No description, website, or topics provided.
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
CYCLE_IDENTITY_GAN
RED_CNN
WGAN_VGG
cyclegan
README.md
inout_util.py

README.md

ct-denoising

Denoising Model

  • RED_CNN
  • WGAN_VGG
  • CYCLEGAN

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)

The dcm_path directory should look like:

dcm_path
├── L067
│   ├── quarter_3mm
│   │       ├── L067_QD_3_1.CT.0004.0001 ~ .IMA
│   │       ├── L067_QD_3_1.CT.0004.0002 ~ .IMA
│   │       └── ...
│   └── full_3mm
│           ├── L067_FD_3_1.CT.0004.0001 ~ .IMA
│           ├── L067_FD_3_1.CT.0004.0002 ~ .IMA
│           └── ...
├── L096
│   ├── quarter_3mm
│   │       └── ...
│   └── full_3mm
│           └── ...      
...
│
└── L506
    ├── quarter_3mm
    │       └── ...
    └── full_3mm
            └── ...     

[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)