Compressed Sensing: Novel Applications
Click on to view the abstract pdf. Click on to view the recorded presentation.
Monday 7 May 2012
Plenary Hall  10:45 - 12:45 Moderators: Craig H. Meyer, Krishna S. Nayak

10:45 0006.   Compressed Sensing Multi-Spectral Imaging of the Post-Operative Spine
Pauline Wong Worters1, Kyunghyun Sung1, Kathryn J Stevens1, Kevin M Koch2, and Brian A Hargreaves1
1Stanford University, Stanford, CA, United States, 2ASL, GE Healthcare, Waukesha, WI, United States

Multi-spectral imaging (MSI) methods such as MAVRIC, SEMAC and Hybrid have been developed in recent years to provide distortion-free MRI of tissue around metallic implants. However, acquisition times remain lengthy (5-15 minutes) and limit the achievable spatial resolution in routine clinical use. In this work, we demonstrate the feasibility of using compressed sensing (CS) to reduce acquisition time in a retrospective application to patient data with spinal hardware. Results show that retrospective CS-MSI are the same as or better than the original MSI images. We also show that fully sampled MSI and prospectively undersampled T2-weighted CS-MSI (42% scan time reduction) are comparable in terms of image contrast and quality.

10:57 0007.   A combined approach to Compressed Sensing and Parallel Imaging for Fat-Water Separation with R2* estimation
Curtis N Wiens1, Colin M McCurdy2,3, and Charles A McKenzie1,3
1Department of Physics and Astronomy, University of Western Ontario, London, Ontario, Canada, 2Department of Physics, University of Guelph, Guelph, Ontario, Canada, 3Department of Medical Biophysics, University of Western Ontario, London, Ontario, Canada

R2*-corrected chemical shift based fat-water separation techniques have been used to accurately quantify fat. These are time-consuming and require image acceleration techniques. The following work describes a method of separating fat and water from undersampled multi-channel datasets. This method simultaneously applies compressed sensing, parallel imaging, and fat-water separation. To illustrate this technique, net acceleration factors of up to 4 are shown in the abdomen. Including the R2* term improves the accuracy of the fat-water separation. The proposed method improves image quality over sequential parallel imaging and fat-water separation at high acceleration factors.

11:09 0008.   Accelerated Echo-Planar Correlated Spectroscopic Imaging in the Human Calf Muscle using Compressed Sensing permission withheld
Jon Furuyama1, Chris Roberts2, and M. Albert Thomas1
1Radiology, UCLA, Los Angeles, CA, United States, 2School of Nursing, UCLA, Los Angeles, CA, United States

Recently, a four-dimensional Echo-Planar Correlated Spectroscopic Imaging (EP-COSI) sequence was shown to produce spatially localized two-dimensional spectra. Despite the speed improvements of the echo-planar readout, the sequence still requires significant scan time to collect all four dimensions. We show that the use of Compressed Sensing (CS) techniques can reconstruct under-sampled datasets with as little as 33% of the original data. Such a reduction in scan time allows for the deployment of 4D spectroscopic imaging sequences in a clinically feasible time frame. Examples of CS reconstruction are shown in the study of diabetes in the human calf muscle.

11:21 0009.   Direct Diffusion Tensor Estimation Using Joint Sparsity Constraint Without Image Reconstruction
Yanjie Zhu1,2, Yin Wu1,2, Ed X. Wu3,4, Leslie Ying5, and Dong Liang1,2
1Paul C. Lauterbur Research Centre for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Shenzhen, Guangdong, China,2Key Laboratory of Health Informatics, Chinese Academy of Sciences, Shenzhen, China, 3Laboratory of Biomedical Imaging and Signal Processing, 4Department of Electrical and Electronic Engineering, The University of Hong Kong, Pokfulam, Hong Kong, 5Department of Electrical Engineering and Computer Science, University of Wisconsin-Milwaukee, WI, Milwaukee, United States

The joint sparsity constraint is integrated into the model-based method to improve the accuracy of direct diffusion tensor estimation from highly undersampled k-space data. The method, named model-based method with joint sparsity constraint (MB-JSC), effectively incorporates the prior information on the joint sparsity of different diffusion weighted images in solving the nonlinear equation of tensors. Experimental results demonstrate that the proposed method is able to estimate the diffusion tensors more accurately than the existing method when a high net reduction factor is used.

11:33 0010.   Highly accelerated dynamic contrast enhanced imaging with prospective undersampling
R. Marc Lebel1, Jesse Jones2, Jean-Christophe Ferré2, Samuel Valencerina2, Krishna S. Nayak1, and Meng Law2
1Department of Electrical Engineering, University of Southern California, Los Angeles, CA, United States, 2Department of Radiology, University of Southern California, Los Angeles, CA, United States

Dynamic contrast enhanced (DCE) imaging requires high spatial and temporal resolution and large volume coverage; traditionally, these are mutually exclusive. We present prospective undersampled DCE imaging of brain tumors with high spatiotemporal resolution using l1-SPIRiT (compressed sensing and parallel imaging). We employ multiple spatial and temporal reconstruction constraints, including a novel temporal constraint that promotes low frequency signal changes, to achieve high accelerations without compromising resolution or dynamic information.

11:45 0011.   Joint Reconstruction of Under-Sampled Multiple Contrast Images Using Mutual Information
Eric Wong1
1Radiology/Psychiatry, UC San Diego, La Jolla, CA, United States

In clinical MRI, images are often acquired of the same anatomy with multiple forms of contrast. Mutual information has long been used as a criterion for aligning images with different contrast, as it is high for any pair of aligned images. We explore here the maximization of mutual information as a criterion in the joint reconstruction of pairs of under-sampled images with different contrasts. For T1 and T2 weighted images that were under-sampled by a factor of 1.9, maximizing mutual information resulted in excellent reconstructions in less than one second of reconstruction time.

11:57 0012.   High-frame-rate Multislice Speech Imaging with Sparse Samping of (k,t)-space
Maojing Fu1,2, Anthony G. Christodoulou1,2, Andrew T. Naber3, David P. Kuehn4, Zhi-Pei Liang1,2, and Bradley P. Sutton2,3
1Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States, 2Beckman Institute for Advanced Science and Technology, Urbana, IL, United States, 3Department of Bioengineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States,4Department of Speech and Hearing Science, University of Illinois at Urbana-Champaign, Urbana, IL, United States

Dynamic MRI can provide a valuable tool to quantitatively assess changes in oropharyngeal dynamics during speech. In this work we present 5-slice dynamic speech imaging at a frame rate of 20 fps with 2.2 mm × 2.2 mm × 8.0 mm spatial resolution. It was successfully performed by incorporating parallel imaging methods, a composite spiral / Cartesian sampling strategy, and a reconstruction scheme exploiting the partial separability and the spatial-spectral sparsity of the speech image sequence. Changing tongue shape is observed over a speech sample in volunteer subjects.

12:09 0013.   An Application of Regularization by Model Consistency Condition to Accelerated Contrast-Enhanced Angiography
Julia V Velikina1, and Alexey A Samsonov2
1Medical Physics, University of Wisconsin - Madison, Madison, Wisconsin, United States, 2Radiology, University of Wisconsin - Madison

A novel regularization by a model consistency condition is adapted for application in time-resolved contrast-enhanced intracranial angiography. The temporal behavior model is learned from low resolution training data by principal component analysis. The proposed method is shown to distinguish different filling patterns of healthy and pathological vasculature.

12:21 0014.   Continuous table movement MRI in a single breath-hold: Highly undersampled radial acquisitions with nonlinear iterative reconstruction and joint coil estimation permission withheld
Michael O. Zenge1, Martin Uecker2,3, Gerald Mattauch1, and Jens Frahm2
1Healthcare Sector, Siemens AG, Erlangen, Germany, 2Biomedizinische NMR Forschungs GmbH am Max-Planck-Institut für biophysikalische Chemie, Göttingen, Germany, 3Electrical Engineering and Computer Science, University of California, Berkeley, United States

Continuous table movement MRI is an emerging technique for a variety of clinical applications. The achievable acceleration with parallel imaging, however, is not yet sufficient to scan an extended FOV in a single breath-hold. Radial scanning in combination with innovative iterative image reconstruction and joint coil estimation promises significantly higher acceleration factors. This method was implemented for moving table abdominal MRI in a single breath-hold and was evaluated in 5 healthy volunteers. It was proven that highly undersampled radial images can be reconstructed with very little streaking artifacts which justifies further investigation in volunteers and patients.

12:33 0015.   T1 Map Reconstruction from Under-sampled KSpace Data using a Similarity Constraint
Mohammad H Kayvanrad1, A. Jonathan McLeod1, John S. H. Baxter1, Charles A McKenzie1, and Terry M Peters1
1Robarts Research Institute, The University of Western Ontario, London, Ontario, Canada

The similarity between images, in problems involving multiple acquisitions with different imaging parameters, is used as an additional reconstruction constraint beside sparity to further increase the quality of reconstruction/k-space under-sampling. This is of particular interest in reconstruction of T1/T2 maps. From a clinical perspective, this means a reduction in acquisition time.