Description
Extract relations between clinical events in terms of time. If an event occurred before, after, or overlaps another event.
Predicted Entities
AFTER, BEFORE, OVERLAP
Live Demo Open in Colab Copy S3 URI
How to use
...
documenter = DocumentAssembler()\
.setInputCol("text")\
.setOutputCol("document")
sentencer = SentenceDetector()\
.setInputCols(["document"])\
.setOutputCol("sentences")
tokenizer = sparknlp.annotators.Tokenizer()\
.setInputCols(["sentences"])\
.setOutputCol("tokens")
pos_tagger = PerceptronModel()\
.pretrained("pos_clinical", "en", "clinical/models") \
.setInputCols(["sentences", "tokens"])\
.setOutputCol("pos_tags")
words_embedder = WordEmbeddingsModel() \
.pretrained("embeddings_clinical", "en", "clinical/models") \
.setInputCols(["sentences", "tokens"]) \
.setOutputCol("embeddings")
ner_tagger = MedicalNerModel.pretrained("ner_events_clinical", "en", "clinical/models")\
.setInputCols("sentences", "tokens", "embeddings")\
.setOutputCol("ner_tags")
ner_converter = NerConverter() \
.setInputCols(["sentences", "tokens", "ner_tags"]) \
.setOutputCol("ner_chunks")
dependency_parser = DependencyParserModel() \
.pretrained("dependency_conllu", "en") \
.setInputCols(["sentences", "pos_tags", "tokens"]) \
.setOutputCol("dependencies")
#Set a filter on pairs of named entities which will be treated as relation candidates
re_ner_chunk_filter = RENerChunksFilter() \
.setInputCols(["ner_chunks", "dependencies"])\
.setMaxSyntacticDistance(10)\
.setOutputCol("re_ner_chunks")
#.setRelationPairs(['SYMPTOM-EXTERNAL_BODY_PART_OR_REGION'])
re_model = RelationExtractionDLModel()\
.pretrained("redl_temporal_events_biobert", "en", "clinical/models") \
.setPredictionThreshold(0.5)\
.setInputCols(["re_ner_chunks", "sentences"]) \
.setOutputCol("relations")
pipeline = Pipeline(stages=[documenter, sentencer, tokenizer, pos_tagger, words_embedder, ner_tagger, ner_converter, dependency_parser, re_ner_chunk_filter, re_model])
text = "She is diagnosed with cancer in 1991. Then she was admitted to Mayo Clinic in May 2000 and discharged in October 2001"
data = spark.createDataFrame([[text]]).toDF("text")
p_model = pipeline.fit(data)
result = p_model.transform(data)
...
val documenter = DocumentAssembler()
.setInputCol("text")
.setOutputCol("document")
val sentencer = SentenceDetector()
.setInputCols("document")
.setOutputCol("sentences")
val tokenizer = sparknlp.annotators.Tokenizer()
.setInputCols("sentences")
.setOutputCol("tokens")
val pos_tagger = PerceptronModel()
.pretrained("pos_clinical", "en", "clinical/models")
.setInputCols(Array("sentences", "tokens"))
.setOutputCol("pos_tags")
val words_embedder = WordEmbeddingsModel()
.pretrained("embeddings_clinical", "en", "clinical/models")
.setInputCols(Array("sentences", "tokens"))
.setOutputCol("embeddings")
val ner_tagger = MedicalNerModel.pretrained("ner_events_clinical", "en", "clinical/models")
.setInputCols(Array("sentences", "tokens", "embeddings"))
.setOutputCol("ner_tags")
val ner_converter = NerConverter()
.setInputCols(Array("sentences", "tokens", "ner_tags"))
.setOutputCol("ner_chunks")
val dependency_parser = DependencyParserModel()
.pretrained("dependency_conllu", "en")
.setInputCols(Array("sentences", "pos_tags", "tokens"))
.setOutputCol("dependencies")
// Set a filter on pairs of named entities which will be treated as relation candidates
val re_ner_chunk_filter = RENerChunksFilter()
.setInputCols(Array("ner_chunks", "dependencies"))
.setMaxSyntacticDistance(10)
.setOutputCol("re_ner_chunks")
// .setRelationPairs(Array("SYMPTOM-EXTERNAL_BODY_PART_OR_REGION"))
// The dataset this model is trained to is sentence-wise.
// This model can also be trained on document-level relations - in which case, while predicting, use "document" instead of "sentence" as input.
val re_model = RelationExtractionDLModel()
.pretrained("redl_temporal_events_biobert", "en", "clinical/models")
.setPredictionThreshold(0.5)
.setInputCols(Array("re_ner_chunks", "sentences"))
.setOutputCol("relations")
val pipeline = new Pipeline().setStages(Array(documenter, sentencer, tokenizer, pos_tagger, words_embedder, ner_tagger, ner_converter, dependency_parser, re_ner_chunk_filter, re_model))
val data = Seq("She is diagnosed with cancer in 1991. Then she was admitted to Mayo Clinic in May 2000 and discharged in October 2001").toDF("text")
val result = pipeline.fit(data).transform(data)
import nlu
nlu.load("en.relation.temporal_events").predict("""She is diagnosed with cancer in 1991. Then she was admitted to Mayo Clinic in May 2000 and discharged in October 2001""")
Results
+--------+-------------+-------------+-----------+-----------+-------------+-------------+-----------+------------+----------+
|relation| entity1|entity1_begin|entity1_end| chunk1| entity2|entity2_begin|entity2_end| chunk2|confidence|
+--------+-------------+-------------+-----------+-----------+-------------+-------------+-----------+------------+----------+
| BEFORE| OCCURRENCE| 7| 15| diagnosed| PROBLEM| 22| 27| cancer|0.78168863|
| OVERLAP| PROBLEM| 22| 27| cancer| DATE| 32| 35| 1991| 0.8492274|
| AFTER| OCCURRENCE| 51| 58| admitted|CLINICAL_DEPT| 63| 73| Mayo Clinic|0.85629463|
| BEFORE| OCCURRENCE| 51| 58| admitted| OCCURRENCE| 91| 100| discharged| 0.6843513|
| OVERLAP|CLINICAL_DEPT| 63| 73|Mayo Clinic| DATE| 78| 85| May 2000| 0.7844673|
| BEFORE|CLINICAL_DEPT| 63| 73|Mayo Clinic| OCCURRENCE| 91| 100| discharged|0.60411876|
| OVERLAP|CLINICAL_DEPT| 63| 73|Mayo Clinic| DATE| 105| 116|October 2001| 0.540761|
| BEFORE| DATE| 78| 85| May 2000| OCCURRENCE| 91| 100| discharged| 0.6042761|
| OVERLAP| DATE| 78| 85| May 2000| DATE| 105| 116|October 2001|0.64867175|
| BEFORE| OCCURRENCE| 91| 100| discharged| DATE| 105| 116|October 2001| 0.5302478|
+--------+-------------+-------------+-----------+-----------+-------------+-------------+-----------+------------+----------+
Model Information
| Model Name: | redl_temporal_events_biobert |
| Compatibility: | Healthcare NLP 2.7.3+ |
| License: | Licensed |
| Edition: | Official |
| Language: | en |
Data Source
Trained on temporal clinical events benchmark dataset.
Benchmarking
Relation Recall Precision F1 Support
AFTER 0.332 0.655 0.440 2123
BEFORE 0.868 0.908 0.887 13817
OVERLAP 0.887 0.733 0.802 7860
Avg. 0.695 0.765 0.710