Description
Assign assertion status to clinical entities extracted by NER based on their context in the text. Also, this model is trained on a list of clinical and biomedical datasets curated in-house
Predicted Entities
Absent, Past, Present, Family, Hypothetical, Possible
How to use
document_assembler = DocumentAssembler()\
.setInputCol("text")\
.setOutputCol("document")
sentence_detector = SentenceDetectorDLModel.pretrained("sentence_detector_dl_healthcare", "en", "clinical/models")\
.setInputCols(["document"])\
.setOutputCol("sentence")
tokenizer = Tokenizer()\
.setInputCols(["sentence"])\
.setOutputCol("token")\
.setSplitChars(["-", "\/"])
word_embeddings = WordEmbeddingsModel.pretrained("embeddings_clinical", "en", "clinical/models")\
.setInputCols(["sentence", "token"])\
.setOutputCol("embeddings")
# ner_oncology
ner_oncology = MedicalNerModel.pretrained("ner_oncology", "en", "clinical/models")\
.setInputCols(["sentence", "token", "embeddings"])\
.setOutputCol("ner_oncology")
ner_oncology_converter = NerConverterInternal()\
.setInputCols(["sentence","token","ner_oncology"])\
.setOutputCol("ner_chunk")
few_shot_assertion_converter = FewShotAssertionSentenceConverter()\
.setInputCols(["sentence", "token", "ner_chunk"])\
.setOutputCol("assertion_sentence")
e5_embeddings = E5Embeddings.pretrained("e5_base_v2_embeddings_medical_assertion_oncology", "en", "clinical/models")\
.setInputCols(["assertion_sentence"])\
.setOutputCol("assertion_embedding")
few_shot_assertion_classifier = FewShotAssertionClassifierModel()\
.pretrained("fewhot_assertion_oncology_e5_base_v2_oncology", "en", "clinical/models")\
.setInputCols(["assertion_embedding"])\
.setOutputCol("assertion")
assertion_pipeline = Pipeline(stages=[
document_assembler,
sentence_detector,
tokenizer,
word_embeddings,
ner_oncology,
ner_oncology_converter,
few_shot_assertion_converter,
e5_embeddings,
few_shot_assertion_classifier
])
sample_text= [
"""The patient is suspected to have colorectal cancer. Her family history is positive for other cancers. The result of the biopsy was positive. A CT scan was ordered to rule out metastases."""
]
data = spark.createDataFrame([sample_text]).toDF("text")
result = assertion_pipeline.fit(data).transform(data)
val document_assembler = new DocumentAssembler()
.setInputCol("text")
.setOutputCol("document")
val sentence_detector = SentenceDetectorDLModel.pretrained("sentence_detector_dl_healthcare", "en", "clinical/models")
.setInputCols(Array("document"))
.setOutputCol("sentence")
val tokenizer = new Tokenizer()
.setInputCols(Array("sentence"))
.setOutputCol("token")
val word_embeddings = WordEmbeddingsModel.pretrained("embeddings_clinical", "en", "clinical/models")
.setInputCols(Array("sentence", "token"))
.setOutputCol("embeddings")
# ner_oncology
val ner_oncology = MedicalNerModel.pretrained("ner_oncology", "en", "clinical/models")
.setInputCols(Array("sentence", "token", "embeddings"))
.setOutputCol("ner_oncology")
val ner_oncology_converter = new NerConverterInternal()
.setInputCols(Array("sentence","token","ner_oncology"))
.setOutputCol("ner_chunk")
val few_shot_assertion_converter = new FewShotAssertionSentenceConverter()
.setInputCols(Array("sentence", "token", "ner_chunk"))
.setOutputCol("assertion_sentence")
val e5_embeddings = E5Embeddings.pretrained("e5_base_v2_embeddings_medical_assertion_oncology", "en", "clinical/models")
.setInputCols(Array("assertion_sentence"))
.setOutputCol("assertion_embedding")
val few_shot_assertion_classifier = FewShotAssertionClassifierModel()
.pretrained("fewhot_assertion_oncology_e5_base_v2_oncology", "en", "clinical/models")
.setInputCols(Array("assertion_embedding"))
.setOutputCol("assertion")
val pipeline = new Pipeline().setStages(Array(
document_assembler,
sentence_detector,
tokenizer,
word_embeddings,
ner_oncology,
ner_oncology_converter,
few_shot_assertion_converter,
e5_embeddings,
few_shot_assertion_classifier))
val data = Seq("The patient is suspected to have colorectal cancer. Her family history is positive for other cancers. The result of the biopsy was positive. A CT scan was ordered to rule out metastases.").toDF("text")
val result = pipeline.fit(data).transform(data)
Results
| chunks | begin | end | entities | assertion | confidence |
|:------------------|--------:|------:|:-----------------|:------------|-------------:|
| colorectal cancer | 33 | 49 | Cancer_Dx | Possible | 0.581282 |
| Her | 52 | 54 | Gender | Present | 0.9563 |
| cancers | 93 | 99 | Cancer_Dx | Family | 0.234656 |
| biopsy | 120 | 125 | Pathology_Test | Past | 0.957321 |
| CT scan | 143 | 149 | Imaging_Test | Past | 0.95717 |
| metastases | 175 | 184 | Metastasis | Possible | 0.549866 |
Model Information
| Model Name: | fewhot_assertion_oncology_e5_base_v2_oncology |
| Compatibility: | Healthcare NLP 5.3.3+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [assertion_embedding] |
| Output Labels: | [assertion] |
| Language: | en |
| Size: | 25.4 KB |
Benchmarking
label precision recall f1-score support
Absent 0.86 0.86 0.86 343
Family 1.00 0.86 0.92 35
Hypothetical 0.72 0.71 0.71 216
Past 0.90 0.95 0.93 2024
Possible 0.76 0.79 0.77 80
Present 0.93 0.87 0.90 1842
accuracy - - 0.90 4540
macro-avg 0.86 0.84 0.85 4540
weighted-avg 0.90 0.90 0.90 4540