SSinyu/ct-denoising

ct-denoising

Denoising Model

• RED_CNN

• paper :https://arxiv.org/ftp/arxiv/papers/1702/1702.00288.pdf

• WGAN_VGG

• paper : https://arxiv.org/pdf/1708.00961.pdf
• original code:
  • vgg : https://github.com/machrisaa/tensorflow-vgg
  • WGAN : https://github.com/jiamings/wgan

• CYCLEGAN

• paper : https://arxiv.org/abs/1703.10593
• original code: https://github.com/xhujoy/CycleGAN-tensorflow

I/O (DICOM file -> .npy)

• Input data Directory
  • DICOM file extension = ['.IMA', '.dcm']

$ os.path.join(dcm_path, patent_no, [LDCT_path|NDCT_path], '*.' + extension)

[Common] Main file(main.py) Parameters

• Directory

• dcm_path : dicom file directory
• LDCT_path : LDCT image folder name
• NDCT_path : NDCT image folder name
• test_patient_no : default : L067,L291
• checkpoint_dir : save directory - trained model
• test_npy_save_dir : save directory - test numpy file
• pretrained_vgg : pretrained vggnet directory(only WGAN_VGG)

• Image info

• patch_size : patch size (WGAN_VGG, RED_CNN)
• whole_size : whole size
• img_channel : image channel
• img_vmax : max value
• img_vmin : min value

• Train/Test

• model : red_cnn, wgan_vgg, cyclegan (for image preprocessing)
• phase : train | test

• others

• is_mayo : summary ROI sample1,2
• save_freq : save a model every save_freq (iterations)
• print_freq : print_freq (iterations)
• continue_train : load the latest model: true, false
• gpu_no : visible devices(gpu no)