/medaCy_model_clinical_notes

A medaCy compatible model for clinical note information extraction.

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

spaCy

medaCy

🏥 Clinical Notes Model for medaCy 🏥

This repository contains a versioned, medaCy compatible Model for information extraction from clinical notes.

alt text

Description

This is the light-weight version (no metamap) of medaCy's model for extracting 9 unique entities from clinical notes:

Drug, Strength, Duration, Route, Form, ADE, Dosage, Reason, Frequency

Results

Model generalization ability is evaluated over 202 patient clinical note files not seen during training. Strict indicates exact matches of spans, Lenient indicates a fuzzy matching of spans (model predictions are off by single characters).

Entities Strict Lenient
Drug
Strength
Duration
Route
Form
ADE
Dosage
Reason
Frequency
System

Training Data

N2C2 2018 Shared Task The data used to induce this model is protected by HIPAA privacy regulations and thus cannot be published.

Authors

Andriy Mulyar and Bridget McInnes

Acknowledgments