面向磁共振影像超分辨的WGAN方法研究  

WGAN for super-resolution megnatic resonance imaging

作　　者：黎玥嵘 武仲科[1] 王学松[1] 申佳丽[1] 王醒策[1] LI Yuerong;WU Zhongke;WANG Xuesong;SHEN Jiali;WANG Xingce(School of Artificial Intelligence,Beijing Key Laboratory of Digital Preservation and Virtual Reality for Cultural Heritage,Engineering Research Center of Virtual Reality and Applications,Ministry of Education:Beijing Normal University,100875,Beijing,China)

机构地区：[1]北京师范大学人工智能学院,文化遗产数字化保护与虚拟现实北京市重点实验室,虚拟现实应用教育部工程研究中心,北京100875

出　　处：《北京师范大学学报（自然科学版）》2021年第6期896-904,共9页Journal of Beijing Normal University(Natural Science)

基　　金：国家自然科学基金资助项目(61972041,62072045)。

摘　　要：针对磁共振成像(magnetic resonance imaging,MRI)超分辨率重构任务,提出了Wasserstein生成式对抗网络(Wasserstein generative adversarial network,WGAN),构建了合适的网络模型与损失函数;基于残差U-net WGAN后端上采样超分模型,设计了感知、纹理和对抗损失,用于恢复低分辨率MRI影像中的细节信息.此网络在2D-MRI的3000张脑影像上获得的峰值信噪比(peak signal to noise ratio,PSNR)是33.09 dB,结构相似度(structural similarity index measure,SSIM)的平均值为0.95;PSNR与SSIM的值与经典超分法相比较,分别增加了4.09 dB和0.06.这表明:网络能更好地学习MRI从低分辨率到高分辨率影像之间的映射关系;该方法有效稳定,可以广泛应用于相似系统.Magnetic resonance imaging(MRI),a tomography method,is widely used in clinical medicine for diagnosis of disease state.However,raw images often need to be further enhanced for better resolution.For superresolution reconstruction,we propose a residual U-net WGAN back-sampling and super-sampling model,with perceptual,texture and adversarial loss function.A comparative experiment was conducted on 3000 MRI images,resulting in averaged PSNR of 33.09 and SSIM of 0.95,improving PSNR by 4.09 and SSIM by 0.06,proving the learning validity of low to high resolution images.This algorithm is stable,robust and could be widely used.

关 键 词：生成式对抗网络 超分辨率重构 磁共振成像 深度学习 

分 类 号：TP231.2[自动化与计算机技术—检测技术与自动化装置]