README.md

BrainTumorSegmentation

Spread sheet including Results and related works is available in here.

Dataset

Dataset is available in here.

1. Training: 3D MRI images from 369 patients (Height=240, Width=240, Depth=155)
   • Images: 4 modalities
   • Masks: 4 classes(0, 1, 2, and 4)

2. Validation: 3D MRI images from 125 patients (Height=240, Width=240, Depth=155)
   • Images: 4 modalities

Preprocess and Augmentation

1. Flip

2. Brightness

3. Rotation

4. Elastic

5. Shift

6. Swirl

Models

1. UNet3D

2. AttUnet3D

3. GAN (Vox2Vox)

4. AttGAN

5. UNet3D with Deformable Convolution

Metrics

1. Dice Loss

2. Generator Loss

3. Discriminator Loss

Requirements

• tensorflow
• tensorflow-addons
• tensorflow-cpu
• numpy
• matplotlib
• elasticdeform
• scikit-learn
• scipy
• nibabel
• SimpleITK

Install the requirements using command below:

pip install -r requirements.txt 

Run

Use the following command to see the arguments needed for running:

python -m scripts.main -h

For running UNet3D model use this:

python -m scripts.main -bs 4 -ps 128 -a 5 -ne 1 -ef 0.25 -lr 1e-3 -b1 0.9 -ds 100 -np 2 -aug 1 -m unet

For running AttUnet3D model use this:

python -m scripts.main -bs 4 -ps 128 -a 5 -ne 1 -ef 0.25 -lr 1e-3 -b1 0.9 -ds 100 -np 2 -aug 1 -m att_unet

For running GAN model use this:

python -m scripts.main -bs 4 -ps 128 -a 5 -ne 1 -ef 0.25 -lr 1e-3 -b1 0.9 -ds 100 -np 2 -aug 1 -m gan

For running AttGAN model use this:

python -m scripts.main -bs 4 -ps 128 -a 5 -ne 1 -ef 0.25 -lr 1e-3 -b1 0.9 -ds 100 -np 2 -aug 1 -m att_gan

For running UNet3D_DCN model use this:

python -m scripts.main -bs 4 -ps 16 -a 5 -ne 1 -ef 0.25 -lr 1e-3 -b1 0.9 -ds 100 -np 1 -aug 1 -m unet_dc

For running AttUnet3DDCN model use this:

python -m scripts.main -bs 4 -ps 16 -a 5 -ne 5 -ef 0.25 -lr 1e-3 -b1 0.9 -np 1 -aug 1 -m att_unet_dc

For running AttGANDCN model use this:

python -m scripts.main -bs 1 -ps 16 -a 5 -ne 5 -ef 0.25 -lr 1e-3 -b1 0.9 -np 1 -aug 1 -m att_gan_dc

References

• Vox2Vox: 3D-GAN for Brain Tumour Segmentation
• Attention Unet
• Deformable Convolution