基于多尺度密集嵌套型网络的磁共振图像重建

Magnetic resonance images reconstruction based on multi-scale dense nested network

目的:研究一种从欠采样K空间数据中重建高质量磁共振图像的算法。方法:利用密集连接网络与多尺度思想设计一种网络模型来实现磁共振图像的高质量重建。首先以密集连接网络为基础框架,将不同空洞率的组合分别放置在密集连接单元;然后基于切片间的先验信息来建立相邻切片间特征信息的传输通道,并嵌套在密集连接网络当中;最后对K空间进行数据保真,并在网络迭代层加入密集连接机制。结果:对于12.5%和25%的K空间数据,重建的磁共振图像峰值信噪比(PSNR)分别为36.12 dB,40.45 dB。结论:在Calgary-Campinas数据集上的实验结果表明,与传统网络模型相比,所提模型重建精度更高、收敛更快。

Aims:This paper aims to study a high-quality magnetic resonance image reconstruction algorithm from undersampling K-space data.Methods:Utilizing the dense connection network and the multi-scale ideology,a network model was designed to reconstruct the high-quality magnetic resonance images.Firstly,based on the dense connection network,the combination with different void rates were placed on the dense connection units.Secondly,based on the prior information between slices,the transmission channel containing the feature information among adjacent slices was established,which was then nested in the dense network;Finally,the fidelity for K-space data and the dense connection mechanism in the network iteration layer were added.Results:For 12.5%and 25%K-space data,the PSNR of reconstructed magnetic resonance images were 36.12 dB and 40.45 dB respectively.Conclusions:The results on a Calgary-Campinas data set show that the proposed model can significantly improve the reconstruction accuracy and the convergence rate compared with the traditional network model.