Description
This vision-language model is trained to understand medical images and extract key details such as patient demographics, clinical conditions, and prescribed medications, etc.
How to use
from sparknlp.base import DocumentAssembler, ImageAssembler
from sparknlp_jsl.utils import vision_llm_preprocessor
from sparknlp_jsl.annotator import MedicalVisionLLM
from pyspark.ml import Pipeline
!mkdir -p images
!wget -O images/prescription_01.png -q "https://raw.githubusercontent.com/JohnSnowLabs/spark-nlp-workshop/master/healthcare-nlp/data/ocr/prescription_01.png"
!wget -O images/prescription_02.png -q "https://raw.githubusercontent.com/JohnSnowLabs/spark-nlp-workshop/master/healthcare-nlp/data/ocr/prescription_02.png"
prompt = """
# Template:
{
"Patient Name": "string",
"Patient Age": "integer",
"Patient Gender": "string",
"Hospital Number": "string",
"Episode Number": "string",
"Episode Date": "date-time"
}
# Context:
<image>
"""
input_df = vision_llm_preprocessor(
spark=spark,
images_path="images",
prompt=prompt,
output_col_name="prompt"
)
document_assembler = (
DocumentAssembler()
.setInputCol("prompt")
.setOutputCol("caption_document")
)
image_assembler = (
ImageAssembler()
.setInputCol("image")
.setOutputCol("image_assembler")
)
medicalVisionLLM = (
MedicalVisionLLM.pretrained("jsl_meds_ner_vlm_2b_q8_v1", "en", "clinical/models")
.setInputCols(["caption_document", "image_assembler"])
.setOutputCol("completions")
.setChatTemplate("")
.setBatchSize(4)
.setNGpuLayers(99)
.setNCtx(4096)
.setNPredict(512)
.setTemperature(0.0)
.setTopK(1)
.setTopP(1)
.setRepeatLastN(256)
.setRepeatPenalty(1.1)
.setStopStrings(["</s>", "User:", "<|im_start|>"])
)
pipeline = Pipeline().setStages([
document_assembler,
image_assembler,
medicalVisionLLM
])
model = pipeline.fit(input_df)
result = model.transform(input_df)
result.selectExpr(
"reverse(split(image.origin, '/')) as image_name", "completions.result"
).show(truncate=False)
from johnsnowlabs import nlp, medical
!mkdir -p images
!wget -O images/prescription_01.png -q "https://raw.githubusercontent.com/JohnSnowLabs/spark-nlp-workshop/master/healthcare-nlp/data/ocr/prescription_01.png"
!wget -O images/prescription_02.png -q "https://raw.githubusercontent.com/JohnSnowLabs/spark-nlp-workshop/master/healthcare-nlp/data/ocr/prescription_02.png"
prompt = """
# Template:
{
"Patient Name": "string",
"Patient Age": "integer",
"Patient Gender": "string",
"Hospital Number": "string",
"Episode Number": "string",
"Episode Date": "date-time"
}
# Context:
<image>
"""
input_df = nlp.vision_llm_preprocessor(
spark=spark,
images_path="images",
prompt=prompt,
output_col_name="prompt"
)
document_assembler = (
nlp.DocumentAssembler()
.setInputCol("prompt")
.setOutputCol("caption_document")
)
image_assembler = (
nlp.ImageAssembler()
.setInputCol("image")
.setOutputCol("image_assembler")
)
medicalVisionLLM = (
medical.MedicalVisionLLM.pretrained("jsl_meds_ner_vlm_2b_q8_v1", "en", "clinical/models")
.setInputCols(["caption_document", "image_assembler"])
.setOutputCol("completions")
.setChatTemplate("")
.setBatchSize(4)
.setNGpuLayers(99)
.setNCtx(4096)
.setNPredict(512)
.setTemperature(0.0)
.setTopK(1)
.setTopP(1)
.setRepeatLastN(256)
.setRepeatPenalty(1.1)
.setStopStrings(["</s>", "User:", "<|im_start|>"])
)
pipeline = nlp.Pipeline().setStages([
document_assembler,
image_assembler,
medicalVisionLLM
])
model = pipeline.fit(input_df)
result = model.transform(input_df)
result.selectExpr(
"reverse(split(image.origin, '/')) as image_name", "completions.result"
).show(truncate=False)
import com.johnsnowlabs.nlp.base._
import com.johnsnowlabs.nlp.annotators._
import org.apache.spark.sql.functions._
import org.apache.spark.ml.Pipeline
import sys.process._
"mkdir -p images".!
"wget -O images/prescription_01.png https://raw.githubusercontent.com/JohnSnowLabs/spark-nlp-workshop/master/healthcare-nlp/data/ocr/prescription_01.png".!
"wget -O images/prescription_02.png https://raw.githubusercontent.com/JohnSnowLabs/spark-nlp-workshop/master/healthcare-nlp/data/ocr/prescription_02.png".!
val prompt = """
prompt = """
# Template:
{
"Patient Name": "string",
"Patient Age": "integer",
"Patient Gender": "string",
"Hospital Number": "string",
"Episode Number": "string",
"Episode Date": "date-time"
}
# Context:
<image>
"""
"""
val data = ImageAssembler.loadImagesAsBytes(spark, "/content/images")
.withColumn("caption", lit(prompt))
val documentAssembler = new DocumentAssembler()
.setInputCol("caption")
.setOutputCol("caption_document")
val imageAssembler = new ImageAssembler()
.setInputCol("image")
.setOutputCol("image_assembler")
val medicalVisionLLM = MedicalVisionLLM
.pretrained("jsl_meds_ner_vlm_2b_q8_v1", "en", "clinical/models")
.setInputCols(Array("caption_document", "image_assembler"))
.setOutputCol("completions")
.setChatTemplate("")
.setBatchSize(4)
.setNGpuLayers(99)
.setNCtx(4096)
.setNPredict(512)
.setTemperature(0.0f)
.setTopK(1)
.setTopP(1f)
.setRepeatLastN(256)
.setRepeatPenalty(1.1f)
.setStopStrings(["</s>", "User:", "<|im_start|>"])
val pipeline = new Pipeline().setStages(Array(
documentAssembler,
imageAssembler,
medicalVisionLLM
))
val model = pipeline.fit(data)
val result = model.transform(data)
result.selectExpr(
"reverse(split(image.origin, '/'))[0] as image_name",
"completions.result"
).show(false)
Results
{
"Patient Name": "Ms RUKHSANA SHAHEEN",
"Patient Age": 56,
"Patient Gender": "Female",
"Hospital Number": "MH005990453",
"Episode Number": "030000528270",
"Episode Date": "2021-07-02T08:31:00"
}
Model Information
| Model Name: | jsl_meds_ner_vlm_2b_q8_v1 |
| Compatibility: | Healthcare NLP 6.1.0+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [caption_document, image_assembler] |
| Output Labels: | [completions] |
| Language: | en |
| Size: | 2.3 GB |