Seong-Eun Kim¹, John A. Roberts¹, Richard Wiggins¹, Bradley D. Bolster², and Dennis L. Parker^1
1UCAIR, Department of Radiology, University of Utah, Salt Lake City, Utah, United States, ²Siemens Healthcare, Salt Lake City, Utah, United States

In this work, we present a shim cycling technique to eliminate the banding artifacts in 3D bSSFP inner ear images in which multiple image sets are acquired with different shim currents during the acquisition of each set to change the local field compensation. These preliminary results give a strong indication that shim adjustments can sufficiently shift the banding artifact in 0.4 isotropic resolution 3D TrueFISP images, allowing very uniform composite images to be obtained. It would appear that banding artifacts in higher resolution images, with even longer TR, could be corrected. This work will be important for very high-resolution 3D TrueFISP imaging of the inner ear.

Kevin Koch¹, Eric Printz¹, and Dan Xu^1
1GE Healthcare, Milwaukee, Wisconsin, United States

It has previously been demonstrated that EPI acquisitions with reversed phase-encode blip polarities can be used to correct image distortions. A by-product of these methods is a shift map that maps the two source images together. Here, we demonstrate that one of the algorithms previously utilized for distortion correction purposes can be modified to produce high-SNR volumetric field maps for shimming purposes. These results suggest that such methods could be used to collect spatially accurate volumetric field maps for shimming purposes in a matter of seconds.

Signe Johanna Vannesjo¹, Bertram J. Wilm¹, Yolanda Duerst¹, Benjamin E. Dietrich¹, David Otto Brunner¹, Christoph Barmet^1,2, Thomas Schmid¹, and Klaas P. Pruessmann^1
1Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland, ²Skope Magnetic Resonance Technologies, Zurich, Switzerland

Breathing- and motion-related field fluctuations can affect brain imaging at 7T. The resulting artifacts can be reduced by concurrent field monitoring. However, gradient dephasing and signal decay of the field probes set limits to the image resolution. Here we assume the field perturbations to be slow, and thus a single field measurement per readout suffices for correction. It is shown that with this approach good image quality can be recovered in T2*-weighted images, that display strong ghosting when not corrected. Unlike full k-space monitoring, the approach is applicable also to high-resolution imaging.

Hiroyuki Takeda¹ and Boklye Kim^2
1Radiology, University of Michigan, Ann Arbor, Michigan, United States, ²Radiology, University of Michigan, ann arbor, MICHIGAN, United States

In practice, the field map dynamically changes with head motion during the scan, and such a changing field leads to variations in geometric distortions in EPI. A previous retrospective approach of approximating a dynamic field map by applying rigid body transformations to an observed static field map may be insufficient in the presence of significant out-of-plane rotations. Our approach is to retrospectively estimate the object’s susceptibility map from an observed high-resolution static field map using an estimator derived from a probability density function of non-uniform noise. This approach is capable of finding the susceptibility map regardless of the wrapping effect.

Martin Krämer¹ and Jürgen R. Reichenbach^1
1Medical Physics Group, Institute of Diagnostic and Interventional Radiology I, Jena University Hospital - Friedrich Schiller University Jena, Jena, Germany

Novel reconstruction technique for PROPELLER-EPI which requires no additionally measured field map for distortion correction. By modulating the input data with constant frequency offsets the effective resonance frequency on which the image is reconstructed is modulated, enabling autocalibrated PROPELLER-EPI reconstruction. The theory of the technique is explained in detail including a proposed algorithm for automatic detection of on resonant image parts. We compare the proposed method to standard PROPELLER-EPI image reconstruction using statically measured field maps showing that comparable results can be achieved.

Jonathan Bishop¹, R Mark Henkelman^1,2, and Brian J. Nieman^1,2
1Hospital for Sick Children, Toronto, ON, Canada, ²Medical Biophysics, University of Toronto, Toronto, ON, Canada

Phase errors generated by the phase encoding gradients of a 3D fast spin-echo sequence for preclinical microscopy are measured with a novel prescan sequence and applied as a retrospective correction at reconstruction to improve image quality.

Tijl A. van der Velden¹, Vincent Oltman Boer¹, Peter R. Luijten¹, and Dennis W.J. Klomp^1
1Radiology, University Medical Center Utrecht, Utrecht, Utrecht, Netherlands

Field probes have been used for shot-to-shot B₀ correction multi shot DWI images of the human breast at 7T. B₀ correction is necessary for multi shot imaging to prevent ghosting due to temporal B₀ fluctuations caused by physiological motions. In this study we demonstrate that the information simultaneously obtained from field probes reduce ghosting artefacts in multi-shot DWI at 7T in the human breast.

Seung-Kyun Lee¹, Ek T. Tan¹, Ambey Govenkar², and Ileana Hancu^1
1GE Global Research, Niskayuna, NY, United States, ²Extenprise, Pune, India

Dynamic slice-dependent update of linear shim gradients and center frequency was implemented in 3 T breast imaging. The method was applied to axial bilateral breast DWI on four volunteers. In all volunteers the measured B0 maps showed significantly improved homogeneity. Anatomy-referenced ADC maps also showed reduced image registration error obtainable with the proposed method.

Erik B. Beall¹, Myung-Ho In², Oliver Speck², Ken E. Sakaie¹, and Mark J. Lowe^1
1Imaging Institute, Cleveland Clinic, Cleveland, OH, United States, ²Biomedical Magnetic Resonance, Otto-von-Guericke University, Magdeburg, Germany

Eddy current artifacts reduce resolving power of DTI scans and obliterate small fiber tracts. By acquiring forward and reverse phase-encodes for every direction and using new algorithms for unwarping these together, both the static and dynamic (diffusion direction, breathing and head position-dependent changes) field inhomogeneity can be removed. This dramatically improves the tensor fit error and uncovers structures invisible with only static unwarping.

Rachel Wai-chung Chan¹, Sebastian Kozerke^2,3, Daniel Giese³, Jack Harmer³, Christian T. Stoeck², Constantin von Deuster^2,3, Andrew Peter Aitken³, and David Atkinson^1
1Centre for Medical Imaging, University College London, London, United Kingdom, ²Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland, ³Division of Imaging Sciences, King's College London, London, United Kingdom

In diffusion tensor imaging (DTI), time-varying eddy-currents over long readout durations cause images from different directions to be misregistered. Using a field camera with 16 NMR probes, higher-order spatial phase offsets from eddy-currents were measured and used for correction of misregistration artifacts in renal DTI. The unipolar Stejskal-Tanner and velocity-compensated bipolar sequences were compared. Phantom experiments reveal that higher-order correction is beneficial for the unipolar sequence. In an in vivo experiment where both kidneys in a healthy volunteer were simultaneously imaged, the fractional anisotropy (FA) maps showed improved image quality with eddy-current correction.

Joseph Y. Cheng¹, Shreyas S. Vasanawala², Michael Lustig³, and John M. Pauly^4
1Electrical Engineering, Stanford University, Stanford, California, United States, ²Radiology, Stanford University, Stanford, California, United States, ³Electrical Engineering and Computer Sciences, University of California, Berkeley, California, United States, ⁴Electrical Engineergin, Stanford University, Stanford, California, United States

There have been major advancements in reducing motion corruption especially for rigid-body motion. Even with these methods, there may be residual artifacts due to motion measurement error. Automatic correction methods can be applied without any motion information. We propose a novel automatic approach using phase-retrieval and sparsity constraints. This approach is incorporated with parallel imaging and compressed sensing to help guide the correction. First, the accuracy of the algorithm is demonstrated in a simulation head study. Afterwards, the method is used to correct motion from conventional in vivo scans.

Saikat Sengupta¹, Sasidhar Tadanki², John C . Gore¹, and E. Brian Welch^1
1Radiology and Radiological Sciences, Vanderbilt University, Nashville, TN, United States, ²Vanderbilt University, Nashville, TN, United States

We present the development and use of inductively coupled wireless NMR markers for prospective rigid body motion correction at 7 Tesla with the ultimate goal of real time head motion correction. Three 5 mm markers mounted on a phantom are located using a linear navigator. Motion is estimated in real time and imaging geometry is updated prospectively to compensate for motion with 6 degrees of freedom. Effective real time correction of complex motions in 1mm³ voxel size gradient echo imaging is demonstrated. Inductively coupled markers add significant benefits in flexibility, comfort, size and receiver channel requirements while maintaining performance.

Mathias Engström^1,2 and Stefan Skare^1,2
1Department of Clinical Neuroscience, Karolinska Institutet, Stockholm, Stockholm, Sweden, ²Department of Neuroradiology, Karolinska University Hospital, Stockholm, Stockholm, Sweden

This work details a two step in-plane motion correction method for diffusion-weighted 3D Multi-Slab EPI, using the navigator readouts and the overlapping area between adjacent slabs.

Brian Robert Keating¹ and Thomas Ernst^2
1Department of Medicine, Unversity of Hawaii, Honolulu, HI, United States, ²Department of Medicine, University of Hawaii at Manoa, Honolulu, HI, United States

Subject motion introduces unwanted phase variance in the MR signal, resulting in image/spectral artifacts. We performed single-voxel brain scans (PRESS) while subject head movement was monitored by an external optical tracking system. The motion-induced phase evolution was estimated based on tracking data and knowledge of the gradient waveforms. Phase corrections were applied in post-processing, resulting in improved phase coherence and increased SNR on real spectra. These results demonstrate the promise of a high-accuracy motion tracking system for reducing movement-related artifacts in phase-sensitive modalities.

Jieying Luo¹, R. Reeve Ingle¹, and Dwight G. Nishimura^1
1Electrical Engineering, Stanford University, Stanford, California, United States

An automatic method for ROI specification and reference frame selection has been developed to enable automatic motion estimation from two-dimensional image navigators. Filtering and feature region searching techniques are used to automatically detect an ROI covering the heart. A novel principal component analysis technique is used to extract respiratory information from the image navigators. This respiratory information is used to identify an image at end expiration to use as a reference for subsequent motion estimation. This method is applied to a free-breathing coronary MR angiography acquisition to enable automatic display of motion-corrected images after a scan.

James G. Pipe¹, Nicholas Ryan Zwart¹, Michael Schar², Wende N. Gibbs³, and John P. Karis^3
1Neuroimaging Research, Barrow Neurological Institute, Phoenix, Arizona, United States, ²Philips Healthcare, Phoenix, Arizona, United States, ³Neuroradiology, Barrow Neurological Institute, Phoenix, Arizona, United States

A novel method for estimating motion from PROPELLER data is given. A modification to the original algorithm allows one to simultaneously solve for correlations between all blade pairs at the same time. Comparison to the original method shows a small but consistent improvement in image quality.

Jason K. Mendes¹, Dennis L. Parker¹, Robb Merrill¹, and J. Rock Hadley^1
1Radiology, University of Utah, Salt Lake City, Utah, United States

Carotid MRI still suffers from blurring and ghosting artifacts due to patient swallowing and associated palatal motion. A simple pneumatic device coupled to a respiration monitoring system (available on most clinical scanners) can be used to detect patient swallowing and improve image quality.

Zhongyuan Bi¹, Martin Uecker², Dengrong Jiang³, Michael Lustig², and Kui Ying^3
1Biomedical Engineering, Tsinghua University, Beijing, Beijing, China, ²Electrical Engineering and Computer Science, University of California Berkeley, Berkeley, California, United States, ³Engineering Physics, Tsinghua University, Beijing, Beijing, China

Auto-calibration parallel imaging (acPI) is based on local correlations in k-space. It is known to perform robustly in practice, especially when accurate sensitivity information is hard to obtain. However, corruption of ACS data, e.g. by motion, often leads to serious artifacts in the reconstructed images. In this work, we propose to exploit the redundancy in k-space to detect and correct sparse corruptions in ACS data, which could result from random, time-limited motion in clinical practice (e.g. swallowing, jerk, etc). Our work is based on low-rank matrix completion with sparse errors.

A. Alhamud¹, B. Laughton², Khader M. Hasan³, André J. W. van der Kouwe⁴, and Ernesta M. Meintjes^1
1Human Biology, MRC/UCT Medical Imaging Research Unit, University of Cape Town, Cape Town, Western Cape, South Africa, ²Paediatrics and Child Health, Stellenbosch University and Tygerberg Hospital, Cape Town, Western Cape, South Africa, ³University of Texas Health Science Center at Houston, Houston, Texas, United States, ⁴Department of Radiology, Massachusetts General Hospital, Charlestown, Massachusetts, United States

Many diffusion tensor imaging (DTI) studies have reported changes in DTI measures especially fractional anisotropy (FA) when studying brain development or neuro-pathological diseases. Moreover, other studies have found that motion and retrospective motion correction may introduce a positive or negative bias to DTI data. In this study we exploit the navigated diffusion sequence (Alhamud et al., 2012) to measure the patterns of head motion in 5-year old children and their effect on DTI data. The influence of retrospective motion correction with and without rotating the diffusion table on DTI data acquired using the standard diffusion sequence was also investigated.

Zhe Liu¹, Zhe Zhang¹, Sheng Fang², Juan Wei³, Chun Yuan^1,4, and Hua Guo^1
1Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China, ²Institute of nuclear and new energy technology, Tsinghua University, Beijing, China, ³Philips Research Asia Shanghai, Beijing, China, ⁴Department of Radiology, University of Washington, Seattle, WA, United States

In PROPELLER, reference data plays a critical role for rigid motion correction. In current practice, there are two methods to generate the reference, single-blade reference method and combined-blade reference method. However, these two methods may fail in certain scenarios. In this study, we propose a new method, grouped¨Cblade reference, for reference generation. Instead of using a single blade or forcing all blades to one direction, our method groups blades with the similar orientations together. Preliminary results show that the new method needs less iteration to converge, and is able to give results with higher quality compared to other methods.

Onur Afacan¹, Ali Gholipour¹, Benoit Scherrer¹, and Simon K. Warfield^1
1Computational Radiology Laboratory, Department of Radiology, Boston Children's Hospital and Harvard Medical School, Boston, Massachusetts, United States

The sensitivity of diffusion imaging to motion combined with this increased scan time creates a need for a motion correction strategy, especially with uncooperative patients such as children. Here in this work, we demonstrated that the information from the phase encoding correction lines acquired with an EPI acquisition can be used to detect motion in real time, and can be used to improve the quality of long diffusion scan when there is substantial motion during the scan.

Julien Senegas¹, Christian Buerger¹, Torbjoern Vik¹, Peter Mazurkewitz¹, and Peter Koken^1
1Philips Research Laboratories, Hamburg, Germany

The focus of this work is on the pelvis area, as MRI is being more and more established as imaging modality to assess and monitor prostate cancer and is expected to play an increasing role in the adaptive planning of radiation therapy. We investigated in volunteers a method to correct for bulk motion between consecutive examinations and analyzed the amplitude of local, non-rigid deformations as the consequence of bladder and rectum filling, with particular focus on the prostate. We think that the methodology proposed in this work to control bulk motion and assess local deformations between consecutive imaging sessions has important applications in therapy planning and image-guided therapy delivery, especially in the field of radio-therapy.

Andreas Hock¹, Spyros S. Kollias², Peter Boesiger¹, and Anke Henning^1,3
1Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland, ²Institute of Neuroradiology, University Hospital of Zurich, Zurich, Switzerland, ³Max Planck Institute for Biological Cybernetics, Max Plank Institute, Tuebingen, Germany

Subject motion is one of the major problems for MR spectroscopy since it often leads to inaccurate results. The detection of patient motion by observing exclusively the spectra is sometimes impossible. Therefore, in this investigation, 1D navigator acquisitions interleaved with non-water-suppressed MRS measurements are used for real-time subject-motion detection and correction in the spinal cord. Interleaved navigators allow precise real-time motion detection for MRS without the need of additional scan-time and additional hardware. Therefore, it allows early intervention (e.g. asking the subject to lay still) and in combination with non-water-suppressed MRS for a retrospective motion correction even in very small regions of interest like the spinal cord.

Guobin Li¹, Maxim Zaitsev¹, Esther Meyer², Dominik Paul², and Jürgen Hennig^1
1University Medical Center Freiburg, Freiburg, Germany, ²Siemens Healthcare, Erlangen, Germany

The measurement time of 3D MR acquisition with turbo spin echo (TSE) sequence is usually long, and very prone to patient movement during the scanning. This leads to degrading of reconstructed image by motion artifacts. An compressed sensing based acquisition scheme is proposed to address this problem, which has these features: scanning stops at any time when motion is detected by built-in navigators; the distribution of acquired k-space points is always optimal whenever the scanning is interrupted; Artifact-free images are reconstructed if a minimum amount of data has been acquired.

Keigo Kawaji^1,2, Pascal Spincemaille³, Thanh D. Nguyen³, Mitchell A. Cooper^1,3, Martin R. Prince³, and Yi Wang^1,3
1Department of Biomedical Engineering, Cornell University, New York, NY, United States, ²Department of Medicine, Beth Israel Deaconess Med. Ctr. and Harvard Medical School, Boston, MA, United States, ³Department of Radiology, Weill Cornell Medical College, New York, NY, United States

2D navigator image processing during MR data acquisition for real-time motion tracking applications such as prospectively gated and/or motion corrected coronary artery imaging is severely constrained by a short processing window of <50 ms for motion extraction. We present a real-time and interactive processing method that performs on a standard clinical hardware which does not require any additional dedicated hardware components for processing. This approach allows interactive 2D navigator setup by the scanner operator during data acquisition, facilitates rapid motion extraction from raw 2D navigator k-space data, and adjusts the scan instruction within 20 milliseconds based on the extracted motion information.

Wyger M. Brink¹, Peter Börnert^1,2, Kay Nehrke², and Andrew Webb^1
1Radiology, Leiden University Medical Center, Leiden, Netherlands, Zuid-Holland, Netherlands, ²Philips Research Laboratories, Hamburg, Hamburg, Germany

High resolution B1+ mapping techniques in the brain at high field often show “residual” structure from, in particular, the ventricles. This can be partially explained by the high contrast in electrical conductivity of CSF with respect to white and grey matter, in addition to long relaxation time-related effects. This factor should be considered when assessing the accuracy of various B1+ mapping sequences.

Jay Moore^1,2, William A. Grissom^1,3, and John C. Gore^1,2
1Vanderbilt University Institute of Imaging Science, Vanderbilt University, Nashville, TN, United States, ²Department of Radiology and Radiological Sciences, Vanderbilt University, Nashville, TN, United States, ³Department of Biomedical Engineering, Vanderbilt University, Nashville, TN, United States

B₁⁺ distributions in the brain are predicted by matching head shapes as determined from survey scans to those in a database consisting of pre-existing B₁⁺ maps and associated survey scans. Results are evaluated through comparison with actual maps as well as through the performance of RF pulses designed from both actual and predicted maps.

Patrik Wyss¹, Shaihan J. Malik², Vincent Oltman Boer³, Johanna J. Bluemink³, Alexander Raaijmakers³, Mustafa Cavusoglu⁴, Peter R. Luijten³, Johannes J. Hoogduin³, Giel Mens⁵, and Anke Henning^1,6
1Institute for Biomedical Engineering, UZH and ETH Zurich, Zurich, Switzerland, ²Imaging Sciences and Biomedical Engineering, Kings College London, London, United Kingdom,³University Medical Center Utrecht, Utrecht, Netherlands, ⁴Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland, ⁵Philips Healthcare, Best, Netherlands, ⁶Max Planck Institute for Biological Cybernetics, Tübingen, Germany

To achieve a sufficient dynamic range for single channel B1+ mapping required for calibration of multi-channel transmit arrays two distinct solutions have been suggestion recently: (1) Bloch-Siegert effect based B1+ mapping and (2) B1+ mapping using linear combinations of transmit channels. In this work, both approaches are combined and interferometric Bloch-Siegert B1+ mapping is introduced. This approach is cross-validated against two recently introduced interferometric B1+ mapping methods based on actual flip angle imaging (AFI) and a multi flip-angle pre-pulse method as well as against single-channel AFI based B1+ mapping.

Mohammad Mehdi Khalighi¹, Jason H. Su^2,3, and Brian K. Rutt^3
1Applied Science Lab, GE Healthcare, Menlo Park, California, United States, ²Department of Electrical Engineering, Stanford University, Stanford, California, United States,³Department of Radiology, Stanford University, Stanford, California, United States

Adiabatic Bloch-Siegert (ABS) B₁⁺ mapping with spiral readout has been recently introduced as a fast and efficient way of volumetric B₁⁺ mapping with no T1 or T2 dependency. Here we have analyzed the angle-to-noise ratio (ANR) of the ABS method and show phantom and brain ANR maps at 7T. We also compare analytically the phase-based ABS using spiral readout with the magnitude-based DREAM method. Our comparison shows that ABS generates >3 times ANR compared to DREAM for similar scan times.

Yuehui Tao¹, Aaron T. Hess¹, Graeme A. Keith¹, Christopher T. Rodgers¹, and Matthew D. Robson^1
1University of Oxford, Oxford, United Kingdom

Cardiac MRI at 7T has several potential advantages over 1.5T and 3T, such as higher signal-to-noise ratio and higher resolution. A quantitative B1 map is routinely required at 7T to allow the management of large B1 variations across the heart. Conventional methods based on rectangular saturation pulse are not accurate in this case because of large B0 inhomogeneity and low flip angle from low maximum B1 power. We propose to use a non-selective broad-band full-passage HS8 pulse that is operating outside its adiabatic state to obtain B1 maps in situations where the B1 is low and the B0 is inhomogeneous such as for cardiac applications at 7T.

KyungHyun Sung¹, Manoj Saranathan², Bruce L. Daniel², and Brian Andrew Hargreaves^2
1Radiological Sciences, UCLA, Los Angeles, California, United States, ²Radiology, Stanford University, Stanford, California, United States

Variable flip angle (VFA) imaging is a common choice to measure T1 since it can provide fast volumetric T1 mapping, but is highly sensitive to flip angle variation. We describe a novel way to simultaneously measure T1 and B1 maps using fat-only VFA images, assuming the T1 relaxation times in fat to be globally uniform, and the B1 inhomogeneity is smoothly varying across the object. We showed B1 maps using the proposed method are similar those using the conventional double angle method. Additionally, we demonstrated we can reduce a T1 estimation error by using our simultaneous T1 and B1 mapping method.

Kosuke Ito¹, Masahiro Takizawa¹, and Tetsuhiko Takahashi^1
1MRI system division, Hitachi Medical corporation, Kashiwa, Chiba, Japan

Fast B₁ mapping method for multi channel transmit coil is developed. In this method, B₁ map acquisition is only one time with using all transmit channels. The acquired B₁ map is decomposed to B₁ map for each transmit channel by using the information of phase difference between transmit channels. Without repeating B₁ map acquisition, short scan time is achieved. Also, only B₁ map for all channel transmission is used, this method relaxes the required accuracy of B₁ map. For 4-channel transmit coil, scan time is only 1 sec.

Kosuke Ito¹, Masahiro Takizawa¹, and Tetsuhiko Takahashi^1
1MRI system division, Hitachi Medical corporation, Kashiwa, Chiba, Japan

A new fast B₁ mapping method (multi Td method) has developed. 3 images are used to calculate B₁ maps in this method, images acquired without pre-pulse, and images acquired at two different delay times (Td₁ and Td₂) from pre-pulse. The scan time is short due to the use of single shot scan. No approximation about T₁ relaxation is needed, and the dynamic range of B₁ map calculation is wide. Scan time is only 980 ms per slice.

Daniel Brenner¹, Kaveh Vahedipour¹, Tony Stöcker¹, Jörg Felder¹, Frank Geschewski¹, and Nadim Jon Shah^1,2
1INM-4, Forschungszentrum Juelich, Juelich, Germany, ²JARA - Faculty of Medicine, RWTH Aachen University, Aachen, Germany

UHF MRI at 9.4T suffers from strong RF inhomogeneities. Measurement of the field distribution of a pTX system is crucial for RF shims and pulse design focussing on removing these problems. A robust 3D whole brain calibration protocol, yielding B1 and B0 maps and an approximate brain mask in approximately 5 minutes is demonstrated. The B1 information is corrected for the effect of off-resonances and limited dynamic range.

Zhen Feng¹, Feng Liu², Stuart Crozier², and He Guo^1
1Dalian University of Technology, Dalian, Liaoning, China, ²The University of Queensland, Brisbane, Queensland, Australia

In the sequential combination of parallel imaging (PI) and compressed sensing (CS) MRI, the CS procedure is conventionally performed on individual coils. In fact, the individual coil data are sensitivity-weighted maps of the whole MRI image, therefore signal overlapping exists between coil data. In this work, we propose a novel sparsity basis to improve CS reconstruction through the exploitation of the inter-coil spatial redundancies. In addition, by introducing a new filter that separates the measured and reconstructed data during L2-norm optimization, noise and errors can be minimized in the sequential PI-CS method. The brain image study showed the promise of the new PI-CS scheme.

Daniel S. Weller¹, Sathish Ramani¹, Jon-Fredrik Nielsen², and Jeffrey A. Fessler^1
1EECS, University of Michigan, Ann Arbor, MI, United States, ²BME, University of Michigan, Ann Arbor, MI, United States

We apply a Monte-Carlo method for estimating Stein's Unbiased Risk Estimate (SURE) to regularization parameter selection for L1-SPIRiT auto-calibrating parallel imaging reconstruction. We validate the error criterion against observed mean-squared error and demonstrate the L1-SPIRiT reconstruction quality using the SURE-optimal regularization parameter for a range of noise levels using fully-sampled multi-channel real data.

Stephen F. Cauley¹, Yuanzhe Xi², Berkin Bilgic³, Kawin Setsompop^4,5, Jianlin Xia², Elfar Adalsteinsson^4,6, V. Ragu Balakrishnan⁷, and Lawrence L. Wald^4,8
1A.A. Martinos Center for Biomedical Imaging, Dept. of Radiology, Massachusetts General Hospital, Charlestown, MA, United States, ²Department of Mathematics, Purdue University, West Lafayette, IN, United States, ³Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, United States, ⁴A.A. Martinos Center for Biomedical Imaging, Dept. of Radiology, MGH, Charlestown, MA, United States, ⁵Harvard Medical School, Boston, MA, United States,⁶Department of Electrical Engineering and Computer Science, MIT, Cambridge, MA, United States, ⁷School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, United States, ⁸Harvard-MIT Division of Health Sciences and Technology, Cambridge, Massachusetts, United States

The far reaching adoption of compressed sensing for clinic MRI hinges on the ability to accurately produce images in a reasonable time-frame. Multiple contrast studies have been successfully combined with joint Bayesian reconstruction for improved image quality. However, current techniques have prohibitive computational requirements. We consider a joint Bayesian approach that approximates point spread functions to exploit sparse matrix methods. We leverage hierarchical matrix analysis and compression schemes to facilitate scalable and accurate CS reconstruction. Our approach is over 100x faster than other multiple contrast approaches while still improving image accuracy by over 35% compared to single image CS techniques.

Satoshi Ito¹, Kazuki Nakamura¹, and Yoshifumi Yamada^1
1Research Division of Intelligence and Information Sciences, Utsunomiya University, Utsunomiya, Tochigi, Japan

We present a new Compressed Sensing reconstruction that is robust to phase variations in MR images. When the signal trajectory in k-space is symmetrical with respect to its origin, the k-space signal corresponding to the real and imaginary parts of the complex image can be synthesized independently by restricting the k-space signal to an even function or an odd function. The proposed method involves random but symmetrical k-space acquisition and independent reconstruction of the real and imaginary parts of images using the real-valued constraint.

Hao Gao^1,2, Longchuan Li³, and Xiaoping P. Hu^3
1Department of Mathematics and Computer Science, Emory University, Atlanta, Georgia, United States, ²Department of Radiology and Imaging Sciences, Emory University, Atlanta, Georgia, United States, ³Department of Biomedical Engineering, Emory University and Georgia Institute of Technology, Atlanta, Georgia, United States

In another submitted abstract “Compressive Diffusion MRI – Part 1: Why Low-Rank?”, we compared several sparsity models and found that the low-rank (LR) model is the most suitable for diffusion MRI. This abstract introduces the Prior-image Constrained LR (PCLR) model, through which prior images can be efficiently incorporated to improve LR. In addition, a simple-to-implement and efficient algorithm has been developed to solve PCLR. The application of PCLR to diffusion MRI, with the prior images that are different from the images to be reconstructed, showed that PCLR performs better than LR.

Sheng Fang¹, Wenchuan Wu², Kui Ying³, and Hua Guo^2
1Institute of nulcear and new energy technology, Tsinghua University, Beijing, China, ²Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China, ³Department of engineering physics, Tsinghua University, Beijing, China

A Nonlocal Total variation (NLTV) that automatically refines the image-dependent weights was proposed for reconstructing highly undersampled variable density spiral (VDS) imaging data. Unlike existing NLTV-related method, the proposed method doesn¡¯t rely on a reference image for weight map estimation. Instead, it automatically updates the weight based on a filtered intermediate image. The wavelet soft shrinkage method is used for the filtering step. Since the aliasing artifact of VDS is incoherent, it can be expected that the shrinkage can perturbation of aliasing artifact and increase the accuracy of weight computation. The in vivo VDS experiment demonstrates that the proposed method can effectively suppress noises amplification and perverse better image details than TV.

Srikant Kamesh Iyer^1,2, Tolga Tasdizen³, Nathan Burgon⁴, Ganesh Adluru², and Edward V.R. DiBella^2
1Electrical and Computer Engineering, University of Utah, Salt Lake City, Utah, United States, ²UCAIR/Radiology, University of Utah, Salt Lake City, Utah, United States, ³SCI, University of Utah, Salt Lake City, Salt Lake City, Utah, United States, ⁴CARMA, Department of Internal Medicine, University of Utah, Salt Lake City, Salt Lake City, Utah, United States

Acquiring Late Gadolinium Enhanced (LGE) images of the left atrium is a valuable tool in assessing the degree of fibrosis in the left atrium. The current method of acquiring high resolution 3D Cartesian inversion recovery data with ECG gating and respiratory navigator is inherently time consuming. Advances in compressed sensing have made it possible to speed up acquisition by acquiring less data while maintaining image quality by using prior information about the underlying image as constraints in the reconstruction. Total variation is one such popular constraint used. The nonlinearity and poor conditioning of such L1 regularization based reconstruction schemes makes minimization using traditional schemes like gradient descent very slow. We propose to use the Split Bregman approach to reconstruct LGE images of the LA in a compressed sensing framework to achieve rapid reconstructions for high acceleration factors

Il Yong Chun¹ and Thomas Talavage^1,2
1School of Electrical and Computer Engineering, Purdue University, West Lafayette, Indiana, United States, ²Weldon School of Biomedical Engineering, Purdue University, West Lafayette, Indiana, United States

We propose a fast constrained L(p,¥å)-L2-norm (L(p,¥å) is an approximated Lp-qusi-norm) minimization algorithm, based on 1) p- and ¥å-dependent weighting techniques, and 2) an efficient split Bregman-based (known to have rapid convergence, especially with an L1-norm ) reweighted L1-minimization algorithm. This L(p,¥å)-L2-norm minimization achieves exact reconstruction from fewer measurements than are required for the L1-L2-norm case.

Jordan Woehr¹ and Michael Smith^1,2
1Electrical and Computer Engineering, University of Calgary, Calgary, Alberta, Canada, ²Radiology, University of Calgary, Calgary, Alberta, Canada

Compressed sensing (CS) applied to under-sampled k-space is used in magnetic resonance to decrease 2D and 3D imaging times while maintaining image resolution. We present the Gardner transform (GT) as a potential sparsifying transform for use with CS with under-sampled or truncated signals in a non-k-space 4^th dimension, e.g. time or frequency. New classes of signals can be sparsified with the GT such as sums of exponentials, Lorentzians, etc. We discuss the practical issues associated with GT-CS reconstruction of a simulated signal assumed to have two exponential components, MTT_GM and MTT_WM, and an ideal GT of two Dirac delta functions.

Enhao Gong¹, Feng Huang², Kui Ying³, Wenchuan Wu⁴, Shi Wang³, Chun Yuan^4,5, and George Randy Duensing^6
1Electrical Engineering, Stanford University, Stanford, CA, United States, ²Philips Healthcare, Shanghai, China, ³Department of Engineering Physics, Tsinghua University, Beijing, China, ⁴Center for Biomedical Imaging Research, Department of Biomedical Engineering, Tsinghua University, Beijing, China, ⁵Department of Radiology, University of Washington, Seattle, WA, United States, ⁶Philips Healthcare, Gainesville, FL, United States

In this work, PROMISE (Parallel Reconstruction with Optimized acquisition for Multi-contrast Imaging in the context of compressed Sensing) is proposed to use manifold sharable information between multi-contrast scans for fast imaging. With the assumption that the same FOV is scanned for multi-contrast MRI, coil sensitivity maps, image structural information and optimal acquisition trajectory were extracted in seconds from previously acquired/reconstructed data to enhance the reconstruction of the following scans. Compared to previous work, PROMISE used more sharable information and resulted in lower artifact/noise level at higher reduction factors. Moreover, PROMISE is able to tolerate inter-scan motions better and is more clinically applicable.

Leo K. Tam¹, Gigi Galiana², Jason P. Stockmann³, Andrew Dewdney⁴, Terence W. Nixon², Dana C. Peters², and Robert Todd Constable^2,5
1Biomedical Engineering, Yale University, New Haven, CT, United States, ²Diagnostic Radiology, Yale University, New Haven, CT, United States, ³Martinos Center, Massachusetts General Hospital, Boston, Massachusetts, United States, ⁴Siemens AG Healthcare, Erlangen, Bavaria, Germany, ⁵Neurosurgery, Yale University, New Haven, CT, United States

O-space imaging has shown distributed artifacts due to non-linear encoding via spatially-varying center placements (CPs). The success of non-linear encoding methods in the image domain lead to development of an approach to maximize incoherence in a sparse transform domain such as the Daubeuchies wavelets. By pseudo-randomizing CPs, an incoherence optimized O-space acquisition produced superior reconstructions under a compressed sensing framework.

Sheng Fang¹, Wenchuan Wu², Kui Ying³, and Hua Guo^2
1Institute of nulcear and new energy technology, Tsinghua University, Beijing, China, ²Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China, ³Department of engineering physics, Tsinghua University, Beijing, China

An ultra-fast variable density spiral (VDS) imaging technique using multiscale CORNOL (coherence regularization using a nonlocal operator) reconstruction is proposed. The multiscale CORNOL circumvents the conflict of large-scale artifact suppression and fine-scale structure perseveration of nonlinear reconstruction by sequentially handling these two tasks with smoothness constraint at different scales. This sequential procedure enables further reduction both residual aliasing artifact and ring-like artifact of VDS with well-preserved image details. Both simulation and in vivo experiment results demonstrate that the proposed method can suppress both large-scale artifacts and noise while preserving image details well at high sampling reduction factors.

Sampada Bhave¹, Jinsuh Kim¹, Casey P. Johnson¹, and Mathews Jacob^1
1University of Iowa, Iowa City, Iowa, United States

A novel method based on multishot variable density spiral acquisitions and compressive sensing is introduced to reduce the scan time in high spatial resolution quantitative CEST imaging. Variable density acquisitions provide acceptable signal to noise ratio, even when high resolution acquisitions are used. The k-space data of each of the z-planes is undersampled by skipping the interleaves of the spiral trajectory. The recovery of the entire spatial spectral dataset is posed as a sparse optimization scheme. The total variation prior is used for spatial regularization, while the l1 norm of the second order temporal derivatives are used to exploit the smoothness of z-spectra.

Daniel Stäb¹, Tobias Wech¹, Dietbert Hahn¹, and Herbert Köstler^1
1Institute of Radiology, University of Würzburg, Würzburg, Bavaria, Germany

In CS, metrics for evaluating the image quality are hardly available. In this work, a simple Monte Carlo approach is presented that allows evaluating systematic and statistical reconstruction errors. Based on a fully sampled reference acquisition and a noise measurement, multiple pseudo measurements are synthesized. By comparing their reconstructions to the reference, systematic signal deviations can be quantified. In addition, the extent of statistical fluctuations can be estimated. The proposed evaluation was applied to a x-f-space CS reconstruction in myocardial perfusion MRI and systematic flattening of the signal intensity time courses was detected.

Florian Knoll¹, Rafael O'Halloran², Kristian Bredies³, Rudolf Stollberger¹, and Roland Bammer^2
1Institute of Medical Engineering, Graz University of Technology, Graz, Austria, ²Radiology, Stanford University, Palo Alto, California, United States, ³Department of Mathematics and Scientific Computing, University of Graz, Graz, Styria, Austria

Diffusion Tensor Imaging is a demanding application requiring the acquisition of many image volumes to extract the desired tensor parameters. k-space undersampling is a straightforward method that can be used to reduce the total scan time, however, if the undersampled data is reconstructed with conventional methods such as gridding, artifacts result. Parallel imaging and compressed sensing are successful in reducing undersampling but it is not clear what effect nonlinear regularization terms have with respect to quantitative evaluation of the images, as performed in Diffusion Tensor Imaging. Here the quantitative accuracy of a 3D spiral acquisition using nonlinear regularization is evaluated in a simulated atlas-based DTI phantom.

Kieren Grant Hollingsworth¹ and David M. Higgins^2
1Institute of Cellular Medicine, Newcastle University, Newcastle upon Tyne, Tyne and Wear, United Kingdom, ²Philips Healthcare, Guildford, Surrey, United Kingdom

Compressed sensing (CS) and parallel imaging (PI) have been successfully applied to the problem of water-fat separation, providing valuable savings in acquisition time. However, it has not been shown that quantitative fat fraction is preserved. Fully sampled and prospectively undersampled (4.4x) 3D gradient echo scans were performed on the lower leg of a volunteer, reconstructed by CS and PI and fat fraction maps produced. 4 regions-of-interest were considered across 5 axial levels and compared individually and by Bland-Altman analysis. No significant difference was found in the fat fraction for the regions and no bias between fully-sampled and undersampled data.

Huisu Yoon¹, Ganesh Adluru², Edward V.R. DiBella², and Jong Chul Ye^1
1Department of Bio and Brain Engineering, KAIST, Daejeon, Korea, ²Utah Center for Advanced Imaging Research, Department of Radiology, University of Utah, Salt Lake City, UT, United States

k-t FOCUSS with motion estimation and compensation is a promising tool for highly accelerated dynamic MRI. One of the limitations of k-t FOCUSS with ME/MC is, however, that the motion residual signals are often contaminated with background noises, so the reconstruction of the residual signal using standard l1 sparsity constraint are often inefficient in capturing the physiological features and results in temporal blurring. In this work, we exploit the reordering algorithm to make the residual reconstruction more efficient. Using cardiac perfusion imaging, the spatio-temporal constrained reconstruction using re-ordering was found effective for reconstruction of the motion residual in k-t FOCUSS with ME/MC. By combining k-t FOCUSS ME/MC, the ordering based residual reconstruction may be a useful tool for compressed sensing MRI.

Wingchi Edmund Kwok^1,2, Yue Hu³, Zhigang You¹, and Mathews Jacob^4
1Department of Imaging Sciences, University of Rochester, Rochester, NY, United States, ²Rochester Center for Brain Imaging, University of Rochester, Rochester, NY, United States, ³Department of Electrical and Computer Engineering, University of Rochester, Rochester, NY, United States, ⁴Department of Electrical and Computer Engineering, University of Iowa, Iowa City, IA, United States

MR angiography (MRA) of fingers is challenging due to the small size of blood vessels. Though high-resolution MRA may be obtained with dedicated RF coils, the associated long scan time and limited coverage hinder applications. Therefore, we evaluated the feasibility of applying compressed sensing on high-resolution finger MRA to save scan time. Our results show that compressed sensing can significantly reduce scan time while preserving blood vessel information, facilitating the clinical application of high-resolution finger MRA. Accelerated high-resolution finger MRA with compressed sensing should be useful for the diagnostic evaluation and pathogenesis studies of systemic sclerosis and arthritis.

Jieying Luo¹, Taehoon Shin¹, Tao Zhang¹, Bob S. Hu², and Dwight G. Nishimura^1
1Electrical Engineering, Stanford University, Stanford, California, United States, ²Palo Alto Medical Foundation, Palo Alto, California, United States

An accurate measurement of lower extremities perfusion is potentially of significant help in the assessment of peripheral arterial disease. This work investigates and optimizes the use of low-rank matrix completion reconstruction for this application. As verified using both numerical simulations and retrospectively undersampled in-vivo data, reconstruction performance is improved by the use of reference images and a complementary uniformly random undersampling pattern. With this method, volumetric perfusion imaging of the lower extremities with temporal resolution of 2 seconds can be achieved.

Fan Lam^1,2, Bo Zhao^1,2, Yinan Liu³, Zhi-Pei Liang^1,2, Michael Weiner^3,4, and Norbert Schuff^3,4
1Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ²Beckman Institute, University of Illinois at Urbana-Champaign, Urbana, IL, United States, ³Center for Imaging of Neurodegenerative Diseases, Department of Veteran Affairs Medical Center, San Francisco, CA, United States, ⁴Department of Radiology and Biomedical Imaging, University of California, San Francisco, CA, United States

We present a new method for image reconstruction from undersampled data for accelerating fMRI data acquisition. The proposed method integrates a low-rank model of the fMRI image series and a sparsity constraint in a unified mathematical formulation, enabling high quality reconstruction of fMRI images from highly undersampled data. Representative results from simulations based on experimental data were used to demonstrate the performance of the proposed method.

Julio Duarte-Carvajalino¹, Christophe Lenglet¹, Junqian Xu², Essa S. Yacoub¹, Kamil Ugurbil¹, Steen Moeller¹, Lawrence Carin³, and Guillermo Sapiro^3
1Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN, United States, ²Mount Sinai School of Medicine, New York, NY, United States, ³Dept. of Electrical and Computer Engineering, Duke University, Durham, NC, United States

This work introduces a novel multi-task Bayesian compressive sensing approach for the direct and joint estimation of white matter fiber orientation distribution function and diffusion-weighted volumes from under-sampled HARDI data.

Sajan Goud Lingala¹, Edward V.R. DiBella², and Mathews Jacob^1
1The University of Iowa, Iowa city, IA, United States, ²University of Utah, Salt lake city, UT, United States

We propose a novel motion compensated compressed sensing reconstruction scheme for myocardial perfusion MRI. We develop an efficient energy minimization framework that jointly estimates the motion and the dynamic images from undersampled data. Our preliminary results show that proposed scheme is able to considerably reduce motion related artifacts in temporally constrained compressed sensing reconstruction.

Matthew J. Muckley^1,2, Scott J. Peltier^1,2, Douglas C. Noll^1,2, and Jeffrey A. Fessler^3
1Biomedical Engineering, University of Michigan, Ann Arbor, MI, United States, ²Functional MRI Laboratory, University of Michigan, Ann Arbor, MI, United States, ³Electrical Engineering and Computer Science, University of Michigan, Ann Arbor, MI, United States

A novel application of low rank methods combined with temporal Fourier sparsity regularization and random sampling for removal of physiological noise in functional MRI is presented. This approach has the potential to recover high temporal frequency characteristics of the physiological noise while sampling these signals well below the Nyquist rate on average. The method is validated in a resting state connectivity task, where it is used to reconstruct a data set with high spatiotemporal resolution before removing physiological noise using low pass filtering.

Tao Zhang¹, Marcus T. Alley², Michael Lustig³, Xiaodong Li⁴, John Pauly¹, and Shreyas S. Vasanawala^2
1Electrical Engineering, Stanford University, Stanford, California, United States, ²Radiology, Stanford University, Stanford, California, United States, ³Electrical Engineering and Computer Sciences, UC Berkeley, Berkeley, California, United States, ⁴Mathematics, Stanford University, Stanford, California, United States

Dynamic contrast enhanced MRI is commonly used to detect and characterize lesions. In this work, a method named Sparsity and Low-Rank enhanced SPIRiT (SLR-SPIRiT) is proposed to push the limit of spatial-temporal resolution in 3D DCE MRI. SLR-SPIRiT exploits parallel imaging, transform-sparsity, and locally low-rank property in the dynamic image series to reconstruct highly accelerated DCE datasets. The proposed method has been validated on a pediatric DCE dataset with 1x1x2 mm3 spatial resolution and 3-second temporal resolution (actual acceleration 19.7).

Anders Garpebring¹, Maarten J. Versluis^2,3, and Matthias J.P. van Osch^2,3
1Radiation Sciences, Umeå University, Umeå, Sweden, Sweden, ²Radiology, Leiden University Medical Center, Leiden, Zuid-Holland, Netherlands, ³CJ Gorter Center for high field MRI, Leiden University Medical Center, Leiden, Zuid-Holland, Netherlands

Ultra high field MRI time-of-flight angiograms can often be seriously degraded by pulsation artifacts. A method based on random order sampling, retrospective gating and L1-SPIRiT reconstruction was developed and tested on a 7 T scanner. The results confirmed that the proposed method can reduce the flow artifacts.

Da Wang¹, Stanislas Rapacchi², Hao Gao³, and Peng Hu^4
1Biomedical Physics/Radiological Sci, UCLA, Los Angeles, CA, United States, ²UCLA, Los Angeles, CA, United States, ³Emory University, Atlanta, GA, United States, ⁴University of California Los Angeles, Los Angeles, CA, United States

A novel compressed sensing MRI reconstruction method has been proposed for dynamic MRI using Prior Rank, Intensity and Sparsity Model (PRISM). By using a low rank decomposition, PRISM can extract the stationary background component from dynamic images to further promote sparsity of the motion component for L1 norm minimization. The combination of parallel MRI methods with compressed sensing methods has shown great potential to improve the reconstructed image quality and acceleration rate. We propose to further improve PRISM compressed sensing algorithm by using TGRAPPA to fill in additional data lines in the k-space before feeding to the PRISM algorithm.

MingJian Hong¹, Feng Liu², XiaoHong Zhang¹, and YongXin Ge^1
1School of Software Engineering, ChongQing University, ChongQing, ChongQing, China, ²School of Information Technology & Electrical Engineering, The University of Queensland, Brisbane, Queensland, Australia

In this work, a sparsity-enforced Kalman filter technique for dynamic cardiac imaging is presented. The Kalman filter is firstly casted into a framework of optimization, and then a sparsity constraint is incorporated to the framework for better motion capture of the imaging object. Applications to cardiac dynamic MRI clearly demonstrated the strength of the proposed method.

Sehoon Lim¹ and Dosik Hwang^2
1SRI International Sarnoff, Princeton, NJ, United States, ²School of Electrical and Electronic Engineering, Yonsei University, Seoul, Korea

A set of different contrast MR images are usually necessary for proper diagnosis. However, the multiple acquisitions of several contrast images require long scanning time. In this study, we propose an efficient multimode compressed sensing (CS) framework to reduce the total scanning time by undersampling the k-space data of each contrast (mode) image and reconstructing artifact-minimized images. In contrast to other groups, we used a wavelet based multiscale selection CS technique to alleviate computational burden. In our studies, the running time of the proposed wavelet multimode CS is presented as a minute, which is promising for agile and compact applications.

Joonsung Choi¹, Yeji Han¹, and HyunWook Park^1
1Department of Electrical Engineering, Korean Advanced Institute of Science and Technology (KAIST), Daejeon, Korea

The highly-constrained projection reconstruction (HYPR)-based algorithms provided high spatio-temporal images by using the composite information. However, HYPR-based methods have limitation that only can reconstruct magnitude image. In the proposed method, we provide a novel method to recontruct complex-valued image and apply the method to phase contrast imaging.

Mark Murphy¹, Michal Zarrouk², Kurt Keutzer², and Michael Lustig^2
1Google, Mountain View, CA, United States, ²EECS, UC Berkeley, Berkeley, CA, United States

We present a fast, autotuned, Gridding-based non-uniform FFT library with parallel implementions on CPUs and GPUs for reconstructing from non-Cartesian data. The influence of a nuFFT implementation and parameter selection on the resulting runtime is non-trivial. Our auto-tuning approach empirically selects an optimal implementation per trajectory by searching over algorithms and parameters, and saves it for future reconstructions (i.e. parallel imaging). We show that the optimal implementation depends also on the target platform and the sampling pattern itself. We also present a heuristic for near-optimal selection when exhaustive search is prohibitively expensive.

Jesse I. Hamilton¹, Katherine L. Wright¹, Kestutis Barkauskas¹, Vikas Gulani², and Nicole Seiberlich^1
1Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, ²Radiology, Case Western Reserve University, Cleveland, OH, United States

Previous work has shown that through-time non-Cartesian GRAPPA can reconstruct images with high radial acceleration factors. Here we demonstrate that 3D through-time radial GRAPPA can reconstruct data accelerated not only in-plane but also through-plane to yield high spatiotemporal resolutions. Reconstruction was performed with a 3D kernel, and calibration data were amassed using fully-sampled data and repeating the kernel through-k-space, through-time, and through-partitions to calculate accurate non-Cartesian GRAPPA weights. Time-resolved renal MR angiography data with total acceleration R=12 (R=6 in-plane, R=2 through-plane) were reconstructed at 1.45x1.45x3.00 mm³ spatial resolution and 2 s/frame temporal resolution (after retrospectively undersampling partitions).

Markus Weiger¹, David Otto Brunner¹, Martin Tabbert², Matteo Pavan¹, Thomas Schmid¹, and Klaas P. Pruessmann^1
1Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland, ²Bruker BioSpin MRI GmbH, Ettlingen, Germany

MRI of short T2 samples can be very efficiently performed with zero echo time (ZTE) imaging. ZTE data are incomplete in the k-space centre due to the initial RF dead time Δ, which can be addressed by algebraic reconstruction. The ZTE approach has been used for imaging T2s of several hundreds of µs. Targeting even shorter T2, however, increases the bandwidth and in turn the relative ∆, and images with large ∆ generally exhibit low-frequency artefacts. Therefore, the spatial response of ZTE as a function of ∆ is investigated and the practical bandwidth limits are explored by simulations and experiments.

Marcel Gutberlet¹, Mario Zeller², Frank Wacker¹, and Herbert Köstler^2
1Institute of Radiology, Medical School Hannover, Hannover, Lower Saxony, Germany, ²Institute of Radiology, University of Würzburg, Würzburg, Bavaria, Germany

In the theory of SENSE imaging the reconstruction matrix is defined by choosing a set of desired voxel functions. The resulting voxel functions are either generated in the strong voxel approach (SVA) by a least squares minimization or in the weak voxel approach (WVA) by constraining the orthonormality relation to the desired voxel functions. In k-space density weighted imaging for the chosen desired voxel function the signal-to-noise ratio (SNR) is maximized by applying an optimized k-space sampling. In this work, the WVA and SVA of SENSE imaging are evaluated in respect of SNR efficiency depending on the desired voxel functions.

Gengsheng Lawrence Zeng¹ and Edward V.R. DiBella^1
1Radiology, University of Utah, Salt Lake City, UT, United States

A non-iterative Bayesian reconstruction algorithm is derived to reconstruct dynamic undersampled MRI images. The k-space is radially sampled and 24 lines are acquired at each time frame. The Bayesian constraint uses the combination of immediately-before, current, and immediately-after data (referred to as the secondary data) to assist the image reconstruction. Unlike the ad hoc HYPR-type methods, the proposed algorithm is analytically derived and is able to track the object motion. The secondary data must be pre-filtered with a ramp filter before a small fraction of it is added to the current data for image reconstruction, with a modified ramp filter.

Michael Carl¹, James H. Holmes², and Graeme C. McKinnon^3
1GE Healthcare, San Diego, CA, United States, ²GE Healthcare, Madison, WI, United States, ³GE Healthcare, Waukesha, WI, United States

In radial out sequences, isotropic undersampling the number of k-space spokes is a popular way to accelerate the acquisition time. In 3D Cones, the trajectories inherently lie on conical surfaces and therefore allow k-space trajectories to support asymmetric FOVs and undersampling. Here we investigated the image quality tradeoffs for asymmetric undersampling under constrain of a fixed total same scan time. We found that increasing the undersampling along the symmetry axis of the Cones surfaces (z-axis) increases the artifact appearance, while reduced artifacts were observed when the overall undersampling was performed preferentially in the x-y plane.

Florian Knoll¹, Kai Tobias Block², Kristian Bredies³, Clemens Diwoky¹, Leon Axel⁴, Daniel K. Sodickson⁴, and Rudolf Stollberger^1
1Institute of Medical Engineering, Graz University of Technology, Graz, Styria, Austria, ²Center for Biomedical Imaging, NYU Langone Medical Center, New York, NY, United States, ³Department of Mathematics and Scientific Computing, University of Graz, Graz, Styria, Austria, ⁴Center for Biomedical Imaging, New York University School of Medicine, New York, NY, United States

Iterative parallel-imaging methods are highly promising for MR image reconstruction from undersampled data due to their flexibility to incorporate a priori knowledge using regularization. However, these methods are computationally very expensive and memory demanding. Consequently, most implementations so far used acquisition schemes that allow separating the reconstruction into smaller sub-problems, e.g. by reconstructing 3D volumes slice by slice. This comes at the expense of loosing acceleration capability in this direction, which limits the achievable overall scan efficiency. Furthermore, for certain imaging techniques like 3D radial ultra-short TE (UTE) imaging, separation of the reconstruction is not feasible. In this work, we present a method that treats the whole 3D imaging volume as single data set. This enables completely arbitrary 3D trajectories with acceleration in any dimension and incorporation of fully 3D regularization functionals.

Xiaodong Ma¹, Feng Huang², Chun Yuan^1,3, George Randy Duensing⁴, and Hua Guo^1
1Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China, ²Philips Healthcare, Beijing, China,³Department of Radiology, University of Washington, Seattle, WA, United States, ⁴Philips Healthcare, Gainesville, FL, United States

Since using all the calibration signals from images with different contrasts potentially provides more information of coil sensitivities, we propose to use multi-contrast information to enhance JSENSE. The same coil sensitivities as well as multiple contrast images are jointly reconstructed in the new model. Preliminary results demonstrated that the multi-contrast JSENSE algorithm with more accurate initialization results in images with improved quality, while costs no more computational time than original JSENSE algorithm

Meng Liu¹, Yunmei Chen¹, Yuyuan Ouyang¹, Xiaojing Ye², Xiaodong Ma³, and Feng Huang^4
1Department of Mathematics, University of Florida, Gainesville, FL, United States, ²School of Mathematics, Georgia Institute of Technology, Atlanta, GA, United States,³Center for Biomedical Imaging Research, Department of Biomedical Engineering, Tsinghua University, Beijing, China, ⁴Philips Healthcare, Shanghai, China

Partially parallel imaging has been used routinely for many MR applications. SENSE is one of the most commonly used methods, theoretically resulted in the optimal signal-to-noise ratio. However, SENSE reconstruction is highly depending on the accuracy of coil sensitivity maps. Several iterative methods were proposed to jointly reconstruct image and estimate sensitivity maps, have demonstrated the improved accuracy of coil sensitivity maps and the SENSE reconstruction quality. However, they suffer two numerical problems. One is the sensitivity to choosing regularization parameters; the other is the high computational cost. The target of this work is to tackle these two existing issues.

Rita Schmidt¹, Bikash Baishya¹, Noam Ben-Eliezer², Amir Seginer¹, and Lucio Frydman^1
1Chemical Physics, Weizmann Institute of Science, Rehovot, Israel, ²Center for Biomedical Imaging, New York University, New York, NY, United States

Recent studies described the potential of single shot methods based on spatiotemporal encoding (SPEN) principles. SPEN displays a higher robustness to frequency offsets. An important step still needed to endow SPEN scanning involves incorporating into the sequence the parallel imaging capabilities –without compromising the achievements of the SPEN super resolution reconstruction. The present work demonstrates the use of multi-band chirp pulses, to simultaneously encode multiple partial field-of-views. This approach is combined with a super-resolved SENSE-based method to reconstruct the full FOV image. The performance of the scanning and the reconstruction was explored in phantoms and in human imaging at 3T.

Eric A. Borisch¹, Paul T. Weavers¹, and Stephen J. Riederer^1
1Mayo Clinic, Rochester, MN, United States

Improving the performance of 2D-accelerated 3D acquisitions by determining an individualized acceleration selection, (R_Y, R_Z) for SENSE or (R_Y, R_Z,Delta) for CAIPIRINHA, for each patient/coil combination is an area of active research. Determining what is the optimal choice of accelerations is a compute- and time-intensive step. To enable a clinical practice-compatible (fast) process, we have implemented a GPU-based optimization routine, enabling a wide range of potential acceleration choices to be examined in under 10 seconds. The process of converting these calculations to a GPU as well as observed performance improvements are discussed.

Jiexun Xu¹, Nicolas Pannetier², and Ashish Raj^3
1Department of Computer Science, Cornell University, Ithaca, New York, United States, ²Department of Radiology and Department of Veterans Affairs Medical Center, University of California at San Francisco, San Francisco, CA, United States, ³Department of Radiology, Weill Medical College of Cornell University, New York, NY, United States

Among recent parallel imaging techniques, a Bayesian method that uses Cartesian under-sampling and sophisticated edge-preserving priors (EPP) have demonstrated its success in clinical applications. Recent compressive sensing related methods have proposed random under-sampling schemes that makes denoising and removing aliasing artifacts much easier. In this work we combine the strengths of both methods and propose a novel algorithm to solve the resulting problem, and demonstrate that our algorithm out performs popular existing methods.

Dengrong Jiang¹, Kui Ying¹, and Feng Huang^2
1Engineering Physics, Tsinghua University, Beijing, Beijing, China, ²Philips Healthcare, Beijing, Beijing, China

Partially Parallel Imaging (PPI) has been widely used in clinical practice to accelerate acquisition, but at the cost of reduced Signal-to-Noise Ratio (SNR). Often, regularization schemes are used to preserve SNR. However, existing regularization schemes have the difficulty to balance SNR and the preservation of boundaries and fine structures, especially when the acceleration factor is high. In this work, we adopted non-local sparse as the regularization term for PPI and achieved better balance of SNR and fine structure preservation compared with CS-SENSE. Low errors image was reconstructed at acceleration factor as high as 8 with an 8-channel head coil.

Claudio Santelli^1,2, Tobias Schaeffter¹, and Sebastian Kozerke^1,2
1Imaging Sciences and Biomedical Engineering, King’s College London, London, United Kingdom, ²Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland

A modified coil-sensitivity based calibration operator was incorporated into non-Cartesian CG-like SPIRiT. While maintaining image quality, significant reduction in reconstruction time has been demonstrated for simulated spiral and radial data. In addition, the exchangeability of the two consecutive k-space interpolation steps with a diagonal matrix multiplication has been shown. Depending on the number of k-space samples, reconstruction times on the order of the highly optimized NUFFT gridder are achieved.

Xi Peng^1,2, Leslie Ying³, Xin Liu^1,2, and Dong Liang^1,2
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, Guangdong, China, ³Department of Biomedical Engineering, Department of Electrical Engineering, The State University of New York at Buffalo, Buffalo, New York, United States

The problem of reconstructing a high-spatial-temporal-resolution MR image sequence occurs in various MR applications, such as interventional imaging, dynamic contrast enhanced imaging, cardiac imaging, where a static reference image can be obtained with relative ease before the whole dynamic process. This work addresses the problem by integrating the generalized series (GS) model, in which the reference prior is incorporated, with standard compressed sensing (CS) and parallel imaging (PI) techniques. The proposed method is validated in a Monte-Carlo study and is shown to provide superior imaging quality with decreased g-factor to existing CS and PI based reconstruction methods.

Martin Uecker¹ and Michael Lustig^1
1Electrical Engineering and Computer Sciences, University of California, Berkeley, Berkeley, California, United States

The purpose of this work to develop memory-efficient implementations of iterative reconstruction algorithms, such as SENSE, nonlinear inversion, and ESPIRiT. Although iterative reconstruction provides faster imaging and improved image quality, high computational demand currently limits its application for large 3D data sets. Graphical processing units can reduce computation time, but do not have enough memory for conventional algorithms. The proposed technique exploits the typical structure of iterative reconstruction algorithms to divide the computation into small independent blocks to reduce memory consumption.

Jonathan R. Polimeni¹, Himanshu Bhat², Thomas Benner³, Thorsten Feiweier⁴, Souheil J. Inati⁵, Thomas Witzel¹, Keith A. Heberlein⁶, and Lawrence L. Wald^1,7
1A.A. Martinos Center for Biomedical Imaging, Department of Radiology, Harvard Medical School, Massachusetts General Hospital, Charlestown, MA, United States, ²Siemens Medical Solutions, Charlestown, MA, United States, ³Siemens AG, Healthcare Sector, Erlangen, Bavaria, Germany, ⁴Siemens AG, Erlangen, Bavaria, Germany, ⁵National Institute of Mental Health, Bethesda, MD, United States, ⁶Siemens Healthcare USA, Charlestown, MA, United States, ⁷Harvard-MIT Division of Health Sciences and Technology, MIT, Cambridge, MA, United States

Conventional auto-calibration data acquisition strategies for accelerated EPI seek to match the echo spacing of the image data, and therefore employ a segmented EPI acquisition. This approach is vulnerable to patient movement or respiration-induced dynamic B0 changes. Here we introduce a new auto-calibration method based on acquiring multi-shot, multi-slice auto-calibration data where the multiple segments in a given slice are acquired sequentially in time, shortening the time interval between segments. Our results show that the proposed method provides higher temporal SNR and reduced motion sensitivity that the conventional approach, while maintaining the echo spacing of the image acquisition.

Christian Binter¹, Rebecca Ramb², Bernd Jung², and Sebastian Kozerke^1,3
1Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland, ²Dept. of Radiology, Medical Physics, University Medical Center, Freiburg, Germany,³Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom

Spatiotemporal undersampling methods allow for significant speed-up of image acquisition. This work aims at providing an analytical assessment of noise behavior and temporal fidelity of k-t SENSE and k-t PCA. The g-factor formalism introduced for parallel imaging (SENSE, GRAPPA) is extended to the temporal frequency domain. Using in-vivo data it is demonstrated that the proposed g-factor shows good agreement with results obtained by pseudo-replica analysis for both k-t SENSE and k-t PCA, and also matches the temporal transfer function for k-t SENSE.

Eric Y. Pierre¹, Nicole Seiberlich¹, Vidya Nadig², and Mark A. Griswold^3
1Biomedical Engineering, Case Western Reserve University, Cleveland, Ohio, United States, ²Heart and Vascular Center, MetroHealth Campus of Case Western Reserve University, Cleveland, Ohio, United States, ³Radiology, Case Western Reserve University, Cleveland, Ohio, United States

Dictionaries with spatial or temporal information shared with a target image are often used as a sparsifying transform to improve Compressed Sensing reconstruction results. However in accordance to Parallel Imaging principles, the coil sensitivity information within these dictionaries could also be directly exploited to reconstruct undersampled images with a light computational load. A new reconstruction scheme which exploits image-block dictionaries extracted from a calibration data is proposed. In vivo reconstruction results are shown for a dynamic cardiac experiment with radial acquisition accelerated by a factor 8.

Marcel Gutberlet¹, Frank Wacker¹, and Herbert Köstler^2
1Institut of Radiology, Medical School Hannover, Hannover, Lower Saxony, Germany, ²Institute of Radiology, University of Würzburg, Würzburg, Bavaria, Germany

In the generalized theory of SENSE imaging the reconstruction matrix is defined by choosing a set of desired voxel functions. Either in the strong voxel approach (SVA) the resulting voxel functions are generated by a least squares minimization or in the weak voxel approach (WVA) by constraining the orthonormality relation to the desired voxel functions. In this work, the reconstruction accuracy of the SVA and WVA of SENSE imaging depending on the choice of the desired voxel functions is evaluated.

Suhyung Park¹ and Jaeseok Park^1
1Department of Brain and Cognitive Engineering, Korea University, Seoul, Seoul, Korea

Combination of partially parallel imaging (PPI) and compressed sensing (CS) [1-4] employs complementary properties of the two competitive methods. Among them, direct combination approaches [1-2], which jointly consider both CS and PPI constraints, potentially suffer from image artifacts at high acceleration, because sparsifying transform are less coherent with sensitivity encoding than Fourier encoding. Then, combination of CS and PPI in a sequential fashion [3-4] was recently introduced, demonstrating its feasibility in overcoming the aforementioned problems. In this work, we develop a novel, multi-scale subband weighted PPI algorithm, wherein 1) CS is utilized to yield multi-scale sparse solutions, 2) Subbands in each scale are employed to produce multiple de-noised filtered k-spaces, 3) Join estimation of PPI convolution kernels and k-spaces are performed, considering both inter-subband correlation and spatial correlation over multiple coils.

Rebecca Ramb¹, Christian Binter², Felix A. Breuer³, Gerrit Schultz¹, Maxim Zaitsev¹, Sebastian Kozerke^2,4, and Bernd Jung^1
1Dept. of Radiology, Medical Physics, University Medical Center, Freiburg, Germany, ²Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland,³MRB Research Center, Würzburg, Germany, ⁴Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom

The goal of this work is the analytical quantification of the noise enhancement in k-t-GRAPPA reconstructions from time-resolved k-t-accelerated acquisitions. Image reconstruction in x-f-space by transforming the three-dimensional convolution kernel is explained, the g-factor with respect to temporal frequencies is introduced and the method is validated in in-vivo-data by statistical noise estimation using the pseudo-replica method. In k-t-reconstruction temporal filtering is observed. The g-factor quantification in x-f-space for k-t-GRAPPA further reflects this and thus constitutes a measure for noise enhancement as well as the amount of temporal filtering introduced in the reconstruction process.

Jun Liu¹, Axel Loewe¹, Michael O. Zenge², Alban Lefebvre¹, Edgar Mueller², and Mariappan S. Nadar^1
1Imaging and Computer Vision, Siemens Corporation, Corporate Technology, Princeton, NJ, United States, ²MR Application & Workflow Development, Siemens AG, Healthcare Sector, Erlangen, Bavaria, Germany

Parallel imaging exploits the difference in sensitivities between individual coil elements in a receive array to reduce the number of gradient encodings required for imaging. SENSE and GRAPPA are two representative approaches. In this abstract, a new approach, called Elora is proposed. Elora implicitly uses coil sensitivities by estimating a low rank subspace from the calibration data, and then works by enforcing the low rank constraint on the sliding blocks of the k-space data.

Jingyuan Lyu¹, Yuchou Chang², and Leslie Ying^1
1Department of Biomedical Engineering, Department of Electrical Engineering, The State University of New York at Buffalo, Buffalo, NY, United States, ²Department of Electrical Engineering and Computer Science, University of Wisconsin-Milwaukee, Milwaukee, WI, United States

In GRAPPA, the computational time increases with the number of channels and the amount of ACS data. To address this issue, different from the existing approaches that compress the large number of physical channels to fewer virtual channels, we propose to use random projections to reduce the dimension of the problem in the calibration step. Experimental results show that randomly projecting the data onto a lower-dimensional subspace yields results comparable to those of traditional GRAPPA, but is computationally less expensive.

Mai T. Le¹, Sathish Ramani¹, and Jeffrey A. Fessler^1
1EECS, University of Michigan, Ann Arbor, MI, United States

SENSE reconstruction for parallel MRI with random undersampling requires spatial regularization for improved image quality. Compressed sensing methods utilize sparsity promoting regularizers that demand computation intensive, non-linear optimization algorithms. Previous variable splitting based algorithms ignored prior information that patients are not rectangular. We formulate a regularized SENSE reconstruction that explicitly includes a support constraint in the problem formulation. We propose a specific variable splitting strategy that when combined with the augmented Lagrangian framework and alternating minimization yields an algorithm with simple, efficient, non-iterative update steps. Experiments with in-vivo data demonstrate the improved performance of this method.

Il Yong Chun¹ and Thomas Talavage^1,2
1School of Electrical and Computer Engineering, Purdue University, West Lafayette, Indiana, United States, ²Weldon School of Biomedical Engineering, Purdue University, West Lafayette, Indiana, United States

We propose two pre-computation-allowable and non-iterative MAP SENSE reconstruction algorithms based on 1) a Gaussian Random Field (GRF) with non-zero mean and 2) a Huber-Markov Random Field (HMRF) with non-zero mean. Simulation results show that the non-iterative HMRF MAP regularization technique is more effective for edge preservation and residual aliasing artifact reduction than non-iterative GRF MAP and Tikhonov-type regularization methods.

Il Yong Chun¹ and Thomas Talavage^1,2
1School of Electrical and Computer Engineering, Purdue University, West Lafayette, Indiana, United States, ²Weldon School of Biomedical Engineering, Purdue University, West Lafayette, Indiana, United States

Here, we present a pre-computation-allowable sparse Tikhonov-regularized SENSE MRI reconstruction technique based on QR decomposition, fast regularization parameter estimation using a new L-curve , and sparse matrix representation. The simulation results show that it significantly reduces residual aliasing artifacts and noise amplification for ill-posed cases.

Dara Bahri¹, Martin Uecker¹, and Michael Lustig^1
1Electrical Engineering and Computer Sciences, University of California, Berkeley, Berkeley, California, United States

While receiver arrays with many channels can increase parallel imaging acceleration and provide high signal-to-noise, processing the large datasets they produce is computationally demanding. Coil compression algorithms reduce, and denoise in the process, data from many coils into fewer virtual ones. Huang et al. proposed using principal component analysis to globally compress multi-coil k-space data. Zhang et al. developed an improved technique for Cartesian sampling by compressing locally along fully-sampled directions, but the method suffers in low-SNR sections of k-space. In this work we present an algorithm that compresses locally while remaining noise-robust.

S. Lalith Talagala¹, Joelle E. Sarlls¹, and Souheil J. Inati^2
1NMRF/NINDS, National Institutes of Health, Bethesda, MD, United States, ²FMRIF/NIMH, National Institues of Health, Bethesda, MD, United States

Current EPI based fMRI protocols frequently incorporate accelerated parallel acquisition techniques such as GRAPPA and SENSE . These techniques help to reduce EPI distortions and to increase the number of slices per TR. However, we have observed that the temporal SNR of GRAPPA EPI data can be highly inhomogeneous and highly compromised with certain EPI protocols. In this work, we show that the tSNR of GRAPPA accelerated EPI can be made more spatially uniform and enhanced by using a GRAPPA reference scan based on a FLASH acquisition scheme rather than an EPI acquisition scheme.

Yuchou Chang¹, Dong Liang², and Leslie Ying^3
1Electrical Engineering, University of Wisconsin - Milwaukee, Milwaukee, Wisconsin, United States, ²Paul C. Lauterbur Research Centre for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, Guangdong, China, ³Department of Biomedical Engineering, Department of Electrical Engineering, State University of New York at Buffalo, Buffalo, New York, United States

To reduce the reconstruction complexity in parallel imaging, principal component analysis (PCA) has been used to compress large array coils into a new set of fewer virtual channels. In this study, a novel kernel (nonlinear) PCA approach is proposed to achieve noise suppression and channel reductions simultaneously. Using GRAPPA as the reconstruction method, experimental results demonstrate that the reconstruction from channels compressed by the proposed kernel PCA method has a higher SNR than those compressed by PCA or uncompressed conventional GRAPPA, while the proposed method takes almost the same computation time as the PCA method.

Lawrence Dougherty¹, Walter R.T. Witschey², Robert M, King¹, and Gamaliel Isaac^1
1Radiology, University of Pennsylvania, Philadelphia, PA, United States, ²Surgery, University of Pennsylvania, Philadelphia, PA, United States

An iterative GRAPPA method has been developed for use on non-Cartesian sampled data. The method uses repeated application of Cartesian GRAPPA interpolation following gridding. Optimal kernel size as well as multiple kernels are investigated. Using a radially undersampled data set, GRAPPA was compared to compressed sensing. The iterative GRAPPA approach is simple to implement and executes rapidly.

Steen Moeller¹, Edward J. Auerbach¹, Junqian Xu¹, Christophe Lenglet¹, Kamil Ugurbil¹, and Essa S. Yacoub^1
1Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, Minnesota, United States

The sensitivity of multiband separation to changes in the relative phase between simultaneously acquired slices is demonstrated. A phase estimation procedure is proposed and tested for a diffusion weighted acquisition. With a phase-updating strategy reduced signal fluctuations allows for improved detectability.

Hilde Segers¹, Willem Jan Palenstijn¹, Kees Joost Batenburg^1,2, and Jan Sijbers^1
1Vision Lab - dept. Physics, Universiteit Antwerpen, Antwerpen, Antwerpen, Belgium, ²Center for Mathematics and Informatics, Amsterdam, Noord-Holland, Netherlands

Segmentation refers to the classification of image pixels into distinct classes, typically based on their grey level. It is usually performed as a post-processing step on an MR magnitude image, which is influenced by reconstruction artifacts. In this abstract, we investigate the integration of reconstruction and segmentation into one single procedure. This combination is a regularized reconstruction problem where we exploit prior knowledge about the discreteness of the grey levels. Simulation results show that this integrated method yields better results than the conventional approach when the underlying truth is a discrete image.

Frederik Testud¹, Daniel Gallichan², Kelvin J. Layton³, Anna M. Welz¹, Christoph Barmet⁴, Chris A. Cocosco¹, Jürgen Hennig¹, Klaas P. Pruessmann⁴, and Maxim Zaitsev^1
1Medical Physics, University Medical Center Freiburg, Freiburg, Germany, ²Centre d'Imagerie BioMédicale, Ecole Polytéchnique Fédérale de Lausanne, Lausanne, Switzerland,³Electrical & Electronic Engineering, University of Melbourne, Parkville, Victoria, Australia, ⁴Institute for Biomedical Engineering, University and ETH Zürich, Zürich, Switzerland

Single shot multidimensional imaging using linear and non-linear enocoding fields auch as 4-Dimensional Radial In/Out or North West Echo Planar Imaging allow to combine the preservation the variability of spatial resolution and to avoid the total resolution loss in the center of the field of view. A field camera consisting of 16 field probes is used for trajectory calibration. Analytically derived Maxwell terms of the used PatLoc head insert gradient coil are used to choose an adequate set of basis functions to further improve the image quality.

Jason P. Stockmann^1,2, Clarissa Zimmerman³, Matthew S. Rosen^2,4, and Lawrence L. Wald^4
1Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Charlestown, Massachusetts, United States, ²Department of Physics, Harvard University, Cambridge, MA, United States, ³Electrical Engineering, Massachusetts Institute of Technology, Cambridge, MA, United States, ⁴Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Charlestown, MA, United States

Recently several encoding strategies have been proposed which use nonlinear spatial encoding magnetic fields (SEMs) to perform projection imaging, exploiting iterative matrix solvers for reconstruction. In the present work, we consider the case of generalized spatial encoding using a rotating multipolar field in the transverse plane, showing how a linear offset field and multiple receive coils can break the symmetry of an arbitrary nonlinear SEM, providing encoding throughout the FOV. This flexible encoding/reconstruction approach relaxes the need for a homogenous B0 field and linear gradient fields, opening the door to new, unconventional MR imaging systems.

Matthias Schloegl¹, Florian Knoll¹, Katharina Gruber¹, Franz Ebner², and Rudolf Stollberger^1
1Institute for medical Engineering, TU Graz, Graz, Austria, ²Universitätsklinik für Radiologie, Medizinische Universität Graz, Graz, Austria

This study investigates about the capability of image metrics as objective tool for quality evaluation of non-linear reconstruction methods in the context of compressed sensing. Metric rating of reconstructions with TGV, IRGN-TV, l1-SPIRiT and CGSENSE with Cartesian, radial and random sub-sampled trajectories was compared to that of six experienced radiologists, focusing on overall image quality and recognizability of anatomy and pathology. Datasets from several body regions affected by identified pathologies were selected. Good correlations were found for metrics based upon models of the human visual system as well as for common image metrics when calculated for a region of interest.

Jeffrey Adam Kasten^1,2, François Lazeyras², and Dimitri Van de Ville^1,2
1Institute of Bioengineering, Ecole Polytechnique Fédérale de Lausanne, Lausanne, VD, Switzerland, ²Department of Radiology and Medical Informatics, Université de Genève, Geneva, GE, Switzerland

Model-based MRSI reconstruction often relies upon structural MR images to characterize the sample by specifying spectrally-homogenous compartments. However, either spatio-spectral disparities between the two modalities or model mismatch will lead to additional artifacts. We therefore consider a more data-driven approach in which the raw MRSI data itself is used to estimate the generating signal model, employing a general framework predicated on principal component analysis and spatial regularization. Phantom experiments show that our method can yield highly resolved spatial and spectral components, while simultaneously surmounting a number of limitations associated with traditional Fourier reconstruction.

Owen L. Kaluza¹, Amanda C. L. Ng^2,3, David K. Wright^4,5, Leigh A. Johnston^5,6, John Grundy⁷, and David G. Barnes^2
1Monash e-Research Centre, Monash University, Clayton, Victoria, Australia, ²Monash Biomedical Imaging, Monash University, Clayton, Victoria, Australia, ³Department of Electrical & Electronic Engineering, The University of Melbourne, Parkville, Victoria, Australia, ⁴Centre for Neuroscience, The University of Melbourne, Parkville, Victoria, Australia,⁵Florey Institute of Neuroscience and Mental Health, Parkville, Victoria, Australia, ⁶NeuroEngineering Laboratory, Dept. Electrical & Electronic Engineering, The University of Melbourne, Parkville, Victoria, Australia, ⁷Centre for Complex Software Systems and Services, Swinburne University of Technology, Hawthorn, Victoria, Australia

Diffusion-guided quantitative susceptibility mapping (QSM) is a new technique that promises improved mapping without the need for multiple-orientation (COSMOS) image acquisitions. However, the computation time for realistic image sizes on central-processing unit (CPU)-based supercomputers is prohibitively expensive. We have analysed the dQSM algorithm and developed an OpenCL-based implementation that runs on graphics processing unit (GPU)-based compute clusters. Our implementation yields identical results to the parallel CPU code, in drastically less time. Dual-GPU cluster nodes can compute the dQSM map 8 - 10 times faster when their GPUs are used compared to their multi-core CPUs. With this work, use of dQSM in research imaging facilities becomes practicable on quite modest computational facilities.

Patrick Virtue¹, Martin Uecker¹, Michael Elad², and Michael Lustig^1
1Electrical Engineering and Computer Sciences, University of California, Berkeley, Berkeley, California, United States, ²Computer Science, Technion - Israel Institute of Technology, Haifa, Israel

The results of under-sampling reconstruction algorithms are often compared to a fully-sampled reconstruction. This comparison is overly optimistic because even if the reconstruction removes the aliasing due to under-sampling, we would still have an inherent loss of SNR due to the reduced acquisition time. We present a process to predict image quality for a given reconstruction technique and under-sampling pattern. Using this prediction as a “gold standard” enables a fair comparison for reconstruction results and provides an efficient means of quickly assessing reconstruction algorithms and parameters.

Elad Bergman¹, Arie Yeredor¹, and Uri Nevo^1
1Tel-Aviv University, Tel Aviv, Israel

This work shows the potential of post-processing estimation for SNR improvement in Unilateral NMR, aiding the use of such devices in bio-medical applications. We present a novel post-processing method to improve the SNR of the acquired signal in unilateral NMR scanners. We estimate the signal parameters from the noisy data with the weighted least square approach, and exploit more efficiently the inherently known characteristics of the NMR signal. The method was first develop and tested for T2 measurements with a CPMG-like sequence. Then, using a similar concept we further developed this method to improve the SNR of lateral slice–selective imaging scans.

Wanyong Shin¹, Erik B. Beall¹, and Mark J. Lowe^1
1Radiology Dept., Cleveland Clinic, Cleveland, Ohio, United States

Simultaneous excitation of multiple slices using multi-band (MB) radio-frequency (RF) excitation echo planar imaging (EPI) has shown potential to increase the spatial and temporal resolution. The Slice-GRAPPA method calculates a linear interpolation kernel for each slice of MB accelerated k-space and de-aliases images using the estimated kernel. It is known that the interpolation kernel size, acceleration factor (R value) and the number of reference lines (ACS line number) determine the reconstructed image quality in in-plane parallel imaging (PI) technique. Since Slice-GRAPPA employs the same basic principle as the GRAPPA technique, we hypothesize that the size of the interpolation kernel and data used to estimate the kernel could affect the performance of MB de-aliasing. In this study, we evaluate MB de-aliasing performance while varying the sizes of interpolation kernel size and fitted data using simulation and show considerable reductions in compute time are possible with no apparent loss in data quality.

Tzu-Chao Chuang¹, Hing-Chiu Chang², Hsuan-Hung Huang¹, and Ming-Ting Wu^3,4
1Electrical Engineering, National Sun Yat-Sen University, Kaohsiung, Taiwan, Taiwan, ²Brain Imaging and Analysis Center, Duke University, Durham, NC, United States, ³School of Medicine, National Yang-Ming University, Taipei, Taiwan, Taiwan, ⁴Radiology, Kaohsiung Veteran General Hospital, Kaohsiung, Taiwan, Taiwan

View-sharing PROPELLER (VS-Prop) with a pixel-based optimal blade selection (POBS) algorithm has been proposed to shorten the acquisition window during dynamic contrast imaging, leading to a higher spatiotemporal resolution or/and spatial coverage. In this work, flow phantom experiment was performed to evaluate the temporal and spatial accuracy. In addition, the result of first-pass dynamic contrast-enhanced (DCE) cardiovascular imaging on healthy volunteers was also presented.

Souheil J. Inati¹, Michael Schacht Hansen², and Peter Kellman^2
1National Institute of Mental Health, National Institutes of Health, Bethesda, MD, United States, ²National Heart Lung and Blood Institute, National Institutes of Health, Bethesda, MD, United States

We present an algorithm for adaptive combination of images from an array of MR coils that is suitable for applications in which complex valued images are required. This algorithm enforces smoothness in both the magnitude and phase of the estimated coil sensitivities and overcomes limitations inherent in previous methods.

Andrzej Jesmanowicz^1
1Biophysics, Medical College of Wisconsin, Milwaukee, Wisconsin, United States

A correction method is presented that removes spatial artifacts related to magnetic field drift in MRI systems equipped with ferro-shim elements usually placed close to high power gradient systems. At high duty cycle of gradients the heat generated inside the magnet bore reduces the effectiveness of shim elements by shifting the temperature closer to the Curie point. In response to unequal initial radial distribution of these elements the new magnetic field is no more uniform than it was at room temperature. Higher order corrections are required to remove artifacts from the time-series of phase images.

Qiu Wang¹, Jun Liu¹, Michael O. Zenge², Nirmal Janardhanan¹, Edgar Mueller², and Mariappan S. Nadar^1
1Imaging and Computer Vision, Siemens Corporation, Corporate Technology, Princeton, NJ, United States, ²MR Application & Workflow Development, Siemens AG, Healthcare Sector, Erlangen, Bavaria, Germany

Parallel imaging achieves scan time reduction by utilizing the correlation among an array of receiver coils to reconstruct image from under-sampled data. For SENSE-type reconstruction, explicit estimation of the coil sensitivity map (CSM) is critical in achieving good image quality. When the data is acquired using 3D Cartesian sampling, one way for estimating the coil profiles is to decouple the data along the frequency encoding direction and then perform the estimation in a 2D manner, which, however, ignores the correlation between the calibration data in the frequency encoding direction. In this work, we propose a 3D CSM estimation method, extending the Eigen-Vector approach that was developed for the 2D scenario. Experiments show the coil profiles are smoother with proposed approach compared to the 2D estimation method, and good image quality has been achieved.

Satoshi Ito¹ and Yoshifumi Yamada^1
1Research Division of Intelligence and Information Sciences, Utsunomiya University, Utsunomiya, Tochigi, Japan

We have proposed a new image reconstruction technique in which images of an optional scale can be obtained and hence alias-free images can be reproduced from a single piece of data by applying a quadratic phase modulation to a Fourier imaging technique. Almost all the aliasing artifacts are removed, however, small aliasing artifacts that comes from the higher frequency components contained in the signal but not contribute to the down-scaled image are remained. In this paper, we propose a new alias-free reconstruction technique in which remaining aliasing artifact are reduced using SENSE-like algorithm.

Hien Nguyen¹ and Gary H. Glover^1
1Department of Radiology, Stanford University, Palo Alto, CA, United States

In high resolution functional MRI, it is often desirable to reduce the readout duration to make the acquired data less prone to T2* susceptibility artifacts at the expense of SNR. This can be achieved by undersampling k-space. However, the conventional Fourier transform-based reconstruction method suffers from undersampling artifacts such as high-frequency ringing and loss of resolution. In this work we propose a new imaging approach to fMRI with under-sampled data by exploiting the generalized series constraint in the penalized maximum-likelihood framework. The effectiveness of the method is characterized and illustrated by experiments at 3T.

Dingxin Wang^1,2, Joseph Vu², Essa S. Yacoub², Kamil Ugurbil², and Vibhas Deshpande^3
1Siemens Medical Solution USA, Inc., Minneapolis, MN, United States, ²Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN, United States,³Siemens Medical Solutions USA, Inc., Austin, TX, United States

Our study demonstrates the feasibility of using slice accelerated SE sequence for acquiring high resolution T2-weighted images with whole brain coverage (72 slices, 2mm thickness). The slice acceleration and SAR reduction are the keys to enable increased coverage of the T2-weighted image acquisition within a reasonable time (8 minutes).

Wanyong Shin¹, Erik B. Beall¹, and Mark J. Lowe^1
1Radiology Dept., Cleveland Clinic, Cleveland, Ohio, United States

To accelerate 2D EPI, various methods have been proposed. One such promising method, simultaneous excitation of multiple slices using multi-band (MB) radio-frequency (RF) excitation, has caught the attention of many researchers because the number of excited slice simultaneously, (the MB factor), accelerates the acquisition’s possible temporal resolution by the MB factor. While the MB technique is expected to be beneficial for resting state and task fMRI studies, it has not be thoroughly evaluated yet. We hypothesize that the kernel size choice will produce non-negligible effect on MB-accelerated EPI image reconstruction, altering the result of rs-fMRI and task-fMRI. In this study, we evaluate rs-fMRI and fMRI analysis, and investigate the kernel size sensitivity of MB reconstruction in the simulation. Finally, we demonstrate rs-fMRI and fMRI analysis using MB EPI scans.

Seong Dae Yun¹ and Nadim Jon Shah^1,2
1Institute of Neuroscience and Medicine - 4, Forschungszentrum Jülich, Jülich, Germany, ²JARA - Faculty of Medicine, RWTH Aachen University, Aachen, Germany

The relatively high imaging speed of EPI has led to its widespread use in dynamic MRI studies. For even faster acquisition of multiple slices in EPI, M-EPI (Multiplexed EPI) method has been recently presented (Feinberg et al.). However, because of the intrinsically low bandwidth in the phase encode direction, EPI-based methods are highly sensitive to field inhomogeneities, which results in potentially severe geometric distortions. This problem becomes more challenging at ultra-high fields such as 9.4T. The present work verifies i) the use of M-EPI at 9.4T and ii) demonstrates the application of the PSF (Point Spread Function)-based correction method to the M-EPI images to remove the geometric distortions.

Rita Schmidt¹ and Lucio Frydman^1
1Chemical Physics, Weizmann Institute of Science, Rehovot, Israel

Recent studies based on spatiotemporal encoding (SPEN) principles showed that many of the echo planar imaging (EPI - the leading “ultrafast” experiment) challenges can be alleviated, particularly if implemented in a “full-refocusing” mode. The present work extends this imaging modality by introducing a variety of new 3D multi-slice schemes, employing either inversion or –for lower Specific Absorption Rate– stimulated echo pulses, and are timed to fulfill the demands of full-refocusing. The work confirmed the advantages of these new imaging modalities with in vivo animal and human scans performed at 3T and 7T, demonstrating higher robustness of the SPEN comparing EPI.

Noam Ben-Eliezer¹, Lucio Frydman², and Daniel K. Sodickson^1
1Bernard and Irene Schwartz Center for Biomedical Imaging, Department of Radiology, New York University School of Medicine, New York, NY, United States, ²Department of Chemical Physics, Weizmann Institute of Science, Rehovot, Israel

Recently a new MR acquisition scheme has been emerging, based on progressive point-by-point refocusing in the object’s spatial, rather than frequency domain, through the use of quadratic phase encoding. This technique, termed Spatiotemporal-Encoding (SPEN), is capable of overcoming sizable B₀ and B₁ field distortions, performing single-shot chemical-shift imaging (CSI), and produces reliable images under conditions that preclude the use of conventional acquisition schemes. This work presents a comprehensive analysis of SPEN’s parametric framework and signal-to-noise ratio (SNR) as compared to conventional k-space encoding, showing – theoretically, and experimentally – that the SNR of these two encoding techniques is comparable.

Rita Schmidt¹ and Lucio Frydman^1
1Chemical Physics, Weizmann Institute of Science, Rehovot, Israel

“Single-shot” strategies –foremost among them Echo Planar Spectroscopic Imaging – alleviate the time constrain of the high dimensionality in the spectroscopic imaging. Although a powerful aid, EPSI’s echo trains face limitations related the rapidly oscillating gradients. Among EPSI’s alternatives is a recent proposal using spatiotemporal encoding (SPEN) principles. The acquired data in such experiment carries the spatial information, and its phase modulation that can convey the chemical shift offsets. The present study recover such additional chemical shift dimension using extended super resolution method that deals with the previous limitations and proves it in a in-vivo 7T animal and 3T human imaging.

Joost van Gorp¹, Job G. Bouwman¹, Chris J.G. Bakker¹, and Peter R. Seevinck^2
1Image Sciences Institute, UMC Utrecht, Utrecht, Utrecht, Netherlands, ²UMC Utrecht, Utrecht, Utrecht, Netherlands

Spectroscopic imaging can be used as a means to acquire geometrically accurate images and characterize signal decay curves in the presence of off-resonance effects, without the need for additional post-processing corrections. To decrease the acquisition time for this inherently slow technique and increase its use for in vivo research, compressed sensing was used to prospectively accelerate the sequence up to a factor 5. Retrospective error evaluation of the reconstruction showed that complex averaging of temporal information, which is available at no additional cost in this sequence, decreased the reconstruction error.

Chuan Huang¹, Joyita Dutta¹, Yoann Petibon¹, Timothy G. Reese², Quanzheng Li¹, Ciprian Catana², and Georges El Fakhri^1
1Center for Advanced Radiological Sciences, Department of Radiology, Massachusetts General Hospital, Boston, MA, United States, ²AA Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, United States

Accurate lesion characterization is crucial for initial staging, follow up and assessment of response to treatment in non-small cell lung cancer. Conventional PET-CT suffers from a temporal mismatch between PET and CT, due to longer PET than CT acquisitions needed for achieving good PET SNR. This mismatch yields artifacts that affect PET diagnostic specificity as well as quantitation accuracy. Gated CT can be used to correct these motion artifacts; however, however, this is not usually done routinely due to longer exams and greater radiation dose. In this work, we present a slice-interleaved golden-angle radial FLASH sequence with short-duration slice-projection navigation for lung respiratory motion measurement and respiratory motion tracking for motion-corrected PET reconstruction using simultaneous PET-MR.

Dan Zhu¹, Zhengwei Zhou², Feng Huang³, and Kui Ying^4
1Department of Biomedical Engineering, Tsinghua, Beijign, China, ²Department of Biomedical Engineering, Tsinghua, Beijing, China, ³Philips Healthcare, Beijing, China, ⁴Department of Engineering Physics, Tsinghua, Beijing, China

The speed of imaging remains one of the basic challenges in Magnetic Resonance Imaging. A large amount of work on fast imaging was focused on studying how to reconstruct an image from a fixed undersampling design. Much less work was done on how to choose the sampling design at the first place. Since trajectory optimization is usually computationally expensive, in this work, we focus on studying the reusability of optimized trajectory. This topic was studied from the following 4aspects: whether the optimized trajectory can be shared among images of different contrasts, different subjects, different slices and different coils.

Yong Pang¹ and Xiaoliang Zhang^1,2
1Radiology & Biomedical Imaging, University of California San Francisco, San Francisco, CA, United States, ²UC Berkeley/UCSF Joint Graduate Group in Bioengineering, Berkeley & San Francisco, CA, United States

In this project, we combined the parallel imaging with the interpolated compressed sensing (iCS) method to further accelerate the imaging speed for multi-slice 2-dimensional parallel MR imaging. The raw data of each slice from each channel is multiplied by a weighting function and then used to estimate the missed k-space data of the neighboring slice from the same array channel, which helps improve the image quality of the neighboring slice. In-vivo MR of human has been used to investigate the feasibility of the proposed method, showing obviously increased SNR and CNR.

Rudolph Pienaar^1,2, Kiho Im^3,4, Mathieu Dehaes^3,4, Sara Smith⁵, Barbara Peysakhovic⁵, Bryce Becker⁵, Nora M. Raschle⁵, Patricia Ellen Grant^2,6, and Nadine Gaab^4,5
1Radiology, Children's Hospital Boston, Boston, MA, United States, ²Radiology, Harvard Medical School, Boston, MA, United States, ³Newborn Medicine, Children's Hospital Boston, Boston, MA, United States, ⁴Pediatrics, Harvard Medical School, Boston, MA, United States, ⁵Developmental Medicine, Children's Hospital Boston, Boston, MA, United States,⁶Radiology, Boston Children's Hospital, Boston, MA, United States

We present a curvature-histogram analysis as robust biomarker for characterizing development dyslexia from typically developing subjects.

Yongxia Zhou¹, Yao Wang², Damon Kenul³, Yuanyi Xue², Yulin Ge³, Joseph Reaume³, Robert I. Grossman⁴, and Yvonne W. Lui^3
1Radiology/Center for Biomedical Imaging, New York University Langone Medical Center, New York, NY, United States, ²Electrical & Computer Engineering, Polytechnic Institute of New York University, Brooklyn, NY, United States, ³Radiology/Center for Biomedical Engineering, New York University Langone Medical Center, New York, NY, United States,⁴Radiology/Center for Biomedical Engineering, New York University, New York, NY, United States

The purpose of this study is to design and develop computational techniques to identify mild traumatic brain injury (MTBI) patients that can be used to help predict patient long-term outcome ultimately using multi-dimensional feature space based on several advanced quantitative MR measures. Fourteen imaging features (e.g. kurtosis, magnetic field correlations, thalamic network connectivity and regional volumetry), and nineteen clinical features were tested with different feature selection and classifier algorithms. Our study demonstrates that an automatic classification based on objective physical and imaging measures can achieve a high accuracy of nearly 100% and a robust prediction for the long-term outcome (P¡Ü0.01).

Ali R. Khan¹, Maged Goubran¹, Sandrine de Ribaupierre², and Terry Peters^1
1Robarts Research Institute, London, ON, Canada, ²Clinical Neurological Sciences, Western University, London, ON, Canada

Pre-surgical localization of the epileptogenic zone is challenging with conventional techniques and often diagnostic scans are found to be negative. Quantitative MRI techniques such as relaxometry and diffusion tensor imaging can potentially detect subtle abnormalities through comparison against a healthy control population atlas. Two approaches for this technique, surface-based and voxel-based, are evaluated in performing patient-specific analyses using quantitative T1, T2 relaxometry, and DTI metrics (FA and MD). We show that both methods produce comparable results and can reveal abnormalities in patients with negative MRI findings.

Stephen J. Sawiak^1,2, A Jennifer Morton³, and T. Adrian Carpenter^1
1Wolfson Brain Imaging Centre, University of Cambridge, Cambridge, England, United Kingdom, ²Behavioural and Cognitive Neuroscience Institute, University of Cambridge, Cambridge, England, United Kingdom, ³Department of Pharmacology, University of Cambridge, Cambridge, England, United Kingdom

Support vector machines are used to detect whether high-resolution brain images are from healthy or transgenic Huntington's disease mice. We found that with leave-one-out cross validation the classifier has >98% accuracy at detecting the sick animals and we applied the same trained classifier to older healthy brains, revealing that the changes seen are not confused with healthy aging.

David W. Shattuck¹, Anand A. Joshi², Justin P. Haldar², Chitresh Bhushan², Soyoung Choi³, Andrew C. Krause¹, Jessica L. Wisnowski^4,5, Arthur W. Toga¹, and Richard M. Leahy^2
1Laboratory of Neuro Imaging, University of California, Los Angeles, Los Angeles, CA, United States, ²Signal and Image Processing Institute, University of Southern California, Los Angeles, CA, United States, ³Dana and David Dornsife Cognitive Neuroscience Imaging Institute, University of Southern California, Los Angeles, CA, United States, ⁴Brain and Creativity Institute, University of Southern California, Los Angeles, CA, United States, ⁵Radiology, Children's Hospital of Pittsburgh of UPMC, Pittsburgh, PA, United States

We describe a collection of software tools for jointly processing and visualizing structural and diffusion MRI of the brain. T1-weighted brain MRI are processed to extract models of the cortical surface. A brain atlas labeled with anatomical ROIs is registered to the subject data using a combined surface/volume registration procedure. Diffusion weighted images are processed to produce fiber tract models. The structural and diffusion results are combined to generate a brain connectivity map based on the set of anatomical ROIs. These tools can be applied using scripts or through a user interface that provides sophisticated interactive processing and visualization capabilities.

David S. Smith¹, Stephanie Barnes¹, and Thomas E. Yankeelov^1
1Vanderbilt University, Nashville, TN, United States

We present preliminary efforts that indicate that the Kolmogorov-Smirnov statistical test may be an extremely useful method for automatically separating signal from noise in dynamic imaging data, especially when aliased power should be captured but noise should be ignored. We compare to Otsu's method and demonstrate an improved automatic classification of signal and noise in in vivo tumor-bearing mouse data.

Chris J. Rose¹, James P. O'Connor^1,2, Tim F. Cootes¹, Chris J. Taylor¹, Gordon C. Jayson³, Geoff J. M. Parker¹, and John C. Waterton^1,4
1Center for Imaging Sciences, Manchester Academic Health Science Center, The University of Manchester, Manchester, Greater Manchester, United Kingdom, ²Department of Radiology, The Christie, Manchester, Greater Manchester, United Kingdom, ³Department of Medical Oncology, The Christie, Manchester, Greater Manchester, United Kingdom,⁴AstraZeneca, Alderley Park, Cheshire, United Kingdom

MRI can spatially map biophysical and physiological parameters across organs and tumors, and is often used in natural history studies and preclinical and clinical trials of novel drugs. In such research, there is a need to draw statistical inferences about the population, based on a sample. It is common to perform hypothesis tests by taking averages over each structure, but spatially heterogeneous differences can attenuate statistical power. We compare the recently-proposed indexed distribution analysis (IDA) to the conventional and histogram analysis approaches using well-controlled simulated and clinical data. IDA has several advantages over conventional and histogram analysis methods.

Robert Allan Brown¹, Sridar Narayanan¹, and Douglas L. Arnold^1
1Montreal Neurological Institute, McGill University, Montreal, QC, Canada

Magnetization transfer ratio (MTR) is a magnetic resonance imaging technique that can be used to measure changes in myelin. However, different MTR sequences produce data on different scales, making multi-site or long duration longitudinal studies difficult. We propose a non-linear normalization method to map scanner-specific MTR values to a standard scale. We compare this method with an existing linear normalization method and uncorrected data.

Artem V. Mikheev¹, Henry Rusinek¹, and Graham Wiggins^1
1Radiology, NYU Langone Medical Center, New York, NY, United States

MR signal intensity, especially at high field strength, is affected by inhomogeneity, or shading artifacts, manifested as a smooth spatially varying signal intensity distortions. Correction of this effect is the key to successful implementation of all MR image analyses, including segmentation, registration and functional modeling of dynamic data. We have developed a new method BiCal (Bias Calculation) for non-uniformity correction that uses 3D Canny Edge detection and 3D Legendre polynomials to represent the bias field.

Adrian Martin^1,2, Antonio Marquina³, Juan Antonio Hernandez-Tamames^1,2, Pablo Garcia-Polo^1,2, and Emanuele Schiavi^2,4
1Electronics, Rey Juan Carlos University, Mostoles, Madrid, Spain, ²Alzheimer's Project, Queen Sofia Foundation - CIBERNED, Madrid, Madrid, Spain, ³Applied Mathematics, Valencia University, Burjassot, Comunidad Valenciana, Spain, ⁴Applied Mathematics, Rey Juan Carlos University, Mostoles, Madrid, Spain

In some MRI applications, in particular when co-registration between modalities is needed, such as fMRI and 3DT1-IR, the acquired image needs to be upsampled to a higher resolution so common interpolation methods have been typically applied to increase this new apparent spatial resolution. Here we propose a new Super Resolution (SR) technique which outperforms these interpolation methods. It is based in a variational SR model proposed and validated in MRI by Joshi et al. in which we introduced the concept of the Total Generalized Variation. Using this operator the solutions obtained by the proposed method present a better image quality. A comparison between methods is presented with phantom and real brain MR images.

Kejia Cai¹, Anup Singh¹, Mohammad Haris¹, Ravi Prakash Reddy Nanga¹, Ranjit Ittyerah¹, Damodar Reddy¹, Harish Poptani¹, Hari Hariharan¹, and Ravinder Reddy^1
1University of Pennsylvania, Philadelphia, PA, United States

This conventional quantification method may be confounded by the intrinsic magnetic transfer ratio (MTR) asymmetry as well as the Nuclear Overhauser Effect (NOE) effect. To decouple these confounding effects, we demonstrate a comprehensive Lorentzian fitting of Z spectra in brain tumor for the quantification of multiple slow-exchanging metabolites contributing to Z spectrum acquired with low RF saturating amplitude. Results show an increased amide proton transfer (APT) integral and a decreased CEST integral at 2ppm in tumor compared to normal brain tissue. The novel Z spectral analysis method may be used for differentiating tumor metabolic profile from normal tissue.

Lindsey Alexandra Crowe¹, Azza Gramoun¹, Wolfgang Wirth², Frank Tobalem³, Kerstin Grosdemange⁴, Jatuporn Salaklang⁵, Anthony Redgem⁵, Alke Petri-Fink⁶, Felix Eckstein², Heinrich Hofmann⁷, and Jean-Paul Vallée^1
1Radiology / Faculty of Medicine, Geneva University Hospital, Geneva, Switzerland, ²Institute of Anatomy and Musculoskeletal Research, Paracelsus Medical University, Salzburg, Austria, ³Radiology, Centre Hospitalier Universitaire Vaudois, Lausanne, Switzerland, ⁴Faculty of Medicine, University of Geneva, Geneva, Switzerland, ⁵Adolphe Merkle Institute, Université de Fribourg, Fribourg, Switzerland, ⁶Adolphe Merkle Institute and Chemistry Departement, Université de Fribourg, Fribourg, Switzerland, ⁷Institute of Materials, Powder Technology Laboratory, EPFL, Lausanne, Switzerland

We present quantification of inhomogeneous SPION uptake over a 3D volume using dUTE and customized software in a rat AIA model at 3T. The use of SPION as a contrast agent is leading development of image acquisition and analysis techniques to quantify uptake and persistence. As signal from iron oxide uptake in vivo can be inhomogeneous, quantification of both the volume and signal intensity are needed for the true extent of SPION uptake. Positive contrast dUTE imaging is used due to its concentration dependence of signal intensity for iron oxide containing regions along with validation of a semi-automated segmentation technique.

Jason Langley¹, Daniel Huddleston², Sinyeob Ahn¹, Xiangchuan Chen¹, Christopher Barnum³, and Xiaoping P. Hu^1
1Biomedical Engineering, Emory University & Georgia Institute of Technology, Atlanta, GA, United States, ²Neurology, Emory University, Atlanta, GA, United States, ³Physiology, Emory University, Atlanta, GA, United States

In this abstract, we present a novel landmark-based method for segmentation of the locus coeruleus. The method is then tested on datasets from non-pathologic subjects.

Pierre-Yves Baudin^1,2, Noura Azzabou^3,4, Pierre G. Carlier^3,4, and Nikos Paragios^1,5
1Center for Visual Computing, Ecole Centrale Paris, Châtenay-Malabry, IDF, France, ²SIEMENS Healthcare, Saint-Denis, IDF, France, ³Institute of Myology, Paris, IDF, France,⁴I2BM, MIRCen, IdM NMR Laboratory, CEA, Orsay, IDF, France, ⁵Equipe Galen, INRIA Saclay, Palaiseau, IDF, France

Developing an automatized tool for segmenting the different skeletal muscles in MRI with minimum user intervention is of paramount importance to facilitate muscle studies. Segmentation of skeletal muscles in 3D MRI poses some specific issues: non-discriminative appearance of the muscles, partial contours between them, large inter-subject variations, spurious contours due to fat infiltrations. We propose an automatic segmentation method based on the Random Walks algorithm to which we add a prior shape model based on learning from an annotated data set.

Suresh Anand Sadananthan^1,2, Bhanu Prakash K.N.³, Melvin K-S Leow^1,4, ChinMeng Khoo⁵, Kavita Venkataraman², Eric Khoo Yin Hao⁵, Lee Yung Seng^1,6, Peter Gluckman¹, Tai E Shyong^1,5, and Sendhil S. Velan^1,7
1Singapore Institute for Clinical Sciences, A*STAR, Singapore, ²Department of Obstetrics & Gynaecology, National University of Singapore, Singapore, ³Singapore Bioimaging Consortium, A*STAR, Singapore, ⁴Department of Endocrinology, Tan Tock Seng Hospital, Singapore, ⁵Department of Medicine, National University of Singapore, Singapore,⁶Department of Pediatrics, National University of Singapore, Singapore, ⁷Clinical Imaging Research Centre, A*STAR-NUS, Singapore

Obesity is associated with increased insulin resistance, a risk factor for type 2 diabetes or cardiovascular disease. Accumulation of fat in different depots may have different effects on insulin resistance. Visceral adipose tissue (VAT) is thought to have greater impact on insulin resistance than subcutaneous fat. More recently, it has been observed that subcutaneous adipose tissue (SAT) has two sub-compartments, deep SAT (DSAT) and superficial SAT (SSAT)), which may have different effects on insulin resistance. While there are many automated methods to accurately segment SAT and VAT, there is currently no technique to separate DSAT and SSAT. We have implemented a fully automated approach to segment DSAT and SSAT and evaluated it on normal and overweight Chinese adults.

Radu Mutihac^1,2, Allen Braun³, and Thomas J. Balkin^2
1Department of Physics, University of Bucharest, Bucharest, Bucharest-Magurele, Romania, ²Psychiatry & Neuroscience, Department of Behavioral Biology, Walter Reed Army Institute of Research, Silver Spring, Maryland, United States, ³Language Section, NIDCD / National Institutes of Health, Bethesda, Maryland, United States

Sleep data acquisition was carried out using uncommon low bandwidth in the phase encoding direction and low gradient amplitude for functional EPI readout resulting in low scanner noise, but substantially increasing the static magnetic field inhomogeneity. EPI images are prone to substantial signal dropout and spatial distortion in regions where the field is inhomogeneous. Two major sources of artifacts affect EPI reconstructions: Nyquist ghosts and geometric distortion. Post-processing optimization of geometric distortion correction in EPI FSE images based on paired field maps (magnitude and phase at different echo times)acquired at 3T in terms of spin-echo time difference is presented hereafter.

Anahita Fathi Kazerooni^1,2, Leila Torbati³, Mahrooz Malek³, and Hamidreza Saligheh Rad^1,2
1Quantitative MR Imaging and Spectroscopy Group, Research Center for Cellular and Molecular Imaging, Tehran University of Medical Sciences, Tehran, Iran, ²Medical Physics and Biomedical Engineering, Tehran University of Medical Sciences, Tehran, Iran, ³Imaging Center, Imam Khomeini Hospital, Tehran University of Medical Sciences, Tehran, Iran

Typically, quantification of DCE-MRI of ovary is susceptible to errors caused by motion artifacts and intensity inhomogeneity induced by bias fields. Motion artifacts and bias fields introduce signal intensity variations in the images that must be resolved from intensity changes caused by the passage of contrast agent. Thus, registration of DCE-MRI image sequence is a challenging issue. In this work, we proposed a solution to the misregistration problem of DCE-MR images, by exploiting residual complexity (RC) similarity measure, to account for complex intensity variations in a non-rigid registration approach and for precise quantification of DCE-MRI to characterize ovarian masses.

Eric M. Bultman¹, Ethan K. Brodsky², Debra E. Horng³, Pablo Irarrázabal⁴, William R. Schelman⁵, Walter F. Block^1,3, and Scott B. Reeder^1,6
1Biomedical Engineering, University of Wisconsin-Madison, Madison, WI, United States, ²Medical Physics, University of Wisconsin - Madison, Madison, WI, United States, ³Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, ⁴Electrical Engineering, Pontificia Universidad Catolica de Chile, Santiago, Chile, ⁵Medicine, University of Wisconsin-Madison, Madison, WI, United States, ⁶Radiology, University of Wisconsin-Madison, Madison, WI, United States

Estimating quantitative hepatic perfusion parameters is challenging for several reasons, including the need to acquire data during periods of free breathing. In this work, we demonstrate the feasibility of estimating hepatic perfusion parameters using interrupted DCE-MRI data acquired with a 3D time-resolved radial imaging sequence during sequential breath-holds. Average time-signal curves corresponding to cirrhotic liver and hepatocellular carcinoma ROIs are fitted to a dual-input single-compartment model, and quantitative perfusion parameters are generated. Perfusion characteristics of HCC are shown to be distinctly different from background cirrhotic liver.

Valentin Hamy¹, Shonit Punwani¹, Jesica Makanyanga¹, Stuart Taylor¹, and David Atkinson^1
1Centre for Medical Imaging, University College London, London, United Kingdom

Dynamic Contrast Enhanced MRI is of interest for the detection of small bowel disorders including ulcerative lesions in Crohn’s disease. However misalignments arise due to patient motion during the acquisition. This is likely to alter the time intensity curve shape of a given region of interest and can affect data analysis. Butylscopolamine injection prior to acquisition can limit the effect of bowel peristalsis but correcting for breathing motion in the presence of contrast changes remains challenging. In this study we investigate the application of a registration approach, robust to contrast changes, to obtain accurate realignment of clinically relevant features in small bowel DCE-MRI

Tobias C. Wood¹, David J. Lythgoe¹, and Steven C.R. Williams^1
1Neuroimaging, King's College London, Institute of Psychiatry, London, United Kingdom

We have modified the Brain Extraction Tool (BET) so that it can successfully produce brain masks for rodent images. Previously the algorithm failed due to the different size and shape of rodent brains compared to humans. We have selected a more appropriate initial brain shape and search constants and demonstrate results with rat data.