Feature Request - Presidio Medical Recognizer

[RKapadia01]

Is your feature request related to a problem? Please describe.
I’m working with Presidio in a context where users may input medical and dietary information into a chatbot.

Currently, Presidio does not have built-in support for detecting medical entities such as diseases, medications, and clinical procedures. This limitation required me to implement a custom recognizer to address the need for medical PII detection.

Below is an example of the custom recognizer I’ve built:

from presidio_analyzer import EntityRecognizer, RecognizerResult
from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

class ClinicalBERTRecognizer(EntityRecognizer):
    def __init__(self):
        # Download the model from Hugging Face's model hub
        model_name = "blaze999/Medical-NER"

        # Load the model and tokenizer
        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
        self.model = AutoModelForTokenClassification.from_pretrained(model_name)

        # Create a pipeline for named entity recognition
        self.ner_pipeline = pipeline("ner", model=self.model, tokenizer=self.tokenizer)

        # Define the supported entities
        self.supported_entities = [
            "BIOLOGICAL_ATTRIBUTE",
            "BIOLOGICAL_STRUCTURE",
            "CLINICAL_EVENT",
            "DISEASE_DISORDER",
            "FAMILY_HISTORY",
            "HISTORY",
            "MEDICATION",
            "THERAPEUTIC_PROCEDURE"
        ]

        super().__init__(supported_entities=self.supported_entities)


    def analyze(self, text, entities, nlp_artifacts=None):
        results = []

        # Perform named entity recognition on the input text
        ner_results = self.ner_pipeline(text)

        for entity in ner_results:
            entity_type = entity["entity"].replace("B-", "").replace("I-", "")

            # Check if the entity type is in  the list of supported entities
            if entity_type in self.supported_entities:
                recognizer_result = RecognizerResult(
                    entity_type=entity_type,
                    start=entity["start"],
                    end=entity["end"],
                    score=entity["score"]
                )

                # Create a RecognizerResult object for the entity
                results.append(recognizer_result)

        return results

Describe the solution you'd like
Would there be interest in incorporating a medical domain recognizer into Presidio? If so, I’d be happy to submit a PR with this implementation or a more generalized version.

The recognizer leverages transformer-based models, from Hugging Face, in order to identify clinical entities like diseases, medications, and procedures. This would allow Presidio to support medical and healthcare-related use cases out of the box.

Describe alternatives you've considered

Additional context

[omri374]

Hi, I think that would be a great addition!
It offers a great addition for those interested in PHI and not just PII. Having said that, we should also think about the computational performance and development complexity. Therefore, our suggestion would be to add this to the repo, but not run by default. In addition, we don't install transformers by default to reduce development complexity, so the code should first check if transformers is installed, and skip if it isn't, so that it doesn't break the rest of the package.
For example:

try:
    import transformers
except ImportError:
    transformers = None


is_available = bool(transformers)

We're doing something similar here:

presidio/presidio-analyzer/presidio_analyzer/nlp_engine/transformers_nlp_engine.py

Line 7 in a21a17c

try:

If you're interested in creating a PR, we would be happy to help with anything needed. Thanks!

[RKapadia01]

Thanks for the response. Agree with your inputs. Will raise a PR for some feedback