Meeting Banner
Abstract #2264

Undersampled MRI Reconstruction with Trained Directions from a Guide Image

Xiaobo Qu1, Di Guo2, Bende Ning1, Yingkun Hou3, Shuhui Cai1, Zhong Chen1

1Department of Electronic Science, Fujian Key Laboratory of Plasma and Magnetic Resonance, Xiamen University, Xiamen, Fujian, China; 2Department of Communication Engineering, Xiamen University, Xiamen, Fujian, China; 3School of Information Science and Technology, Taishan University, Taian, Shandong, China

Undersampling the k-space data can speed up magnetic resonance imaging (MRI) at the cost of introducing the aliasing artifacts. In this paper, a patch-based directional wavelets(PBDW) is proposed to sparsify the magnetic resonance (MR) image in undersampled MRI reconstruction. First, a guide image is reconstructed from incomplete k-space data with conventional compressed sensing MRI method. Then, a parameter of PBDW, indicating the geometric direction of each image patch, is trained from the guide image and incorporated into the sparsifying transform to provide the sparse representation for the image to be reconstructed. Simulations demonstrate that trained PBDW leads to better edges than the convetional sparse MRI reconstruction methods do.

Abstract PDF Poster PDF

Keywords

accurate achieves adequate adopted aliasing always amounts approximation arranged artifacts assuming available basis better blocky brain building candidate china coefficients communication compressed constructed cost council decides defined denotes dictionary directional discrete divided doctoral domain edges education electronic enforces enforcing engineering error estimating extracted fidelity flowchart forward fully fund geometric grant graphics ground guide healthy incomplete incorporated indicate initial introduced introducing invariant laboratory lack larger largest leads limited lowest measure measured medicine minimize mitigate norm obviously orthogonal oversea paper parallel partially patch patches pattern people pixels plasma postgraduates preserve preserved produces program promotes proposed quarter reconstructed reconstruction reduced regularization represent representation republic respect respectively sampled sampling scanner scholarship school science selected sensing sharper simulations since slice solving space sparse sparsity specify speed spin stands supported technology term train trained training transactions transform trio truth turbo universities variation volunteer wavelet wavelets