基于生成对抗网络的肺炎CT图像生成  

CT Image Generation of Pneumonia Based on Generative Adversarial Network

作　　者：王家晨 张鸿鑫 刘庆华[1,2] WANG Jia-chen;ZHANG Hong-xin;LIU Qing-hua(Ocean College,Jiangsu University of Science and Technology,Zhenjiang 212100,China;Computer College,Jiangsu University of Science and Technology,Zhenjiang 212100,China)

机构地区：[1]江苏科技大学海洋学院,江苏镇江212100 [2]江苏科技大学计算机学院,江苏镇江212100

出　　处：《计算机与现代化》2023年第7期20-24,共5页Computer and Modernization

基　　金：江苏省六大高峰人才项目(XYDXX-117);国家自然科学基金资助项目(51008143)。

摘　　要：针对以随机噪声为输入的生成对抗网络生成图像模糊、训练不易收敛,以及传统物理数据扩充方法无法融合数据特征的相似性与多样性的问题,提出一种基于特征金字塔和注意力机制CBAM的生成对抗网络——UG-DCGAN。首先以经过遮掩去噪的CT图像作为输入,增强网络的鲁棒性;然后利用特征融合金字塔与注意力机制联合建立生成网络提取和重构CT图像,其中特征融合金字塔仅保留最大尺度融合,并在下采样过程中加入残差结构以提高特征提取能力;最后增加判别器网络的卷积层,提高其监督判断能力。经过实验验证,与最新的StyleGAN2模型生成的CT图像相比,本文算法所生成的CT图像的IS值提高了12.44%,FID值降低了21.98%,表明本文算法对提升CT图像生成的清晰度、特征的相似性和多样性都有较明显的效果。Aiming at the problem of fuzzy image generated by generative adversarial networks with random noise as input,diffi⁃cult convergence in training,and the similarity and diversity of data features cannot be fused by traditional physical data expan⁃sion methods,this paper proposes a generative adversarial network-UG-DCGAN based on feature pyramid and attention mecha⁃nism CBAM.The method first takes the masked and denoised CT image as input to enhance the robustness of the network.Then use the feature fusion pyramid and attention mechanism to jointly establish a generative network to extract and reconstruct CT im⁃ages,in which the feature fusion pyramid only retains the maximum scale fusion,and the residual structure is added in the down-sampling process to improve the feature extraction ability.Finally,the convolution layer of the discriminator network is added to improve its supervised judgment ability.After experimental verification,the results compared with the StyleGAN2 model,the FID value of the CT image generated by the algorithm is reduced by 21.98%,and the IS value is increased by 12.44%.It shows that the algorithm has obvious effects on improving the clarity of CT image generation,the similarity and diversity of features.

关 键 词：深度学习 特征金字塔 GAN 图像生成 CT 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]