dcm-classifier 0.4.4

This is a consolidation of work from NAMIC efforts primarily at the University of Iowa.

Introduction

In this work, we developed a robust, easily extensible classification framework that extracts key features from well-characterized DICOM header fields to identify image modality and acquisition plane. Utilizing classical machine learning paradigms and a heterogeneous dataset of 9121 scans collected at 12 sites, using 23 scanners from 6 manufacturers, we achieved 99.4% accuracy during the K-Fold Cross-Validation for classifying 11 image modalities and 99.96% accuracy on image acquisition plane classification. Furthermore, we demonstrated model generalizability by achieving 98.6% accuracy on out-of-sample animal data. Our proposed framework can be crucial in eliminating error-prone human interaction, allowing automatization, and increasing imaging applications' reliability and efficiency.

This work was submitted for publication at the 2024 SPIE Medical Imaging conference.

This project was supported by Botimageai.

Instructions

Below are useful commands to start using the tool.

Clone git repo

$ git clone https://research-git.uiowa.edu/SINAPSE/dicomimageclassification.git

Navigate to the cloned repo

$ cd <repo path>

Setup virtual environment

$ python3 -m venv <venv_path> && source <venv_path>/bin/activate

Install required packages

$ pip install -r REQUIREMENTS.txt

Run the script!

python3 classify_study.py -m models/rf_classifier.onnx -d <path_to_dicom_session>

Testing

  pytest

Coverage Analysis

 coverage run --concurrency=multiprocessing --parallel-mode -m pytest tests --junitxml=tests/pytest.xml
 coverage combine
 coverage report --format=text -m |tee tests/pytest-coverage.txt
 coverage xml -o tests/coverage.xml
 coverage xml -o tests/coverage.xml

Authors

1. Michal Brzus

   github: mbrzus, email: [e-mail]

2. Hans J. Johnson

   github: BRAINSia, email: [e-mail]