search

SSinyu / RED-CNN

Pytorch Implementation of Low-Dose CT with a Residual Encoder-Decoder Convolutional Neural Network (RED-CNN)
211 stars 54 forks source link

Calculation of indicators PSNR, SSIM and RMSE is different from other papers. #19

Open xwt11 opened 1 year ago

xwt11 commented 1 year ago

Hello, I use the code of this project, and the calculated PSNR and SSIM are smaller than the results of other papers. The value of RMSE is large.

After denormalize_ function, the pixel value range of the image changes to [-1024,3072], but trunc function is used to directly change the pixel value greater than 240 to 240, and the pixel value less than -160 to -160. Why the change? Why is the pixel value negative?

def denormalize_(self, image):
    image = image * (self.norm_range_max - self.norm_range_min) + self.norm_range_min
    return image

def trunc(self, mat):
    mat[mat <= self.trunc_min] = self.trunc_min
    mat[mat >= self.trunc_max] = self.trunc_max
    return mat
Nioolek commented 1 year ago

我猜测，由于测试的患者L506的CT图是腹部图像，而腹部图像在CT中成像HU值都比较接近，一般都会使用-160~240这样的窗宽窗位，所以这里就按照这样来设置了