Low-Dose CT with a Residual Encoder-Decoder Convolutional Neural Network (RED-CNN)

• RED_CNN

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

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

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

• 10 layers (5 conv + 5 deconv)
• shortcut
• remove pooling operation
• filter size : 96 * 9 + 1* 1(last layer)
• kernel size : 5 * 5
• stride : 1 (no padding)

• patch size : 55 * 55
• augumentation(patch? whole? // patch...)

• rotation(45 degrees)
• flipping (vertical & horizontal)
• scaling (0.5, 2)

• learning rate : 10e-4 (slowly decreased down(?))
• initializer : random Gaussian distribution (0, 0.01)
• loss function : MSE
• optimizer : Adam

• Directory

• dcm_path : dicom file directory
• LDCT_path : LDCT image folder name
• NDCT_path : NDCT image folder name
• test_patient_no : test patient id list(p_id1,p_id2...) (train patient id : (patient id list - test patient id list)
• checkpoint_dir : save directory - trained model
• test_npy_save_dir : save directory - test numpy file
• pretrained_vgg : pretrained vggnet directory

• Image info

• patch_size : patch size
• 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)

• Training detail

• num_iter : iterations (default = 200000)
• alpha : learning rate (default=1e-4)
• batch_size : batch size (default=128)

• train

python main.py

• test

python main.py --phase=test