RESSEG: segmentation of postoperative brain cavities on 3D MRI using deep learning

This is the code for Pérez-García et al., 2021, A self-supervised learning strategy for postoperative brain cavity segmentation simulating resections - International Journal of Computer Assisted Radiology and Surgery (IJCARS).

If you use this code or the EPISURG dataset for your research, please cite this publication as:

Pérez-García, F., Dorent, R., Rizzi, M. et al. A self-supervised learning strategy for postoperative brain cavity segmentation simulating resections. Int J CARS (2021). https://doi.org/10.1007/s11548-021-02420-2

BibTeX:

@article{perez-garcia_self-supervised_2021,
	title = {A self-supervised learning strategy for postoperative brain cavity segmentation simulating resections},
	issn = {1861-6429},
	url = {https://doi.org/10.1007/s11548-021-02420-2},
	doi = {10.1007/s11548-021-02420-2},
	language = {en},
	urldate = {2021-06-14},
	journal = {International Journal of Computer Assisted Radiology and Surgery},
	author = {P{\'e}rez-Garc{\'i}a, Fernando and Dorent, Reuben and Rizzi, Michele and Cardinale, Francesco and Frazzini, Valerio and Navarro, Vincent and Essert, Caroline and Ollivier, Ir{\`e}ne and Vercauteren, Tom and Sparks, Rachel and Duncan, John S. and Ourselin, S{\'e}bastien},
	month = jun,
	year = {2021},
}

Installation

$ conda create -n ijcars python=3.7 ipython -y && conda activate ijcars
$ conda install pytorch=1.7.1 torchvision=0.8.2 cudatoolkit=11.0 -c pytorch -y
$ pip install -r requirements.txt

Related projects

resseg

The trained models can be used to segment easily using resseg.

resector

Resections can be simulated on 3D MRI using resector.

EPISURG dataset

See the EPISURG extension for 3D Slicer.

Commands used for training

Using simulated resections only

Augmentation

python main.py with config/simulated/config_simulated_no_augment.yml

Shape

Cuboids

python main.py with config/simulated/shape/config_simulated_shape_cuboid.yml

Ellipsoids

python main.py with config/simulated/shape/config_simulated_shape_ellipsoid.yml

Noisy ellipsoids (baseline)

python main.py with config/simulated/config_simulated_baseline.yml

Texture

Percentile 1

python main.py with config/simulated/texture/config_simulated_texture_dark.yml

Percentile 1, 99

python main.py with config/simulated/texture/config_simulated_texture_random.yml

CSF (baseline)

python main.py with config/simulated/config_simulated_baseline.yml

CSF + WM

python main.py with config/simulated/texture/config_simulated_texture_wm.yml

CSF + BC

python main.py with config/simulated/texture/config_simulated_texture_clot.yml

CSF + WM + BC

python main.py with config/simulated/texture/config_simulated_texture_clot_wm.yml

Using clinical data from hospitals in UK, Italy and France

Train

Load and tune

python main.py with config/real/config_load_train.yml with dataset_name $DATASET