This repository is the work of "HDC-Net: Hierarchical Decoupled Convolution Network for Brain Tumor Segmentation" based on pytorch implementation.The multimodal brain tumor dataset (BraTS 2018) could be acquired from here.

• python 3.6
• pytorch 0.4 or 1.0
• nibabel
• pickle
• imageio
• pyyaml

Download the BraTS2018 dataset and change the path:

experiments/PATH.yaml

Convert the .nii files as .pkl files. Normalization with zero-mean and unit variance .

python preprocess.py

(Optional) Split the training set into k-fold for the cross-validation experiment.

python split.py

Sync bacth normalization is used so that a proper batch size is important to obtain a decent performance. Multiply gpus training with batch_size=10 is recommended.The total training time is about 12 hours and the average prediction time for each volume is 2.3 seconds when using randomly cropped volumes of size 128×128×128 and batch size 10 on two parallel Nvidia Tesla K40 GPUs for 800 epochs.

python train_all.py --gpu=0,1 --cfg=HDC_Net --batch_size=10

You could obtain the resutls as paper reported by running the following code:

python test.py --mode=1 --is_out=True --verbose=True --use_TTA=False --postprocess=True --snapshot=True --restore=model_last.pth --cfg=HDC_Net --gpu=0

Then make a submission to the online evaluation server.

If you use our code or model in your work or find it is helpful, please cite the paper:

1. DMFNet
2. BraTS2018-tumor-segmentation
3. Synchronized-BatchNorm-PyTorch