search

hmshan / CPCE-3D

Low-dose CT via Transfer Learning from a 2D Trained Network, In IEEE TMI 2018
50 stars 16 forks source link

data format, #3

Open chocosando opened 5 years ago

chocosando commented 5 years ago

Hi. Thank you for nice talk!

How can I use this code for my image ?

My images consist of : pair of routine radiation dose CT [ 512 x 512] and low dose CT [ 512 x 512 ]

How can I make N x D x W x H and N x W x H ? What does 'label' means for paired images ?

Thank you!!

hmshan commented 5 years ago

Hi there,

The label here means the routine radiation dose CT image. The code here has a 2D version and 3D version. Since the training process is usually based on image patches, so the W and H are 64 in the paper.

  1. 2D version. The training data and label can both be N x 64 x 64, where N is the number of image patches.
  2. 3D version. The 3D version requires the adjacent slices and 3D convolution. So the input needs to be N x D x 64 x 64, and the label corresponds to the central slice (D//2). For example, if D=3, every time you need to input three adjacent slices and the output of the network corresponds to the central slice. Put differently, the two adjacent slices of the central one will help to denoise the central slice.

image