RADTorch provides a framework of higher level classes and functions that aim at significantly reducing the time needed for implementation of different machine and deep learning algorithms on DICOM medical images.
RADTorch was built by radiologists for radiologists so they can build, test and implement state-of-the-art machine learning algorithms in minutes.
RADTorch is built upon widely used machine learning and deep learning frameworks. These include:
-
PyTorch for Deep Learning and Neural Networks.
-
Scikit-learn for Data Management and Machine Learning Algorithms.
-
PyDICOM for handling of DICOM data.
-
Bokeh, Matplotlib and Seaborn for Data Visualization.
Running a state-of-the-art DICOM image classifier can be run using the Image Classification Pipeline using the commands:
from radtorch import pipeline
classifier = pipeline.Image_Classification(data_directory='path to data')
classifier.run()
RADTorch playground for testing is provided on Google Colab.
Full Documentation at official website: https://docs.radtorch.com
Feature requests are more than welcomed on our discussion board HERE
RADTorch was developed by Mohamed Elbanan, MD: a Radiology Resident at Yale New Haven Health System, Clinical Research Affiliate at Yale School of Medicine and a Machine-learning enthusiast.
Bug reports and pull requests are welcome.
This framework is provided for free and opensource to anyone who would like to use it and/or update it.
Please cite us as :
@software{mohamed_elbanan_2020_3827986,
author = {Mohamed Elbanan},
title = {radtorch/radtorch: 1.1.1},
month = may,
year = 2020,
publisher = {Zenodo},
version = {1.1.1},
doi = {10.5281/zenodo.3827986},
url = {https://doi.org/10.5281/zenodo.3827986}
}