Implementation of Low-Dose CT with a Residual Encoder-Decoder Convolutional Neural Network (RED-CNN)
There is several things different from the original paper.
- The input image patch(55x55 size) is extracted randomly from the 512x512 size image. --> Original : Extract patches at regular intervals from the entire image.
- Masking the Hounsfield unit corresponding to the air and calculating the MSE loss. (The original MSE loss also proceeds)
- 512x512 entire image input is used without extracting the patches.
- “the 2016 NIH-AAPM-Mayo Clinic Low Dose CT Grand Challenge” by Mayo Clinic
https://www.aapm.org/GrandChallenge/LowDoseCT/