Designed after this paper on volumetric segmentation with a 3D U-Net. Currently, the network does not have the B-Spline deformations that are mentioned in the paper. If you figure out a way to apply these to a 3D Keras CNN, let me know! PRs are always welcome!

The code was written to be trained using the BRATS data set for brain tumors, but it can be easily modified to be used in other 3D applications. To adapt the network, you might have to play with the input size to get something that works for your data.

I used Bohdan Pavlyshenko's Kaggle kernel for 2D U-Net segmentation as a base for this 3D U-Net.

1. Download the BRATS 2017 GBM and LGG data. Place the unzipped folders in the brats/data/original folder.
2. Install dependencies:

nibabel,
keras,
pytables,
nilearn,
SimpleITK,
nipype

(the last two are for preprocessing only)

3. Install ANTs N4BiasFieldCorrection and add the location of the ANTs binaries to the PATH environmental variable.

4. Add the repository directory to the PYTONPATH system variable:

$ export PYTHONPATH=${PWD}:$PYTHONPATH

5. Convert the data to nifti format and perform image wise normalization and correction:

$ cd brats

Import the conversion function and run the preprocessing:

$ python
>>> from preprocess import convert_brats_data
>>> convert_brats_data("data/original", "data/preprocessed")

6. Run the training:

$ python train.py

In the training above, part of the data was held out for validation purposes. To write the predicted label maps to file:

$ python predict.py

The predictions will be written in the prediction folder along with the input data and ground truth labels for comparison.

By changing the configuration dictionary in the brats/train.py file, makes it easy to test out different model and training configurations. If you are running out of memory, try reducing the "batch_size" parameter. A smaller batch size will feed smaller chunks of data to the CNN. If the batch size is reduced down to 1 and it still you are still running out of memory, you could also try changing the patch size to (32, 32, 32). Keep in mind, though, that a smaller patch size will likely not perform as well.

If you want to train a 3D UNet on a different set of data, you can copy the brats/train.py file and modify to read in your data rather than the preprocessed BRATS data that is currently setup to train on.