Reconstruction of under-sampled magnetic resonance image based on convolutional neural network
Yi-da WANG,Yang SONG,Hai-bin XIE,Rui TONG,Jian-qi LI,Guang YANG
DOI: https://doi.org/10.12015/issn.1674-8034.2018.06.010
2018-01-01
Abstract: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.