wmh_segmentation_challenge

Documentation

• https://wmh.isi.uu.nl/data/

Implementation

U-NET model for white matter segmentation

Results

• https://drive.google.com/file/d/11OSFLJ2qMyQj5LbVa2l9QQTvjYIcLfbn/view?usp=sharing

Dataset

The dataset is the one used during the WMH Segmentation Challenge In our repository, it should be placed under a datasets/ folder

Generate the data for training

Before training the model, we need to generate the data that will be used, to do this, you will need to launch:

• python generate_data.py <path/to/datasets> <save/dir>

Note that the <save/dir> is the directory where the train.pickle file will be stored.

Usage

cd models/clean

Training

python3 main.py <path/to/dataset.pickle> preprocess:0|1 <3d:2|3>

Please look at the file main.py for more information regarding the parameters

Examples of training usage:

• python3 main.py data/train.pickle 0 2
• python3 main.py data/train.pickle 1 2
• python3 main.py data/train.pickle 0 3
• python3 main.py data/train.pickle 1 3

Testing

python3 main.py <path/to/dataset.pickle> preprocess:0|1 <3d:2|3>

<path/to/dataset.pickle> should be something like */test.pickle

Please look at the file main.py for more information regarding the parameters After generating the weights and the result file, you can test the model.

python3 test_model.py <path/to/dataset> <path/to/test_pickle> <result.pickle>

With:

• <path/to/dataset> The root folder of the dataset
• <path/to/test_pickle> The same argument given to <path/to/dataset.pickle>
• <result.pickle> The file generated by the command above with "python3 main.py ...."