joannabroniarek / dcgan-for-mri-images

Tensorflow 2.0 implementation of DCGAN for 3D MRI

DCGAN-for-MRI-images

Inspiration for this project was the paper: http://www.nlab.ci.i.u-tokyo.ac.jp/pdf/isbi2018.pdf

The idea was to implement and train DCGAN model on the 3D MRI images from BRATS 2019 dataset.

Demo

To see a short explanation of the implementation as well as to generate new images, please see the demo notebook

Repository Content

• train3D.py - the file to be run (with training functions and 'main' function)

• models.py - contains implementation of discriminator and generator

• create_data.py - functions used for data preprocessing and generation of the TfRecords file

• utils.py - additional useful functions used to i.e. plotting images

• docs/ - a folder with gif / figures

Custom Training

$ python3 train3D.py
	--epochs 100
	--batch_size 16
	--lr_g 5e-4
	--lr_d 5e-5
	--rand_seed 42

Sample run for 100 epochs

---

To Do:

• requirements txt
• pre-trained model upload

dcgan-for-mri-images's Issues

Whitening function not found

whitening function to normalize the image is not declared.