MRI super-resolution via realistic downsampling with adversarial learning.
MRI super-resolution via realistic downsampling with adversarial learning.
复制标题
DOI：
10.1088/1361-6560/ac232e
复制
发表时间：
2021-10-05
影响因子：
3.5
通讯作者：
Cai J
中科院分区：
文献类型：
--
作者：
Huang B;Xiao H;Liu W;Zhang Y;Wu H;Wang W;Yang Y;Yang Y;Miller GW;Li T;Cai J
许多深度学习（DL）框架在磁共振成像（MRI）的超分辨率（SR）任务中表现出了最先进的性能，但大多数性能都是通过模拟低分辨率（LR）图像而不是来自真实的采集的LR图像实现的。由于SR网络的有限的泛化能力，由于训练LR图像的不真实性，不能保证对真实的LR图像的增强。在这项研究中，我们提出了一个基于DL的SR框架，重点是数据建设，以实现更好的性能在真实的LR MR图像。该框架包括两个步骤：（a）使用生成对抗网络（GAN）进行下采样训练，以构建更真实和完美匹配的LR/高分辨率（HR）对。下采样GAN输入是真实的LR和HR图像。该生成器将HR图像转换为LR图像，并将合成图像和真实的LR图像之间的块级差异进行区分。(b)超分辨率训练使用增强的深度超分辨率网络（EDSR）进行。在对照实验中，使用我们提出的方法，高斯模糊和k空间零填充训练三个EDSR。至于数据，从24例患者中获得肝脏MR图像，使用屏气系列LR和HR扫描（仅在常规方法中使用HR图像）。k空间零填充组在真实的LR图像上几乎为零增强，而高斯组产生了相当数量的伪影。与其他两种网络相比，该方法具有更好的分辨率增强和更少的伪影。该方法在结构相似性指数（SSIM）和峰值信噪比（PSNR）上分别比高斯方法提高了0.111 ± 0.016和2.76 ± 0.98 dB。传统高斯方法的盲/无参考图像空间质量评价器（BRISQUE）度量为46.6 ± 4.2，新方法的BRISQUE度量为34.1 ± 2.4。
Many deep learning (DL) frameworks have demonstrated state-of-the-art performance in the super-resolution (SR) task of magnetic resonance imaging (MRI), but most performances have been achieved with simulated low-resolution (LR) images rather than LR images from real acquisition. Due to the limited generalizability of the SR network, enhancement is not guaranteed for real LR images because of the unreality of the training LR images. In this study, we proposed a DL-based SR framework with an emphasis on data construction to achieve better performance on real LR MR images. The framework comprised two steps: (a) downsampling training using a generative adversarial network (GAN) to construct more realistic and perfectly matched LR/high-resolution (HR) pairs. The downsampling GAN input was real LR and HR images. The generator translated the HR images to LR images and the discriminator distinguished the patch-level difference between the synthetic and real LR images. (b) Super-resolution training was performed using an enhanced deep super-resolution network (EDSR). In the controlled experiments, three EDSRs were trained using our proposed method, Gaussian blur, and k-space zero-filling. As for the data, liver MR images were obtained from 24 patients using breath-hold serial LR and HR scans (only HR images were used in the conventional methods). The k-space zero-filling group delivered almost zero enhancement on the real LR images and the Gaussian group produced a considerable number of artifacts. The proposed method exhibited significantly better resolution enhancement and fewer artifacts compared with the other two networks. Our method outperformed the Gaussian method by an improvement of 0.111 ± 0.016 in the structural similarity index (SSIM) and 2.76 ± 0.98 dB in the peak signal-to-noise ratio (PSNR). The blind/reference-less image spatial quality evaluator (BRISQUE) metric of the conventional Gaussian method and proposed method were 46.6 ± 4.2 and 34.1 ± 2.4, respectively.
登录
查看更多内容
影响因子：
3.8
作者：
Liu, Yilin;Yin, Fang-Fang;Cai, Jing
通讯作者：
Cai, Jing
影响因子：
3.3
作者：
Chaudhari AS;Fang Z;Kogan F;Wood J;Stevens KJ;Gibbons EK;Lee JH;Gold GE;Hargreaves BA
通讯作者：
Hargreaves BA
影响因子：
2.5
作者：
Luo, Jianhua;Mou, Zhiying;Zhu, Yuemin
通讯作者：
Zhu, Yuemin
通过具有固定跳跃连接的宽残差网络实现 MR 图像超分辨率
DOI：
10.1109/jbhi.2018.2843819
发表时间：
2019-05-01
影响因子：
7.7
作者：
Shi, Jun;Li, Zheng;Yan, Pingkun
通讯作者：
Yan, Pingkun
影响因子：
3.8
作者：
Chun, Jaehee;Zhang, Hao;Park, Justin C.
通讯作者：
Park, Justin C.