PyTorch implementation of the Pix2PixHD network for generating synthetic dynamic contrast enhanced (DCE) subtraction images from native (T1-/T2-weighted) sequences and low-dose contrast DCE subtraction images.

• Split into unilateral breasts and cut volumes into slices along the axial axis
• Resample images to 0.64 x 0.64 mm
• Crop or Zero-Pad Images to 256 X 224 Pixels
• Stack [T2, T2, LowDose] into a 3-channel Image
• Scale Image to [0,1]
• Normalize each channel so that Mean = 0.5 and STD=0.5
• Store files as .tiff in a folder called test_A

• Eventually see instructions given in the Pix2PixHD repository
• Change the following options in /options/base_options.py:
  • --name select desired model from ./checkpoints. Name must match folder name, e.g. 'Lowdose_T2'
  • --dataroot set to directory that contains 'test_A' eg. /mnt/datasets/breast/
• Change the following options in /options/test_options.py:
  • results_dir set to desired output directory eg. '/mnt/datasets/breast/pred_B'
• Run test.py

Please see our paper at:

This code is largely based on the Pix2PixHD repository.