基于残差密集融合对抗生成网络的PET-MRI图像融合
- 分享到:
摘要：
为了增强核磁共振与正电子发射断层扫描图像融合的纹理细节,摆脱人工设计融合规则对先验知识的依赖.提出了自适应的残差密集生成对抗网络(adaptive dense residual generative adversarial network,ADRGAN)来融合两种模态的医学图像.ADRGAN设计了区域残差学习模块与输出级联生成器,在加深网络结构的同时避免特征丢失;然后,设计了基于自适应模块的内容损失函数,强化输出融合图像的内容信息;最后,通过源图像的联合梯度图与融合图像的梯度图构建对抗性博弈来高效训练生成器与鉴别器.实验结果表明,ADRGAN在哈佛医学院MRI/PET数据集的测试中峰值信噪比和结构相似度分别达到55.2124和0.4697,均优于目前最先进的算法;所构建的模型具有端对端和无监督两特性,无需人工干预,也不需要真实数据作为标签.
In this paper,in order to enhance the texture details of fusion of magnetic resonance imaging(MRI)and positron emission computed tomography imaging(PET)images and get rid of the dependence of artificial design fusion rules on a priori knowledge.An adaptive dense residual generative adversarial network(ADRGAN)is proposed to fuse two modes of medical images.ADRGAN designs a regional residual learning module and output cascade generator to deepen the network structure and avoid feature loss.Then,a content loss function based on adaptive module is designed to enhance the content information of the output fused image.Meanwhile,an adversarial game is constructed through the joint gradient map of the source image and the gradient map of the fused image,and an efficient training generator and discriminator are constructed.The experimental results show that the peak signal-to-noise ratio and structural similarity of ADRGAN proposed in this paper are 55.2124 and 0.4697 respectively,which are better than the most advanced algorithms.Moreover,the model in this paper has two characteristics:end-to-end and unsupervised,without manual intervention and real data as labels.
作者：
刘尚旺 杨荔涵
Liu Shangwang;Yang Lihan(College of Computer and Information Engineering,Engineering Lab of Intelligence Business and Internet of Things,Henan Normal University,Xinxiang 453007,China)
机构地区：
河南师范大学计算机与信息工程学院
出处：
《河南师范大学学报(自然科学版)》 CAS 2024年第1期74-83,I0005,共11页
Journal of Henan Normal University(Natural Science Edition)
基金：
国家自然科学基金(U1304607) 河南省高等学校重点科研项目基础研究计划(21A520022)。
关键词：
深度学习 对抗生成网络 多模态图像融合 密集残差网络
deep learning generative adversarial network multi-model image fusion dense residual network
分类号：
TP391 [自动化与计算机技术—计算机应用技术]
