LOCAL_SPECIFICATION, BIOLOGICAL_PARAMETER, PROCESS, MEDICATION, DOSING, DEGREE, MEDICAL_CONDITION, PERSON, TISSUE, STATE_OF_HEALTH, BODY_PART, TREATMENT
Live Demo Open in Colab Copy S3 URI
How to use
Use as part of an nlp pipeline with the following stages: DocumentAssembler, SentenceDetector, Tokenizer, WordEmbeddingsModel, NerDLModel. Add the NerConverter to the end of the pipeline to convert entity tokens into full entity chunks.
...
word_embeddings = WordEmbeddingsModel.pretrained("w2v_cc_300d","de","clinical/models")\
.setInputCols(["document","token"])\
.setOutputCol("embeddings")
clinical_ner = NerDLModel.pretrained("ner_healthcare", "de", "clinical/models") \
.setInputCols(["sentence", "token", "embeddings"]) \
.setOutputCol("ner")
...
nlp_pipeline = Pipeline(stages=[document_assembler, sentence_detector, tokenizer, word_embeddings, clinical_ner, clinical_ner_converter])
light_pipeline = LightPipeline(nlp_pipeline.fit(spark.createDataFrame([['']]).toDF("text")))
annotations = light_pipeline.fullAnnotate("Das Kleinzellige Bronchialkarzinom (Kleinzelliger Lungenkrebs, SCLC) ist ein hochmalignes bronchogenes Karzinom")
...
val word_embeddings = WordEmbeddingsModel.pretrained("w2v_cc_300d","de","clinical/models")
.setInputCols(Array("document","token"))
.setOutputCol("embeddings")
val ner = NerDLModel.pretrained("ner_healthcare", "de", "clinical/models")
.setInputCols("sentence", "token", "embeddings")
.setOutputCol("ner")
...
val pipeline = new Pipeline().setStages(Array(document_assembler, sentence_detector, tokenizer, word_embeddings, ner, clinical_ner_converter))
val data = Seq("Das Kleinzellige Bronchialkarzinom (Kleinzelliger Lungenkrebs, SCLC) ist ein hochmalignes bronchogenes Karzinom").toDF("text")
val result = pipeline.fit(data).transform(data)
import nlu
nlu.load("de.med_ner.healthcare").predict("""Das Kleinzellige Bronchialkarzinom (Kleinzelliger Lungenkrebs, SCLC) ist ein hochmalignes bronchogenes Karzinom""")
Results
+----+-------------------+---------+---------+--------------------------+
| | chunk | begin | end | entity |
+====+===================+=========+=========+==========================+
| 0 | Kleinzellige | 4 | 15 | MEDICAL_SPECIFICATION |
+----+-------------------+---------+---------+--------------------------+
| 1 | Bronchialkarzinom | 17 | 33 | MEDICAL_CONDITION |
+----+-------------------+---------+---------+--------------------------+
| 2 | Kleinzelliger | 36 | 48 | MEDICAL_SPECIFICATION |
+----+-------------------+---------+---------+--------------------------+
| 3 | Lungenkrebs | 50 | 60 | MEDICAL_CONDITION |
+----+-------------------+---------+---------+--------------------------+
| 4 | SCLC | 63 | 66 | MEDICAL_CONDITION |
+----+-------------------+---------+---------+--------------------------+
| 5 | hochmalignes | 77 | 88 | MEASUREMENT |
+----+-------------------+---------+---------+--------------------------+
| 6 | bronchogenes | 90 | 101 | BODY_PART |
+----+-------------------+---------+---------+--------------------------+
| 7 | Karzinom | 103 | 110 | MEDICAL_CONDITION |
+----+-------------------+---------+---------+--------------------------+
Model Information
| Model Name: | ner_healthcare |
| Type: | ner |
| Compatibility: | Healthcare NLP 2.6.0 + |
| Edition: | Official |
| License: | Licensed |
| Input Labels: | [sentence, token, embeddings] |
| Output Labels: | [ner] |
| Language: | [de] |
| Case sensitive: | false |
| Dependencies: | w2v_cc_300d |
Data Source
Trained with augmented version of 2010 i2b2/VA dataset on concepts, assertions, and relations in clinical text with w2v_cc_300d.
Benchmarking
| | label | tp | fp | fn | precision| recall| f1 |
|---:|--------------------:|-------:|------:|-----:|---------:|---------:|---------:|
| 0 | BIOLOGICAL_PARAMETER| 103 | 52 | 57 | 0.6645 | 0.6438 | 0.654 |
| 1 | BODY_FLUID | 166 | 16 | 24 | 0.9121 | 0.8737 | 0.8925 |
| 2 | PERSON | 475 | 74 | 142 | 0.8652 | 0.7699 | 0.8148 |
| 3 | DOSING | 38 | 14 | 31 | 0.7308 | 0.5507 | 0.6281 |
| 4 | DIAGLAB_PROCEDURE | 236 | 58 | 68 | 0.8027 | 0.7763 | 0.7893 |
| 5 | BODY_PART | 690 | 72 | 79 | 0.9055 | 0.8973 | 0.9014 |
| 6 | MEDICATION | 391 | 117 | 167 | 0.7697 | 0.7007 | 0.7336 |
| 7 | STATE_OF_HEALTH | 321 | 41 | 76 | 0.8867 | 0.8086 | 0.8458 |
| 8 | LOCAL_SPECIFICATION | 57 | 19 | 24 | 0.75 | 0.7037 | 0.7261 |
| 9 | MEASUREMENT | 574 | 260 | 222 | 0.6882 | 0.7211 | 0.7043 |
| 10 | TREATMENT | 476 | 131 | 135 | 0.7842 | 0.7791 | 0.7816 |
| 11 | MEDICAL_CONDITION | 1741 | 442 | 271 | 0.7975 | 0.8653 | 0.83 |
| 12 | TIME_INFORMATION | 651 | 126 | 161 | 0.8378 | 0.8017 | 0.8194 |
| 13 | BIOLOGICAL_CHEMISTRY| 192 | 55 | 60 | 0.7773 | 0.7619 | 0.7695 |