cycontext 0.1.3

ConText algorithm using spaCy for clinical NLP

cycontext

A Python implementation of the ConText algorithm for clinical text concept assertion using the spaCy framework

Overview

This package implements the ConText algorithm within the spaCy framework. ConText detects semantic modifiers such as negation, family history, and certainty in clinical text by relating target concepts (such as "pneumonia") with semantic modifiers (such as "no evidence of").

This builds on pyConText, which extracts both targets and modifiers using regular expressions and relates them using a NetworkX graph.

See notebooks/ for detailed explanations and examples.

Before getting started, make sure you've installed spacy and downloaded a model (ie., en_core_web_sm).

Key Features

• cycontext is used as a modular spaCy component
• Modifiers are defined in a knowledge base which are used to modify target spans
• Results are stored in a custom attribute Doc._.context_graph
• Modifier -> target relationships are stored in Doc._.context_graph.edges

Basic Usage

Installation

You can install cycontext using pip:

pip install cycontext

Or clone this repository install cycontext using the setup.py script:

$ python setup.py install

Once you've installed the package and spaCy, make sure you have a spaCy language model installed (see https://spacy.io/usage/models):

$ python -m spacy download en_core_web_sm

In the future cycontext should be made available through conda.

Example

import spacy

from cycontext import ConTextComponent, ConTextItem

nlp = spacy.load("en_core_web_sm")
text = "There is no evidence of pneumonia."

###########################################
# Add code for extracting target concepts #
###########################################
doc = nlp(text)
print(doc.ents)
>>> (pneumonia,)

item_data = [ConTextItem("no evidence of", "DEFINITE_NEGATED_EXISTENCE", rule="forward")]
context = ConTextComponent(nlp)
context.add(item_data)
nlp.add_pipe(context, last=True)

doc = nlp(text)
print(doc._.context_graph.edges)
>>> [(pneumonia, <TagObject> [No evidence of, definite_negated_existence])]

for target in doc.ents:
    print(target._.is_experienced)

>>> False

Visualization

We can use spaCy's visualizers to display the results of cycontext. This NER-style visualization highlights the targets and modifiers in text:

from cycontext import viz
viz.visualize_ent(doc)

While this dependency-style visualization shows the relationships between targets and modifiers:

viz.visualize_dep(doc)

Tests

To run tests:

cd cycontext
pytest tests

Contact Information

Alec Chapman: [e-mail]