CardiSort: a convolutional neural network for cross vendor automated sorting of cardiac MR images
This is a deep learning network which was developed to sort cardiac MRI images by sequence type and imaging plane, facilitating efficient and fully automated post-processing pipelines.
Please find the pre-trained model weights here.
The model can be run to sort a folder of DICOM images as:
>> python cardisort_inference.py [INPUT_FOLDER]
or
>> python cardisort_inference.py [INPUT_FOLDER] [OUTPUT_FOLDER]
This assumes that the input folder contains only one CMR study from one patient. For an idea of how to run on multiple input folders from multiple patients see run_multiple.sh. You will need to update the path to the python environment to fit your own setup.
If you find this code helpful in your research please cite the following paper:
Ruth P Lim, Stefan Kachel, Adriana DM Villa, Leighton Kearney, Nuno Bettencourt, Alistair A Young, Amedeo Chiribiri, and Cian M Scannell. CardiSort: a convolutional neural network for cross vendor automated sorting of cardiac MR images. arXiv preprint arXiv:2109.08479, 2021.
@misc{lim2021cardisort,
title={CardiSort: a convolutional neural network for cross vendor automated sorting of cardiac MR images},
author={Ruth P Lim and Stefan Kachel and Adriana DM Villa and Leighton Kearney and Nuno Bettencourt and Alistair A Young and Amedeo Chiribiri and Cian M Scannell},
year={2021},
eprint={2109.08479},
archivePrefix={arXiv},
primaryClass={eess.IV}
}