This code performs automatic segmentation of L4-L5 and L5-S1 CEPs from sagittal UTE (Ultra-short Echo Time) spine MR images
Segmentation is performed via a U-net (Ronneberger et al., 2015) that was trained on UTE images and manual segmentations.
Set up a conda environment if you do not have all the packages/compatible versions (the list of dependencies is listed in environment.yml).
Create and activate virtual environment using conda
Set-up environment using conda (make sure you are in the CEP-seg code repository):
conda env create -f environment.yml
The default name of the environment is CEPseg. Activate the environment with source activate CEPseg, and deactivate with source deactivate.
Use conda info CEPseg to see more information about the environment and ensure that it was installed properly.
Make sure you are in the CEP-seg directory.
source activate CEPseg
jupyter notebook
Open prediction.ipynb.
Under User inputs, change level ('l4l5inf', 'l4l5sup', 'l5s1inf', 'l5s1sup'), img_path (where MR images are stored) and pred_path (where masks will be created) as desired.
Run prediction.ipynb. Masks will be created in the directory specified as pred_path.