• python2.7
• SimpleITK
• Medpy
• keras
• h5py
• 3DSlicer

Using Brats 2015 dataset

As follows make diretories. Unzip brats dataset in data directory.

U-net3D

data

HGG LGG HGG_LGG output source

1. Run data_rename.py in U-net3D directory.

This will rename brats dataset.

2. Run mha_to_nii.py

This will change MRI ground truth file extension(.mha to .nii) and move file to /data/data directory

3. Using 3D slicer for doing N4ITKBiasFieldCorrection.

N4N4ITKBiasFieldCorrection remove image gradation in your MRI dataset.

cd /U-net3d/data
for n in *.mha; do ~/slicer/lib/Slicer-4.7/cli-modules/N4ITKBiasFieldCorrection "./$n" ./data/"${n%.mha}.nii"; done

3D slicer's file root maybe diffenent.

4. Run pre_process.py

This source contains three pre-processes

Intensity Range Standardization Histogram equalizing Gauss Normalization

Run train.py

Run predict.py

• 윤지석, 석홍일 (2016). 딥러닝 기반의 멀티-모달 MRI 영상에서의 뇌종양 영역 분할. 한국정보과학회 학술발표논문집, 1680-1682
• Kayalibay, B., Jensen, G., & van der Smagt, P. (2017). CNN-based segmentation of medical imaging data. arXiv preprint arXiv:1701.03056.