Dense Recurrent Neural Networks for Accelerated MRI: History-Cognizant Unrolling of Optimization Algorithms.
Dense Recurrent Neural Networks for Accelerated MRI: History-Cognizant Unrolling of Optimization Algorithms.
复制标题
用于加速MRI的密集递归神经网络：优化算法的历史认知展开。
DOI：
10.1109/jstsp.2020.3003170
复制
发表时间：
2020-10
影响因子：
7.5
通讯作者：
Akçakaya M
中科院分区：
文献类型：
--
作者：
Hosseini SAH;Yaman B;Moeller S;Hong M;Akçakaya M
加速磁共振成像的逆问题通常在正则化重建框架中结合关于前向编码算子的域特定知识。最近，物理驱动的深度学习(DL)方法被提出使用神经网络进行数据驱动的正则化。这些方法通过神经网络在特定领域的数据一致性和数据驱动的正则化之间交替，展开迭代优化算法来求解反问题目标函数。然后，对整个展开的网络进行端到端的训练，以学习网络的参数。由于数据一致性更新使用梯度下降步骤的简单性，近端梯度下降(PGD)是展开物理驱动的DL重建方法的一种常见方法。然而，PGD方法具有较慢的收敛速度，需要更多的展开迭代，导致训练中的内存问题和测试中的重建时间较慢。受使用先前迭代历史的PGD方法的有效变体的启发，我们提出了一种具有历史认知的优化算法展开，该算法具有跨迭代的密集连接以提高性能。在我们的方法中，梯度下降步长是以所有先前正则化单元的输出的可训练组合来计算的。我们还将这一思想应用于二次松弛变量分裂方法的展开。我们在快速MRI膝关节数据集的重建结果表明，与传统的未滚动方法相比，所提出的历史认知方法减少了残余混叠伪影，而不需要额外的计算能力或增加重建时间。
Inverse problems for accelerated MRI typically incorporate domain-specific knowledge about the forward encoding operator in a regularized reconstruction framework. Recently physics-driven deep learning (DL) methods have been proposed to use neural networks for data-driven regularization. These methods unroll iterative optimization algorithms to solve the inverse problem objective function, by alternating between domain-specific data consistency and data-driven regularization via neural networks. The whole unrolled network is then trained end-to-end to learn the parameters of the network. Due to simplicity of data consistency updates with gradient descent steps, proximal gradient descent (PGD) is a common approach to unroll physics-driven DL reconstruction methods. However, PGD methods have slow convergence rates, necessitating a higher number of unrolled iterations, leading to memory issues in training and slower reconstruction times in testing. Inspired by efficient variants of PGD methods that use a history of the previous iterates, we propose a history-cognizant unrolling of the optimization algorithm with dense connections across iterations for improved performance. In our approach, the gradient descent steps are calculated at a trainable combination of the outputs of all the previous regularization units. We also apply this idea to unrolling variable splitting methods with quadratic relaxation. Our results in reconstruction of the fastMRI knee dataset show that the proposed history-cognizant approach reduces residual aliasing artifacts compared to its conventional unrolled counterpart without requiring extra computational power or increasing reconstruction time.
登录
查看更多内容
影响因子：
3.3
作者：
Hammernik K;Klatzer T;Kobler E;Recht MP;Sodickson DK;Pock T;Knoll F
通讯作者：
Knoll F
影响因子：
2.1
作者：
Becker, Stephen;Bobin, Jerome;Candes, Emmanuel J.
通讯作者：
Candes, Emmanuel J.
影响因子：
2.7
作者：
Hong, Mingyi;Wang, Xiangfeng;Luo, Zhi-Quan
通讯作者：
Luo, Zhi-Quan
影响因子：
3.7
作者：
Hosseini, Seyed Amir Hossein;Zhang, Chi;Akcakaya, Mehmet
通讯作者：
Akcakaya, Mehmet
影响因子：
5.4
作者：
Giryes, Raja;Eldar, Yonina C.;Sapiro, Guillermo
通讯作者：
Sapiro, Guillermo