This repository contains the source code of the segmentation package. This package provides utility functions to be used for CNN-based segmentation of medical images. The code contained in the repository was used for data processing of the following paper:
"D. Marzorati, M. Sarti, L. Mainardi, A. Manzotti and P. Cerveri, "Deep 3D Convolutional Networks to Segment Bones Affected by Severe Osteoarthritis in CT Scans for PSI-Based Knee Surgical Planning," in IEEE Access, vol. 8, pp. 196394-196407, 2020, doi: [10.1109/ACCESS.2020.3034418](10.1109/ACCESS.2020.3034418)."
- `segmentation.callbacks`: callbacks used during the training of the network
- `segmentation.cnn`: CNN-based models to be used for network architecture
- `segmentation.losses`: loss functions for segmentation tasks
- `segmentation.metrics`: metrics to be used during the training of the network
- `segmentation.mesh`:
- `segmentation.utils`: utils functions
The segmentation.preprocess is a submodule to preprocess dicom or Nifti (.nii) files to adapt data for training and optimize memory consumption: - `segmentation.preprocess.main_preprocess`: this module contains the main preprocessing functions - `segmentation.preprocess.Cycles`: this module contains example functions to loop over files and call the main functions - `segmentation.preprocess.utils`: utils functions used for various purposes
You can install the package directly from GitHub: pip install +git https://github.com/dado93/cnn-segmentation
You can generate your local documentation by running: make docs to update the documentation of the project. Updated documentation can be found in the docs folder. Documentation of the package can be found at the following link: [https://dado93.github.io/cnn-segmentation/](https://dado93.github.io/cnn-segmentation/)
- [x] Documentation with Read The Docs
- [ ] New format for loss functions