Description
The Frailty classifier employs BioBERT within a robust classifier architecture. Trained on a diverse dataset, this model provides accurate label assignments and confidence scores for its predictions. The primary goal of this model is to categorize text into two key labels: ‘High_or_Low_Frailty’ and ‘No_Frailty_or_Unknown.’
High_or_Low_Frailty: This category encompasses statements that indicate the presence of significant concerns or symptoms related to frailty conditions, suggesting either high or low risks. This label reflects substantial worries about frailty and the need for relevant medical measures.
No_Frailty_or_Unknown: This category includes statements that do not contain clear concerns related to frailty or where the frailty condition is unknown. This label indicates the absence of frailty-related issues or insufficient information in the text.
Predicted Entities
High_or_Low_Frailty, No_Frailty_or_Unknown
How to use
document_assembler = DocumentAssembler()\
.setInputCol('text')\
.setOutputCol('document')
tokenizer = Tokenizer()\
.setInputCols(['document'])\
.setOutputCol('token')
sequenceClassifier = MedicalBertForSequenceClassification.pretrained('bert_sequence_classifier_sdoh_frailty', 'en', 'clinical/models')\
.setInputCols(["document",'token'])\
.setOutputCol("prediction")
pipeline = Pipeline(stages=[
document_assembler,
tokenizer,
sequenceClassifier
])
sample_texts=[ "Patient B is a 40-year-old female who was diagnosed with breast cancer. She has received a treatment plan that includes surgery, chemotherapy, and radiation therapy.",
"Post-chemotherapy, the patient was under regular surveillance for osteosarcoma. Recent imaging showed no signs of local recurrence or distant metastasis. Whereas the recovery was challenging, current evaluation confirms patient is in remission.",
"The patient was diagnosed with stage II colon cancer and will be undergoing a treatment regimen that includes both chemotherapy and radiation therapy.",
"Thyroid nodules detected during routine examination; fine-needle aspiration was conducted. Cytology results indicated no malignancy, consistent with a benign thyroid adenoma. However, patient is advised for a follow-up ultrasound in 12 months to monitor nodule size.",
"The patient's persistent lymphadenopathy led to further tests, which confirmed a diagnosis of AIDS.",
"Female patient presented with pelvic discomfort. Ovarian cysts were found during ultrasound; however, CA-125 levels are within normal range, and repeat imaging has shown consistent cyst size. No features of ovarian cancer were present, and a follow-up is scheduled in six months."
]
sample_data = spark.createDataFrame(sample_texts, StringType()).toDF("text")
result = pipeline.fit(sample_data).transform(sample_data)
result.select("text", "prediction.result").show(30,truncate=100)
val documenter = new DocumentAssembler()
.setInputCol("text")
.setOutputCol("document")
val tokenizer = new Tokenizer()
.setInputCols(Array("document"))
.setOutputCol("token")
val sequenceClassifier = MedicalBertForSequenceClassification.pretrained("bert_sequence_classifier_sdoh_frailty", "en", "clinical/models")
.setInputCols(Array("document","token"))
.setOutputCol("prediction")
val pipeline = new Pipeline().setStages(Array(documenter, tokenizer, sequenceClassifier))
val data = Seq(Array("Patient B is a 40-year-old female who was diagnosed with breast cancer. She has received a treatment plan that includes surgery, chemotherapy, and radiation therapy.",
"Post-chemotherapy, the patient was under regular surveillance for osteosarcoma. Recent imaging showed no signs of local recurrence or distant metastasis. Whereas the recovery was challenging, current evaluation confirms patient is in remission.",
"The patient was diagnosed with stage II colon cancer and will be undergoing a treatment regimen that includes both chemotherapy and radiation therapy.",
"Thyroid nodules detected during routine examination; fine-needle aspiration was conducted. Cytology results indicated no malignancy, consistent with a benign thyroid adenoma. However, patient is advised for a follow-up ultrasound in 12 months to monitor nodule size.",
"The patient's persistent lymphadenopathy led to further tests, which confirmed a diagnosis of AIDS.",
"Female patient presented with pelvic discomfort. Ovarian cysts were found during ultrasound; however, CA-125 levels are within normal range, and repeat imaging has shown consistent cyst size. No features of ovarian cancer were present, and a follow-up is scheduled in six months."
)).toDS.toDF("text")
val result = pipeline.fit(data).transform(data)
Results
+----------------------------------------------------------------------------------------------------+-----------------------+
| text| result|
+----------------------------------------------------------------------------------------------------+-----------------------+
|Patient B is a 40-year-old female who was diagnosed with breast cancer. She has received a treatm...| [High_or_Low_Frailty]|
|Post-chemotherapy, the patient was under regular surveillance for osteosarcoma. Recent imaging sh...|[No_Frailty_or_Unknown]|
|The patient was diagnosed with stage II colon cancer and will be undergoing a treatment regimen t...| [High_or_Low_Frailty]|
|Thyroid nodules detected during routine examination; fine-needle aspiration was conducted. Cytolo...|[No_Frailty_or_Unknown]|
| The patient's persistent lymphadenopathy led to further tests, which confirmed a diagnosis of AIDS.| [High_or_Low_Frailty]|
|Female patient presented with pelvic discomfort. Ovarian cysts were found during ultrasound; howe...|[No_Frailty_or_Unknown]|
+----------------------------------------------------------------------------------------------------+-----------------------+
Model Information
| Model Name: | bert_sequence_classifier_sdoh_frailty |
| Compatibility: | Healthcare NLP 5.1.2+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [document, token] |
| Output Labels: | [prediction] |
| Language: | en |
| Size: | 406.4 MB |
| Case sensitive: | true |
| Max sentence length: | 512 |
Benchmarking
label precision recall f1-score support
High_or_Low_Frailty 0.898067 0.891798 0.894921 573
No_Frailty_or_Unknown 0.908148 0.913562 0.910847 671
accuracy - - 0.903537 1244
macro-avg 0.903107 0.902680 0.902884 1244
weighted-avg 0.903505 0.903537 0.903511 1244