parsigs 0.0.4b0

The private, smart and easy Sig (Dosage Instructions) text-parser

Parsigs - the private, smart and easy Sig (Dosage Instructions) text-parser

Parsigs is an open-source project that aims to extract relevant information from doctors' signature text without compromising privacy and PHI (Protected Health Information) using Natural Language Processing.

Features

• Extracts relevant structured information such as dosage, frequency, drug name from doctors' prescription-sig text.
• Protects patient privacy by not extracting any patient-related information or using any external API.
• Uses NLP techniques to accurately extract information from unstructured text.

Installation

For optimal usage, you will need to install the parsigs package and the underlying model (you can provide a different one by yourself, but this is not recommended as the model was trained for the specific task of parsing dosage instructions from Sigs)

pip install parsigs
pip install https://huggingface.co/royashcenazi/en_parsigs/resolve/main/en_parsigs-any-py3-none-any.whl

Usage

Here are some examples of how to use the parse_sig method:

from parsigs.parse_sig_api import StructuredSig, SigParser

sig_parser = SigParser()

sig = "Take 1 tablet of ibuprofen 200mg 3 times every day for 3 weeks"
parsed_sig = sig_parser.parse(sig)

expected = StructuredSig(drug="ibuprofen", form="tablet", strength="200mg", frequencyType="Day", interval=3, singleDosageAmount=1.0, periodType='Week', periodAmount=3)

sig2 = "Take 2 tablets 3 times every month"
parsed_sig = sig_parser.parse(sig)

expected = StructuredSig(drug=None, form='tablets', strength=None, frequencyType='Month', interval=3, singleDosageAmount=2.0, periodType=None, periodAmount=None, takeAsNeeded = False)

The StructuredSig object has the following attributes:

• drug: the name of the drug
• form: the form of the medication (e.g. tablet, solution, pill)
• strength: the strength of the medication (e.g. 200mg, 500mg)
• frequencyType: the time-unit of the frequency (e.g. Day, Week, Month)
• interval: the number of times per frequency time-unit
• singleDosageAmount: the amount of the medication to take at each interval
• periodType: the unit-type of the period which indicates for how long medication should be taken (e.g. Day, Week, Month)
• periodAmount: the number of units per periodType
• takeAsNeeded: a flag indicating if the instructed dosage should be taken only when the patient needs it

Known Issues

The parse-sig project is developed by using the Named Entity Recognition model for tagging different parts in a dosage instruction (Sig) sentence (Duration, Frequency, Dosage, Drug, Form, Strength). The different tags are than being processed by static rules. You should expect some amount of errors as the model was trained on a large but limited data, as aquiring this type of data for it being private is a task of itself. I do intend to add a Dev set in the future and also more examples other than the ones in the test module.

The main target of the project is to identify and structure dosage instructions, as most prescriptions contain the brand name as a part of the definition, it is less important to structure it than the frequency, dosage and period so expect that part not be as complete as dosage instructions identification (this data is often not provided in the Sig, e.g "Take 1 tablet every day")

If you encounter any more issues or have any questions, please don't hesitate to send me an email or file an issue.

Credits

This repository is based on the following resources -

https://odsc.medium.com/training-a-medication-named-entity-recognition-model-from-scratch-with-spacy-e94fdff56022

https://github.com/bpben/spacy_ner_tutorial