改进EPGAN模型的COVID-19 CT图像生成方法  

COVID-19 image generation method of improved EPGAN model

作　　者：高志军 高雅 Gao Zhijun;Gao Ya(School of Computer&Information Engineering,Heilongjiang University of Science&Technology,Harbin 150022,China)

机构地区：[1]黑龙江科技大学计算机与信息工程学院,哈尔滨150022

出　　处：《黑龙江科技大学学报》2023年第5期767-773,共7页Journal of Heilongjiang University of Science And Technology

基　　金：黑龙江省省属高等学校基本科研业务费项目(2019-KYYWF-11,2022-KYYWF-0565)。

摘　　要：为解决COVID-19 CT图像分类任务中需要大量数据的问题,提出了一种基于EPGAN模型的新冠肺炎CT图像生成方法。采用向量到图像的生成模型生成COVID-19 CT图像,通过添加ECA注意力机制到生成器中上采样模块的方法,解决ProjectedGAN模型生成图像肺部轮廓不闭合问题。结果表明,该方法生成的COVID-19 CT图像,FID降低至16.75,PSNR提升至12.62,SSIM提升至0.320。该方法可以提升COVID-19 CT图像中肺部轮廓的质量,生成更高质量的COVID-19 CT图像。This paper proposes a new method for generating COVID-19 CT images based on improved EPGAN model to address the problem of large amount of data needed for COVID-19 image classification.The study involves using a vector-to-image(V2I)generation model to generate COVID-19 CT images;solving lung contour failure problem in ProjectedGAN model by adding ECA attention mechanism to the upsampling module in the generator.The results show that FID decreases to 16.75,PSNR increases to 12.62,and SSIM increases to 0.320 in COVID-19 CT images generated by the proposed method.This method can improve the quality of lung contour in COVID-19 CT images and generate higher-quality COVID-19 CT images.

关 键 词：图像生成 COVID-19 生成对抗网络 注意力机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]