摘 要：为了对新冠肺炎 CT 图像进行精确分割，在 Unet 模型基础上，设计了一种基于主成分分析的改进 Unet 模型。在提出的方法中，先用 PCA 对新冠肺炎 CT 图像进行特征预提取，以去除图像中的噪声等因素干扰，获得更本质的特征。然后将产生的特征图像作为原始图像输入 Unet 中进行分割。同时，在网络层中进行 BN 处理。这个联合了 PCA 和 Unet 的改进模型记为 PCA-Unet。实验表明，相比于单纯使用 Unet，PCA-Unet 有更好的表现。在视觉印象上，PCA-Unet 能够更准确地分割出感兴趣区域，更接近于金标准。在定量比较上，PCA-Unet 在四项指标上都获得了优势。
关键词：新冠 CT 图像；主成分分析；Unet；BN
DOI:10.19850/j.cnki.2096-4706.2022.20.023
基金项目：2020 年国家级大学生创新创业训练计划项目资助（S202010927051）
中图分类号：TP18 文献标识码：A 文章编号：2096-4706（2022）20-0094-04
The Segmentation of CT Images of COVID-19 Using Unet Network Based on PCA
YU Houqiang, XU Yifan, XU Jinglei, CHEN Yao, TANG Xiaoli
(School of Mathematics and Statistics, Hubei University of Science and Technology, Xianning 437100, China)
Abstract: In order to segment CT images of COVID-19 accurately, an improved Unet model based on principal component analysis (PCA) is designed on the baseline of Unet model. In the proposed method, PCA is firstly used to extract the features of CT images of COVID-19, which can remove the interference of noise and other factors in the images and obtain more essential features. Then, as the original images, the generated feature images are input into Unet for image segmentation. Meanwhile, BN processing is carried out in the network layer. This improved model combining PCA and UNET is called PCA-Unet. The experimental results show that PCA-Unet has better performance than using Unet alone. In terms of visual impression, PCA-Unet can segment the interesting region more accurately, which is closer to the gold standard. In quantitative comparison, PCA-Unet has gained advantages in four indicators.
Keywords: CT images of COVID-19; principal component analysis; Unet; BN
参考文献：
[1] CHEN N，ZHOU M，DONG X，et al. Epidemiological and clinical characteristics of 99 cases of 2019 novel coronavirus pneumonia in Wuhan，China：a descriptive study [J].Lancet 2020：395（10223）：507–513.
[2] CHUNG M，BERNHEIM A，MEI X，et al.CT imaging features of 2019 novel coronavirus （2019-ncov） [J].Radiology， 2020，295（1）：202-207.
[3] 赵丽亚 . 肿瘤手术导航中图像分割与配准方法研究 [D].北京：北京工业大学，2016.
[4] SUMMERS R M.Deep learning lends a hand to pediatric radiology [J].Radiology，2018，287（1）：323-325.
[5] NOMURA Y，HIGAKI T，FUJITA M，et al. Effects of iterative reconstruction algorithms on computer-assisted detection （CAD） software for lung nodules in ultra-low-dose CT for lung cancer screening [J].Academic Radiology，2016，24（2）：124-130.
[6] RONNEBERGER O，FISCHER P，BROX T. U-Net：Convolutional networks for biomedical image segmentation [C]// MICCAI 2015：Medical Image Computing and Computer-Assisted Intervention. Cham：Springer，2015：234–241.
[7] HAN C T，JIA K，GAO S H，et al.Pcanet：a simple deep learning baseline for image classification [J].IEEE Trans Image Process，2015，24（12）：5017-32.
[8] IOFFE S，SZEGEDY C. Batch Normalization：Accelerating Deep Network Training by Reducing Internal Covariate Shift [J/OL].（2022-07-02）.https：//arxiv.org/abs/1502.03167.
作者简介：余后强（1979—），男，汉族，湖北咸宁人，副教授，博士，主要研究方向：应用数学。