Meeting Banner
Abstract #4232

Improved Compressed Sensing Reconstruction with Overcomplete Wavelet Transforms

Alicia W. Yang1, 2, Li Feng1, 2, Jian Xu3, 4, Ivan Selesnick4, Daniel K. Sodickson1, Ricardo Otazo1

1Department of Radiology, New York University School of Medicine, New York, NY, United States; 2Sackler Institute for Biomedical Sciences, New York University, New York, NY, United States; 3Siemens Medical Solution USA Inc, NY, United States; 4Polytechnic Institute of New York University, NY, United States

An adaptive decreasing thresholding method is developed to take into consideration the structure of overcomplete transforms by (1) using local thresholds that adapt to the signal power in each band and (2) decreasing the threshold for each step of the iterative reconstruction algorithm. The performance of this method in dual-tree wavelet transforms and curvelets was tested on compressed sensing reconstructions of retrospectively undersampled 3D coronary MRA and brain image datasets, and compared to that of standard Haar wavelet transforms.

Keywords

ability adapt adapting adaptive additionally advantage aforementioned aims aliasing applied array arrays arrows artery artifacts band bands better biomedical birdcage brain broad capture cardiac channel coefficients coil combined complexity compressed computational conjugate constant controls coronary curvatures curves dataset datasets dealing decomposition decreased decreasing density described dimension directionality disadvantages distinguishable domain dual early efficient enforced equipped exploited exploits fast features fold frequency full fully generated globally head healthy illustrates implementation improve improved improvement included indicated instead institute iteration iterations iterative joint lacks local mask matrix medical minimal multiplying operator optimize orthogonal partition patient performance power preserving produces propose radial radiology random receive reconstruct reconstructed reconstruction reconstructions regularization relatively representative represented resolution retrospective retrospectively sampled scanned scanner schemes school selecting sensing several sharp sharper simulation slice soft solution solve space sparse sparsity spatial specifically spokes step take threshold thresholding thresholds together trajectory transform transforms transverse tree twelve typically variable volunteer wavelet yang yielded