[1]黄锦威,萧文鹏,朱思婷,等.基于对抗训练的U-Net神经网络在稀疏投影CT图像增强的应用[J].中国医学物理学杂志,2020,37(5):612-618.[doi:10.3969/j.issn.1005-202X.2020.05.016]
 HUANG Jinwei,XIAO Wenpeng,ZHU Siting,et al.Application of adversarial training-based U-Net neural network in the enhancement of CTimages obtained by sparse projection[J].Chinese Journal of Medical Physics,2020,37(5):612-618.[doi:10.3969/j.issn.1005-202X.2020.05.016]
点击复制

基于对抗训练的U-Net神经网络在稀疏投影CT图像增强的应用()
分享到:

《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]

卷:
37
期数:
2020年第5期
页码:
612-618
栏目:
医学影像物理
出版日期:
2020-05-25

文章信息/Info

Title:
Application of adversarial training-based U-Net neural network in the enhancement of CT images obtained by sparse projection
文章编号:
1005-202X(2020)05-0612-07
作者:
黄锦威1萧文鹏2朱思婷3丘皓怡1陈星宇2刘深泉1
1. 华南理工大学数学学院,广东广州510640;2. 华南理工大学计算机科学与工程学院,广东广州510006;3. 华南理工大学电力学院,广东广州510641
Author(s):
HUANG Jinwei1 XIAO Wenpeng2 ZHU Siting3 QIU Haoyi1 CHEN Xingyu2 LIU Shenquan1
1. School of Mathematics, South China University of Technology, Guangzhou 5106401, China 2. School of Computer Science and Engineering, South China University of Technology, Guangzhou 510006, China 3. School of Electric Power Engineering, South China University of Technology, Guangzhou 510641, China
关键词:
CT图像增强U-Net神经网络对抗训练稀疏投影
Keywords:
CT image enhancement U-Net neural network adversarial training sparse projection
分类号:
TP391;R318
DOI:
10.3969/j.issn.1005-202X.2020.05.016
文献标志码:
A
摘要:
目的:针对稀疏投影的CT重建图像附带噪声和伪影的特性,使用神经网络模型对稀疏投影得到的低质量CT重建 图像进行图像增强。方法:在残差编码-解码卷积神经网络基础上提出一种基于对抗训练的U-Net神经网络模型,并使用 公开数据集TCGA-CESC癌症CT影像进行模型训练和测试。评价模型处理效果的指标包括峰值信噪比(PSNR)、结构相 似性(SSIM)和均方根误差(RMSE)。结果:在对180 次探测的CT重建图像的测试中,模型处理后的图像相比未处理图 像,PSNR、SSIM和RMSE指标平均值分别提升15.10%、37.89%和38.20%。在PSNR和SSIM指标平均值意义下,模型处 理后的图像优于1 800次探测的未处理CT重建图像。结论:本研究提出的神经网络模型能够减少伪影和噪点,对稀疏投 影CT图像增强有一定效果。
Abstract:
Objective For the existence of noise and artifacts in reconstructed CT images obtained by sparse projection, a neural network model is proposed for the enhancement of low-quality CT images obtained by sparse projection. Methods Based on residual encoder-decoder convolutional neural network, an adversarial training-based U-Net neural network was proposed. Model training and testing were carried out using the CT images of cancer from TCGA-CESC dataset. The main indicators for the evaluation of the model processing results included peak signal-to-noise ratio (PSNR), structural similarity (SSIM) and root-mean-square error (RMSE). Results In the test on the reconstructed CT images obtained by 180 detections, the average values of the PSNR, SSIM and RMSE of the images processed by the proposed model were increased by 15.10%, 37.89% and 38.20%, respectively, compared with those of the unprocessed images. Moreover, the average values of PSNR and SSIM indicated that the processed images were superior to the unprocessed images obtained by 1 800 detections. Conclusion The proposed neural network model can reduce artifacts and noise, and has certain effects on the enhancement of CT images obtained by sparse projection.

备注/Memo

备注/Memo:
【收稿日期】2019-12-09 【基金项目】“攀登计划”广东大学生科技创新培育专项资金(pdjh2018-b0043) 【作者简介】黄锦威,研究方向:数学物理,E-mail: 1191572861@qq.com 【通信作者】刘深泉,教授,E-mail: mashqliu@scut.edu.cn
更新日期/Last Update: 2020-06-03