Individual Vertebrae Segmentation on VerSe Dataset

This is the implementation of iterative vertebrae segmentation on VerSe datasets (2019 and 2020). Vertebrae segmentation is accomplished by first finding the spine location on CT volume and then performing an iterative segmentation of vertebrae from top to down:

1. Usage

After moving to the repo directory, the first thing to do is to clone ML_3D_Unet repo which is needed:

$ git clone https://github.com/luiserrador/ML_3D_Unet.git

Download VerSe dataset:

_{Note: We strongly recommend to test the repository with the data available. So, test steps 1.1 and 1.2 before downloanding all the data!}

$ python get_verse_data.py

1.1 Training spine location network

$ python pre_processing_heatmap.py
$ python train_heatmap.py

1.2 Training segmentation network

$ python pre_processing_segmentation.py
$ python train_segmentation.py

BibTeX Citation

If you use this repo in a scientific publication, we would appreciate using the following citations:

@article{Serrador2024,
  title = {Knowledge distillation on individual vertebrae segmentation exploiting 3D U-Net},
  volume = {113},
  ISSN = {0895-6111},
  url = {http://dx.doi.org/10.1016/j.compmedimag.2024.102350},
  DOI = {10.1016/j.compmedimag.2024.102350},
  journal = {Computerized Medical Imaging and Graphics},
  publisher = {Elsevier BV},
  author = {Serrador,  Luís and Villani,  Francesca Pia and Moccia,  Sara and Santos,  Cristina P.},
  year = {2024},
  month = apr,
  pages = {102350}
}