A test implementation for the submitted paper "BPGAN: Bidirectional CT-to-MRI Prediction using Multi-Generative Multi-Adversarial Nets with Spectral Normalization and Localization" (Under reviewing)

python 3.6

pytorch (Pytorch http://pytorch.org/)

torchvision

numpy

time

pandas

scipy

1. Download pre-trained models from BaiduNetdisk. password: ciw9.

2. Download partial test samples from BaiduNetdisk, then put all this data into corresponding dir and extract compressed files.

3. Copy the model (net_G_A: CT predictor, net_G_B: MRI predictor) into your dir

   cp latest_net_G_A.pth ./brain_model/

   cp latest_net_G_B.pth ./brain_model/

4. Test for CT or MRI precition from MRI or CT images in proposed BPGAN

   python test.py --dataroot ./datasets/brain --name brain_model

• The implementation of proposed BPGAN model is based on cycle-GAN (https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix). We improve the cycle-GAN by introducing pathological auxiliary information, spectral normalization, localization and edge retention to achieve the bidirectional prediction between CT and MRI images.
• This is developed on a Linux machine running Ubuntu 16.04.
• Use GPU for the high speed computation.
• Due to partial samples in SPLP dataset related to private information, so please e-mail me (xulimmail@gmail.com) if you need the dataset and I will share a private link with you.