Statistical Iterative CBCT Reconstruction Based on Neural Network

This is an implementation of our article on matlab, matconvnet.

The repository includes:

• data：this folder includes CS phantom projection with $5e^3$ incident photon and original CS phantom without noise.

• cpp：this folder includes 3D-Projection process：based on 3D Forward and Back-Projection for X-Ray CT Using Separable Footprints（compiled files --- updateu_method_cs.mexa64 for linux and updateu_method_cs.mexw64 for windows）

• models：this folder includes trained model

• eval：this folder includes PSNR, SSIM, ISNR functions（our paper use this version evaluation）

• utilities：this folder includes extra functions（e.g. load trained model function）

• Cal_Deblur.m：deblur functions（wrap the denoise phase）

• iter_cs.m：demo of CS Phantom

Preparation

Prerequisites

• matlab（tested on matlab16b(linux)）
• matconvnet：please reference the official installation instructions to install it

Get Starts

step1：clone the code

git clone https://github.com/HUST-Shan/Deblur-CBCT.git
cd Deblur-CBCT

step2：run CS phantom demo

iter_cs

（Note: 1. You can find the results in result folder. 2. the demo do deblur in middle layer, you can apply it in all the layer as well.）

Some Explainations

1. This pre-trained model is coming from IRCNN：the denoise network is a little different from our article, however, you can replace the denoise network with "any denoise network or methods"（we will release a trainable model in our python version）
2. We will upload more simulation digital phantoms soon.
3. We will release a python version soon（Containing training phase.）

Citation

@article{Deblur-CBCT
    Author = {Binbin Chen, Kai Xiang, Zaiwen Gong, Jing Wang*, Shan Tan*},
    Title = {Statistical Iterative CBCT Reconstruction Based on Neural Network},
    Journal = {Transactions on Medical Imaging},
    Year = {2017}
}