基于深度递归级联卷积神经网络的并行磁共振成像方法  

A Deep Recursive Cascaded Convolutional Network for Parallel MRI

作　　者：程慧涛 王珊珊[3] 柯子文 贾森 程静 丘志浪 郑海荣[3] 梁栋[1,2] CHENG Hui-tao;WANG Shan-shan;KE Zi-wen;JIA Sen;CHENG Jing;QIU Zhi-lang;ZHENG Hai-rong;LIANG Dong(Research Center for Medical AI(Shenzhen Institute of Advanced Technology,Chinese Academy of Sciences),Shenzhen 518055,China;University of Chinese Academy of Sciences,Beijing 100049,China;Paul C.Lauterbur Research Center for Biomedical Imaging(ShenzhenInstitute of Advanced Technology,Chinese Academy of Sciences),Shenzhen 518055,China)

机构地区：[1]医学人工智能研究中心(中国科学院深圳先进技术研究院),广东深圳518055 [2]中国科学院大学,北京100049 [3]保罗C.劳特伯生物医学成像研究中心(中国科学院深圳先进技术研究院),广东深圳518055

出　　处：《波谱学杂志》2019年第4期437-445,共9页Chinese Journal of Magnetic Resonance

基　　金：国家重点研发计划(2017YFC0108802)

摘　　要：快速磁共振成像是磁共振研究领域重要的课题之一.随着大数据和深度学习的兴起,神经网络成为快速磁共振技术的重要方法.然而网络性能表现和网络参数量之间较难取得平衡,且对于多通道数据重建的并行成像问题,相关研究较少.本文构建了一种深度递归级联卷积神经网络结构,用于处理并行成像问题.这种网络结构在减少网络参数量的同时,能够尽可能地提高网络的表达能力,提高网络重建的精确度.实验结果表明,相较于传统并行成像方法,通过训练好的神经网络对欠采样磁共振数据进行重建,可以得到更准确的重建结果,且重建时间大大缩短.Fast magnetic resonance imaging(MRI)has been attracting more and more research interests in recent years.With the emergence of big data and development of advanced deep learning algorithms,neural network has become a common and effective tool for image reconstruction in fast MRI.One main challenge to the deep learning-based methods for fast MRI reconstruction is the trade-off between the network performance and the network capacity.Few previous studies have used the deep learning-based methods in parallel imaging.In this work,a deep recursive cascaded convolutional network(DRCCN)architecture was designed for parallel MRI,with reduced number of network parameters while maintaining a satisfactory performance.The experimental results demonstrated that,compared to the classical methods,image reconstruction with the well-trained DRCCN networks were more accurate and less time consuming.

关 键 词：快速磁共振成像 并行成像 深度学习 卷积神经网络 先验信息 

分 类 号：O482.53[理学—固体物理]