Research on low-dose CT image denoising method based on improved Corediff model
10.19745/j.1003-8868.2025080
- VernacularTitle:基于改进Corediff模型的低剂量CT图像去噪方法研究
- Author:
Li-mei SONG
1
;
Hang WU
;
Yi-feng HUANG
;
Qiang WANG
;
Guan-jun LIU
;
Feng CHEN
;
Ming YU
;
Jian-kun SHEN
Author Information
1. 军事科学院系统工程研究院,天津 300161
- Publication Type:Journal Article
- Keywords:
low-dose CT;
image denoising;
diffusion model;
Corediff model;
U-Net network
- From:
Chinese Medical Equipment Journal
2025;46(5):9-13
- CountryChina
- Language:Chinese
-
Abstract:
Objective To propose a low-dose CT image denoising method based on an improved Corediff model to recover the detailed features of the image and enhance the image quality.Methods An RS-Corediff model was established by modifying the key component U-Net network of the Corediff model.Firstly,the residual module was introduced in the network input stage for feature extraction;secondly,a new downsampling module was designed in the U-Net network encoder,which learned the semantic information of the feature map by convolution and maintained the learning state during the downsampling process so as to fully extract the image features;thirdly,the feature splicing processing was used to further enhance the learning effect during the upsampling process of the U-Net network decoder;finally,the convolutional kernel size was modified to adjust the sensory field during the convolutional process of the whole U-Net network structure so as to obtain rich features.The RS-Corediff model was compared with the residual encoder-decoder convolutional neural network(RED-CNN)model and the Corediff model on the public dataset AAPM 2016 in order to verify its effectiveness for low-dose CT image denoising.Results The RS-Corediff model gained advantages over the RED-CNN and Corediff models with a peak signal-to-noise ratio(PSNR)of 41.269 8,structural similarity(SSIM)of 0.953 4 and root mean square error(RMSE)of 17.568 7.Conclusion The proposed method effectively preserves the texture and details of low-dose CT images during the denoising process to improve the overall quality of the images.[Chinese Medical Equipment Journal,2025,46(5):9-13]
