This example implements the segmentation network of the paper Multi-label Whole Heart Segmentation Using CNNs and Anatomical Label Configurations. The results when running the scripts may not be competely the same, as we originally trained the network with caffe and not with tensorflow. See the localization folder for examples of how to train networks for localization.
You need to have the MedicalDataAugmentationTool framework downloaded and in you PYTHONPATH for the scripts to work. If you have problems/questions/suggestions about the code, write me a mail!
Download the files from the challenge website. In order for the framework to be able to load the data, they must be preprocessed first with the following commands using the program c3d. Update the paths accordingly:
The labels need to be sorted according to label value ranging from 1 to 7:
for x in *_label.nii.gz; do c3d ${x} -replace 500 1 600 2 420 3 550 4 205 5 820 6 850 7 -o ${${x:r}:r}_sorted.nii.gz; done
The images need to be converted to .mha files:
for x in *.nii.gz; do c3d ${x} -compress -type short -o ../mr_mha/${${x:r}:r}.mha; done
The images need to be reoriented to RAI:
for x in *_sorted.mha; do c3d ${x} -compress -type short -orient RAI -o ../mr_mha/${x}; done
After these commands, copy the resulting files into the folders mmwhs_dataset/mr_mha and mmwhs_dataset/ct_mha accordingly. For each file from the training dataset, there should be an image and label file (e.g., ct_train_1001_image.mha and ct_train_1001_label.mha). For each file from the testing dataset, there should be an image file (e.g., ct_test_2010_image.mha)
Update: If the program c3d generates error messages, you could also try to run the script reorient.py, which should produce the same results.
Run main.py to train the network. Adapt parameters in the file to use either MR or CT and to define cross validation or the full training/testing.
In order to train and test on other datasets, modify the dataset.py file. See the example files and documentation for the specific file formats. Set the parameter save_debug_images = True in order to see, if the network input images are reasonable.
If you use this code for your research, please cite our paper and the overview paper of the MMWHS2017 challenge
@inproceedings{Payer2018a,
title = {Multi-label Whole Heart Segmentation Using {CNNs} and Anatomical Label Configurations},
author = {Payer, Christian and {\v{S}}tern, Darko and Bischof, Horst and Urschler, Martin},
booktitle = {Statistical Atlases and Computational Models of the Heart. ACDC and MMWHS Challenges. STACOM 2017},
doi = {10.1007/978-3-319-75541-0_20},
pages = {190--198},
year = {2018}
}
@article{Zhuang2019,
title = {Evaluation of Algorithms for Multi-Modality Whole Heart Segmentation: An Open-Access Grand Challenge},
author = {Zhuang, Xiahai and Li, Lei and Payer, Christian and {\v{S}}tern, Darko and Urschler, Martin and Heinrich, Mattias P. and Oster, Julien and Wang, Chunliang and Smedby, {\"{O}}rjan and Bian, Cheng and Yang, Xin and Heng, Pheng-Ann and Mortazi, Aliasghar and Bagci, Ulas and Yang, Guanyu and Sun, Chenchen and Galisot, Gaetan and Ramel, Jean-Yves and Brouard, Thierry and Tong, Qianqian and Si, Weixin and Liao, Xiangyun and Zeng, Guodong and Shi, Zenglin and Zheng, Guoyan and Wang, Chengjia and MacGillivray, Tom and Newby, David and Rhode, Kawal and Ourselin, Sebastien and Mohiaddin, Raad and Keegan, Jennifer and Firmin, David and Yang, Guang},
journal = {Medical Image Analysis},
doi = {10.1016/j.media.2019.101537},
pages = {101537},
volume = {58},
year = {2019}
}