medstruct

Extract structure from medical narrative (Clinical NLP)

Classifiers

Lung cancer classification (TNM-8)

A rule-based algorithm to perform lung cancer t-stage classification on radiology reports 'Lung - Cancer TNM 8th edition'.

Pre-configured languages are Dutch (TN) and English (T-only)

TNM-8 Lung	acc. Dutch (n=100)	acc. English (n=200)
T	~ 0.85	~ 0.85
N	~ 0.90	TODO
M	TODO	TODO

Mentions containing uncertainty are detected and not used for classification!

Configuration of synonyms and context is included for English and Dutch, but should always be validated for your corpus. Reported scores are based on the use of a sectionizer, depending on the structure of your reports, a sectionizer has to be reimplemented.

1. Requirements

docker
git

2. Install

git clone https://github.com/putssander/medstruct-config.git

cd medstruct-config

docker network create medstruct-network

3. RUN

docker-compose pull && docker-compose up -d

3.1 GUI

The medstruct user interface will come available at: http://localhost:8080

3.2 Excel Lung TNM-classification

If you want to perform classification on a excel file containing a batch for reports see maastroclinic/medstruct.

To run the algorithm on an excel sheet.

   docker run -ti --rm --net medstruct-network -v '[LOCAL_PATH_IN_OUT]:/data/' -v ${PWD}/application.yml:/app/application.yml putssander/medstruct:2.0 python medstruct/xlsx_run.py /data/[XLSX_FILE_NAME] [RESULT_NAME]

example

   docker run -ti --rm --net medstruct-network -v ${PWD}/data/:/data/ -v ${PWD}/medstruct/medstruct-application.yml:/app/application.yml registry.gitlab.com/medstruct/medstruct:2.1 python medstruct/xlsx_run.py /data/reports/TNMTest1.xlsx training-set

example 2 (used Zuyderland)

   docker run -ti --rm --net medstruct-network -v ${PWD}/data/:/data/ -v ${PWD}/medstruct/medstruct-application.yml:/app/application.yml registry.gitlab.com/medstruct/medstruct:2.1 python medstruct/xlsx_run.py /data/reports/TNMTest1Zuyderland_TEST50.xlsx training-set

Output can be configured in ${PWD}/medstruct/medstruct-application.yml, example

    medstruct.output-dir: '/data/reports/output.xlsx'

4. Overview & configuration

The medstruct pipeline consists out of several reusable components/micro-services with each its own repository. Detailed configuration options are described in the repositories of each individual service.

Overview of medstruct repositories/services:

Configuration:

putssander/medstruct-config (this repository): configuration repository.

Required components:

maastroclinic/medstruct: pre-processing/management, main-API, text-cleaning, sectionizer, read/write evaluation excel data
putssander/spacy-json-nlp: annotate text with spaCy into the JSON-NLP format
maastroclinic/pycontextnlp: JSON-NLP wrapper around pyContextNLP to annotate clinical targets (clinical concepts) and context (negations/uncertainty)
putssander/medstruct-measurement-extractor: extract and link measurement expressions to clinical concepts
putssander/medstruct-tnm-classifier: lung classification (tnm-8)

Optional components:

putssander/medstruct-gui: medstruct Graphical User Interface

5. Docker Development

To set up a development environment, where docker-compose applications communicate with applications running on your host. For the applications you want to run on your host machine (for development):

In docker-compose.yml place the services in comment
In ./medstruct/medstruct-application.yml set the hostname to host.docker.internal

6. FAQ

"not a directory"

  \\\"not a directory\\\"\"": unknown: Are you trying to mount a directory onto a file (or vice-versa)?

On Windows make sure to use the latest docker version (2.1.0.4 or newer)

putssander / medstruct-config