MRI-Generation

Author: Antoine DELPLACE
Last update: 10/02/2020

This repository corresponds to the source code used for the MRI generation part of my Master Thesis entitled "Segmentation and Generation of Magnetic Resonance Images by Deep Neural Networks".

Method description

The aim of the project is to generate high resolution brain Magnetic Resonance Images (MRIs) from a random latent space (noise) thanks to Generative Adversarial Networks (GANs). Three architectures are tuned and compared : Deep Convolutional GAN (DCGAN), Super Resolution Residual GAN (SRResGAN) and Progressive GAN (ProGAN).

Usage

Dependencies

• Python 3.6.8
• Tensorflow 1.14
• Numpy 1.16.2
• Matplotlib 3.0.3
• Imageio 2.5.0
• Scikit-image 0.15.0 -- analyze_plots_prog.py

File description

1. The training files correspond to dcgan3.py, srresgan1.py and progan5.py in their respective folder.

2. The files analyze_plots*.py and graph_loss*.py enable the monitoring of the training process thanks to plots of the ouputs, the loss functions and the processing time. Note that you need to use the files ending by "_prog.py" for ProGAN.

3. The files *_generate_data.py and *_generate_interpolation.py enable the generation of images from the trained models.

4. The file evaluate_gan.py enables the evaluation of the performance of the model, based on the generated images.

Results

The comparison and analysis of the different architectures can be found in the conference and thesis papers. Here, I just display the result of the training process, especially the animations composed of the generated outputs during training.

DCGAN

analyze_plots.py dcgan3

training_animations/dcgan3.gif

SRResGAN

analyze_plots.py srresgan1

training_animations/srresgan1.gif

ProGAN

analyze_plots_prog.py progan5

training_animations/progan5.gif

References

1. A. Delplace. "Synthetic Magnetic Resonance Images with Generative Adversarial Networks", Conference paper at the University of Queensland, October 2019. arXiv:2002.02527
2. A. Delplace. "Segmentation and Generation of Magnetic Resonance Images by Deep Neural Networks", Master thesis at the University of Queensland, October 2019. arXiv:2001.05447