摘要
目的使用卷积神经网络(convolutional neural network,CNN)从欠采样的磁共振成像K空间数据快速重建出无伪影的高质量图像。材料与方法实验数据包含60位自愿者矢状位、横断位、冠状位全采的T1加权脑部MR图像,使用旋转和拉伸等操作对训练数据进行扩增。不同欠采模式的MR图像和金标准图像分别输入CNN进行训练,学习获得的网络可实现由欠采图像到全采集图像之间的非线性映射。重建时,将CNN重建图像的K空间与原始的K空间数据进行合并保真。实验中利用金标准图像,计算重建图像的峰值信噪比(peak signal to noise ratio,PSNR)、结构相似度(structural similarity,SSIM)和高频误差范数(high frequency error norm,HFEN),定量评价重建结果。结果 (1)CNN重建出的中央采样MR图像的PSNR、SSIM、HFEN分别为31.13、0.93、223.81,优于Tukey窗函数的25.69、0.86、482.75;(2)CNN重建出的伪随机采样MR图像的PSNR、SSIM、HFEN分别为32.78、0.95、195.51,优于压缩感知的31.01、0.93、184.69。结论卷积神经网络可以从欠采数据重建出高质量的磁共振图像,无论是主观的视觉效果还是客观的评价参数都优于传统的处理方法。与K空间中央连续采集相比,伪随机采样模式更有利于CNN重建出高质量的MR图像。
Objective:To reconstruct high quality,artifacts-free magnetic resonance imaging(MRI)images from under-sampled k-space data with convolutional neural network(CNN).Materials and Methods:T1-weighted brain MR images of sagittal,transverse and coronal orientations from sixty volunteers are used.Rotation and stretching were used for data augmentation.CNN was trained with pairs of ground truth and under-sampled MR images to learn the nonlinear mapping between them.In the reconstruction,output of CNN was merged with the sampled k-space data to get the final image.For quantitative evaluation,we used peak signal-to-noise ratio(PSNR),structural similarity(SSIM)and high-frequency error norm(HFEN)to compare results of different methods.Results:(1)PSNR,SSIM,HFEN of center-sampled MR images reconstructed by CNN are 31.13,0.93,223.81,compared with 25.69,0.86,482.75 of Tukey filter.(2)PSNR,SSIM,HFEN of pseudo-random sampled MR images reconstructed by CNN are 32.78,0.95,195.51,compared with 31.01,0.93,184.69 of compressed sensing.Conclusions:CNN can reconstruct high quality MR images from under-sampled data and achieved better results both visually and statistically,compared with traditional methods.For CNN-based reconstruction,pseudo-random sampling is more favorable.
作者
王一达
宋阳
谢海滨
童睿
李建奇
杨光
WANG Yi-da;SONG Yang;XIE Hai-bin;TONG Rui;LI Jian-qi;YANG Guang(School of Physics&Materials Science,East China Normal University,Shanghai Key Laboratory of Magnetic Resonance,Shanghai 200062,China)
出处
《磁共振成像》
CAS
CSCD
2018年第6期453-459,共7页
Chinese Journal of Magnetic Resonance Imaging
基金
国家自然科学基金重点项目(编号:61731009)~~
关键词
磁共振成像
卷积神经网络
图像重建
欠采
压缩感知
Magnetic resonance imaging
Convolutional neural network
Image reconstruction
Under-sampling
Compressed sensing
