Brian Andrew Hargreaves¹ and Karla L. Miller^2
1Radiology, Stanford University, Stanford, CA, United States, ²FMRIB Centre, Oxford University, Oxford, United Kingdom

The extended-phase-graph (EPG) formalism, which allows efficient simulation of the signal from numerous commonly-used pulses sequences, is summarized intuitively, with example applications and software. EPG treats a group of spins within a voxel undergoing integer numbers of cycles of gradient dephasing by using a Fourier basis that reduces a large spin ensemble to a limited number of states. Effects of relaxation, RF nutation, gradient dephasing and rephasing, diffusion and static dephasing are easily included. Links to Matlab code with intuitive examples of how to build up pulse sequence simulations are provided.

Tilman Johannes Sumpf¹ and Markus Untenberger^1
1Biomedizinische NMR Forschungs GmbH am Max-Planck-Institut fuer biophysikalische Chemie, Goettingen, Germany

Especially for development and testing of new reconstruction and post processing algorithms, the Matlab environment became a popular choice in the MRI community. However, the visualization and comparison of multidimensional complex MRI data arrays is still rather uncomfortable with standard Matlab routines. To overcome these limitations, we introduce a freely available tool which has been designed specifically for MRI data analysis. This e-poster gives a quick tour through the concepts and features of the tool and, therewith, offers practical solutions to some common MRI data visualization problems in Matlab.

Dattesh D. Shanbhag¹, Sheshadri Thiruvenkadam¹, Sandeep Kaushik¹, Gaspar Delso², Scott D. Wollenweber³, Sonal Ambwani³, Rakesh Mullick¹, and Florian Wiesinger^4
1GE Global Research, Bangalore, Karanataka, India, ²GE Healthcare, Glattbrugg, Zurich, Switzerland, ³GE Healthcare, Waukesha, WI, United States, ⁴GE Global Research, Garching b. Munchen, Bavaria, Germany

MR-based PET attenuation correction (AC) is a prerequisite for quantitative PET and a key determining factor for the success of PET/MR. RF shading with phased array coils results in segmentation based MR-AC map generation failure. In this work we present a novel approach for MR-AC map generation within the phase field based framework based on joint estimation/correction of the RF shading and tissue segmentation maps using Dixon MRI images. The method provides for parameter variation resilient body contour and tissue class segmentation, obviates the need to “re-tune” the algorithm for specific cohort of data acquisition and coils and results in simplified workflow for PET-MR attenuation map generation.

Audrey Peiwen Fan¹, Berkin Bilgic¹, Louis O. Gagnon^2,3, Thomas Witzel³, Himanshu Bhat³, Bruce R. Rosen^2,3, and Elfar Adalsteinsson^1,3
1Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, United States, ²Health Sciences and Technology, Harvard-MIT, Cambridge, MA, United States, ³Radiology, Athinoula A. Martinos Center for Biomedical Imaging, Charlestown, MA, United States

Comprehensive venograms which map quantitative oxygenation along each vessel are shown for the first time. Susceptibility maps are created from MR phase images and cerebral veins are graphed into a node-edge representation. Quantitative oxygen saturation (SvO₂) along each vessel is then displayed onto a 3D mesh generated from the graph to create oxygenation venograms. We probe the dependence of reconstructed SvO₂ in vivo as a function of vein tilt angle relative to the main field, and found increased SvO₂ at angles undersampled by the dipole imaging kernel. This observation motivates future work to incorporate vessel angle priors from graphing for more accurate model-based reconstruction of SvO₂ venograms.

Yudong Zhang^1,2, Chuanmiao Xie^1,3, Bradley S. Peterson^1,2, and Zhengchao Dong^1,2
1Columbia University, New York, NY, United States, ²New York State of Psychiatric Institute, New York, NY, United States, ³Department of Medical Imaging & Interventional Radiology, Sun Yat-Sen University, Guangzhou, Guangdong, China

We proposed a novel hybrid system to classify an MR brain image as either normal or abnormal. The method employed digital wavelet transform to extract features and used principal component analysis to reduce the dimensionality of the feature space. Afterwards, we constructed a kernel support vector machine with Radial Basis Function kernel, using particle swarm optimization to optimize the parameters in the training function. We tested the method with a dataset of 90 brain images consisting of 17 diseases. A 5-fold cross validation showed that our method achieved 97.78% classification accuracy, higher than 86.22% by BP-NN and 91.33% by RBF-NN.

Venkata Veerendranadh Chebrolu¹, Dattesh D. Shanbhag¹, Arun Govinda Rao², Patrice Hervo³, Marc-Antoine Labeyrie⁴, Catherine Oppenheim^4,5, and Rakesh Mullick^1
1Medical Image Analysis Lab, GE Global Research, Bangalore, Karnataka, India, ²GE Healthcare, Bangalore, Karnataka, India, ³GE Healthcare, Buc, France,⁴Departments of Radiology and Neurology, Centre Hospitalier, Sainte-Anne, Paris, France, ⁵Université Paris Descartes, Paris, France

In semi-automated stroke lesion segmentation based on user defined seed inputs, the location and shape of the input could vary. We developed robust and reproducible semi-automated DWI and PWI lesion segmentation algorithms and evaluated their performance and reproducibility in assessing perfusion-diffusion mismatch in a cohort of acute ischemic stroke patients. Repeated measures ANOVA did not show any statistically significant differences between ground-truth lesion volumes and those obtained using the semi-automated methods (p < 0.05). Mismatch agreement was achieved in 88% of the cases with a kappa (κ) of 0.766.

Ana-Maria Oros-Peusquens¹, Sandra M. Meyers², Alex L. MacKay², and Nadim Jon Shah^1,3
1INM-4, Research Centre Jülich, Jülich, Germany, ²Department of Physics and Astronomy, University of British Columbia, Vancouver, B.C., Canada, ³Faculty of Medicine, JARA, RWTH Aachen University, Aachen, Germany

It is increasingly recognised that myelin is ubiquitous in determining MR contrast in the living brain, especially at high fields. Investigating its distribution is not only useful in the study of several neurodegenerative diseases but can also be used for the in vivo parcellation of the brain. The current “gold standard” method for characterisation of myelin content in vivo is based on the investigation of the water trapped between myelin layers. Separation of the myelin water and tissue water pools, which have different mobility, is possible with an NNLS-based analysis of T2 decay curves. We introduce here a method which allows for discrimination between the two water pools based on their T2* properties. The data are acquired at 3T with either a 2D or a 3D multiple-echo gradient echo sequence and analysed with NNLS. Evaluation of the raw data (no additional resampling or image filtering) allows for separation of myelin (T2* < 20ms) and tissue water (T2*> 20ms) in the majority of white matter voxels. The average value of brain myelin water fraction for 12 volunteers is found to be 6.4%. Several improvements in data analysis are suggested. The method is proposed as an alternative to (high-SAR) spin-echo based myelin water analysis for use with high-field systems.

Takao Goto¹ and Hiroyuki Kabasawa^1
1Global Science Laboratory, GE Healthcare, Hino-shi, Tokyo, Japan

We present a new method of the automated bolus tracker positioning for MRI liver scans. Bolus tracker is used to monitor bolus signal to know the arrival of the bolus. Placing the bolus tracker inside the aorta accurately with looking at scout images is complicated task and one of the factors making operator�fs workflow worse. In proposed method, the aorta was detected using AdaBoost by searching aorta around spine. 234 axial images of 64 volunteers�f datasets were tested and showed 98.3 % success rate of the aorta detection. Automated size of the tracker will be necessarily towards practical use.

Niranchana Manivannan¹, Bradley Dean Clymer¹, Anna Bratasz^2,3, and Kimerly A. Powell^2,3
1Department of Electrical and Computer Engineering, The Ohio State University, Columbus, OH, United States, ²Small Animal Imaging Shared Resources, The Ohio State University, Columbus, OH, United States, ³Department of Biomedical Informatics, The Ohio State University, Columbus, OH, United States

The goal of this research is to explore the effect on image quality of including an additional LR view to the orthogonal super resolution geometry. The visual evaluation of this technique was performed in low resolution data sets simulated from isotropic high resolution data. The results of simulation indicate that the quality of an SRR image based on orthogonal acquisition may be improved by the addition of a fourth view acquired obliquely to the through-plane direction. Additionally, the observed increase in image quality would be worth the minimal increase in acquisition time required for one additional view.

Ruth Oliver¹, Michael A. Chappell^2,3, David Thomas¹, and Xavier Golay^1
1Institute of Neurology, University College London, London, United Kingdom, ²Institute of Biomedical Engineering, University of Oxford, Oxford, United Kingdom, ³FMRIB Centre, University of Oxford, Oxford, United Kingdom

Partial volume effects due to low spatial resolution are known to introduce errors in quantification of perfusion estimates using ASL. This is particularly problematic in patients where atrophy is present and cortical thinning occurs. A comparison is made for single-TI data of an existing method that employs adaptive spatial priors, and a linear regression approach that assumes constant perfusion for each tissue over a specified kernel area. Both methods offer good correction, increasing GM perfusion by a factor of 1.7 on average, although the adaptive spatial prior method introduces less smoothing and better preserves detail.

Mohit Saxena¹, Namita Aggarwal², Bharti Rana², S. Senthil Kumaran³, Ramesh Kumar Agrawal², and Madhuri Behari^1
1Department of Neurology, All India Institute of Medical Sciences, New Delhi, Delhi, India, ²School of Computer & Systems Sciences, Jawaharlal Nehru University, New Delhi, Delhi, India, ³Department of NMR, All India Institute of Medical Sciences, New Delhi, Delhi, India

Support vector machine (SVM) is used to distinguish PD from controls in terms of white matter changes in substantia nigra, thalamus and/ or combination of both areas from the T1-weighted MR images. We extracted voxels from white tissue probability maps of substantia nigra and thalamus region. The performance of the decision system was evaluated in terms of sensitivity, specificity and accuracy. Experimental results demonstrate the importance of white matter change in thalamus and effectiveness of SVM to automatically distinguish PD from controls.

Eli Gibson^1,2, Lanette J. Friesen-Waldner^2,3, Amanda M. Hamilton², Emeline J. Ribot², Trevor P. Wade^2,3, Curtis N. Wiens⁴, Kundan Thind^2,3, Jacqueline K. Harris³, Nica M. Borradaile⁵, Charles A. McKenzie^1,3, and Aaron D. Ward^1,2
1Biomedical Engineering, University of Western Ontario, London, Ontario, Canada, ²Robarts Research Institute, London, Ontario, Canada, ³Medical Biophysics, University of Western Ontario, London, Ontario, Canada, ⁴Physics and Astronomy, University of Western Ontario, London, Ontario, Canada, ⁵Physiology and Pharmacology, University of Western Ontario, London, Ontario, Canada

Validation of quantitative MRI for non-invasive quantification of liver fat ideally uses accurate registration to an accepted reference standard (e.g. histology). We demonstrate accurate co-registration of murine in vivo 3D whole-body water-only and fat-only images to stained histological liver images. Three 129/SVJ mice were fed a high fat diet to induce hepatic steatosis and obesity, imaged using 3D quantitative IDEAL MRI at 3T, then sectioned on a cryomicrotome, where 3D optical and histology images were collected. We successfully co-registered MR-optical images and optical-histology images interactively using thin-plate-spline transformations with mean target registration errors of 0.7 mm and 0.1 mm, respectively.

Tess E. Catherwood¹, Andrew J. Patterson¹, Kamarul Zaki², Mahesh Iddawela², Helena Earl², Carlos Caldas³, Karen Sayal⁴, Martin John Graves¹, and Fiona J. Gilbert^1
1Radiology, University of Cambridge, NHS Foundation Trust, Cambridge, United Kingdom, ²Oncology, Addenbrooke's Hospital and University of Cambridge, NHS Foundation Trust, Cambridge, United Kingdom, ³Cancer Research UK Cambridge Institute, Cambridge, United Kingdom, ⁴University of Cambridge, Cambridge, United Kingdom

Tumor heterogeneity is well recognized in breast cancer and is associated with differential responses to chemotherapy. Voxel-wise analysis of sequential DCE-MRI requires that images be spatially registered. Patients were studied with 3T MRI during neoadjuvant chemotherapy. Mutual information-based rigid and non-rigid registration was used to align dynamic and sequential volumes to account for motion artifacts and larger inter-session deformations. Semi-quantitative perfusion metrics were extracted and generation of spatially registered parameter maps allowed voxel-wise assessment of treatment response. This approach provides an insight into tumor microstructure and physiology and developments will ideally increase statistical sensitivity in differentiating responders and non-responders.

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

Analysis of real-time fMRI time series is subject to temporal dispersion of the hemodynamic response and aliasing of physiological noise. Echo-volumar imaging (EVI), inverse imaging (InI), highly undersampled projection imaging (PI), and compressed sensing (CS) imaging reconstruction enable temporal resolution down to 100 ms. Extremely short acquisition (TR) poses the problem of serial correlations among voxels studied here in the context of autoregressive (AR) models.

Thomas A. Depew¹ and Qing-San Xiang^1,2
1Physics & Astronomy, University of BC, Vancouver, BC, Canada, ²Radiology, University of BC, Vancouver, BC, Canada

3D MRI is becoming increasingly popular in clinical diagnostic imaging and interventional studies. In multi-slice acquisitions, the through-plane resolution is limited by the slice selection profile. We present here a simple and scalable technique to improve through-plane resolution in 3D multi-slice acquisitions that can be employed on any set of MRI data by acquiring a few extra datasets.

Renjie He¹, Joshua P. Yung¹, David T. A. Fuentes¹, R. Jason Stafford¹, and John D. Hazle^1
1Department of Imaging Physics, The University of Texas M. D. Anderson Cancer Center, Houston, Texas, United States

FFT Based phase unwrapping is a straight, flexible and fast method; however, it suffers from the phase distortion in practice. We tried to recover the phase unwrapping exactly by quantizing the dispersed integer filed into an integer field through Powell optimization. The exact phase unwrapping was obtained for phantom phase MR images. This approach is useful in quantitative analysis based on phase images.

Yu Ding¹, Rizwan Ahmad¹, Hui Xue², Lee C. Potter¹, Samuel T. Ting¹, Ning Jin³, and Orlando P. Simonetti^1
1The Ohio State University, Columbus, OH, United States, ²Corporate Research, Siemens Corporation, Princeton, NJ, United States, ³Siemens Healthcare, Columbus, OH, United States

SENSE is a widely used parallel imaging technique. The so-called g-factor represents how noise in the raw data affects the noise in the reconstructed image. However, the g-factor calculation does not take into account the noise in the channel sensitivity map. In this abstract, we present a noise transfer model in SENSE reconstruction that takes into account the noise in both the raw data and the channel sensitivity map. A phantom study showed that the model has satisfactory accuracy. We observed that a large portion (> 35%) of the image noise originates from the noise in the sensitivity map.

Congbo Cai¹, Jiyang Dong¹, Shuhui Cai¹, Jing Li¹, Lin Chen¹, and Zhong Chen^2
1Department of Communication Engineer and Electronic Science, Xiamen university, Xiamen, Fujian, China, ²Department of Communication Engineer and Electronic Science, Xiamen University, Xiamen, Fujian, China

Spatiotemporally-encoded single-scan MRI method is relatively insensitive to field inhomogeneity compared to EPI method. Conjugate gradient (CG) method has been used to reconstruct super-resolved images from the original blurred ones. In this article, a new de-convolution reconstruction method is proposed. Through removing the quadratic phase modulation from the signal acquired with spatiotemporally-encoded MRI, the signal can be described as a convolution of desired super-resolved image and a point spread function. The de-convolution method proposed herein not only is simpler and more efficient than the CG method, but also provides super-resolved images with better quality.

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

High resolution isotropic 3D MR imaging is often repeatedly performed for preoperative evaluation and postoperative follow-up. Usually only a little information is updated each time in a following examination compared to the previous one if neglecting the change of pose. an adaptive synthesis of the prior data and the new acquired data from separate 3D examinations is presented. It shows that the information synthesis can help to reduce residual aliasing artifacts in Compressed Sensing (CS) imaging with highly undersampled k-space.

Linda DeMello¹, Rajan Rakheja¹, Christopher B. Glielmi², Christian Geppert³, Hersh Chandarana⁴, and Kent Friedman^1
1Radiology, New York University School of Medicine, New York, New York, United States, ²MR R&D Collaborations, Siemens Healthcare, New York, New York, United States, ³MR R&D Collaborations, Siemens Medical Systems, New York, New York, United States, ⁴Radiology, NYU Langone Medical Center, New York, New York, United States

This study was performed to compare the accuracy of spatial registration of PET/CT and PET/MR in patients with FDG-avid metastatic lesions. Thirteen patients underwent PET/CT followed by PET/MR, and the spatial coordinates of the visually estimated centers of nineteen metastatic lesions were measured. Total distance between the isocenters was calculated, along with averages and standard deviations. Morphologic PET/MR sequences showed significantly more accurate spatial registration than PET/CT, likely secondary to simultaneous acquisition of PET and MR data. However, EPI-based DWI sequences demonstrated significant misregistration compared to PET/CT, likely due to respiratory motion and the inherent spatial distortion of DWI.

Daniel Hähn^1,2, Nicolas Rannou^1,2, Thomas J. Re^1,3, Patricia Ellen Grant^1,2, and Rudolph Pienaar^1,2
1Radiology, Boston Children's Hospital, Boston, MA, United States, ²Radiology, Harvard Medical School, Boston, MA, United States, ³Radiology, University of Milan, Milan, Italy

Recent advances in web browser technology enable fast loading and visualization of medical imaging data. We developed a parser for MR data in DICOM format which shows workstation-like performance while enabling access to such data across devices (smartphones, tablets, computers).

L. Martyn Klassen¹ and Ravi S. Menon^1
1Robarts Research Institute, The University of Western Ontario, London, ON, Canada

Increased use of multichannel coils and phase sensitive analysis requires coil combination techniques beyond the traditional square root of sum of squares. For multiple echo acquisitions used in B0 and quantitative susceptibility mapping, pixel wise singular value decomposition can be used to generate phase sensitive optimum, in a least squares sense, coil sensitivity and image magnetization estimates.

Tokunori Kimura¹ and Naho Imamura^1
1MRI development department, Toshiba medical systems corp., Otawara, Tochigi, Japan

We proposed and assessed a new partial Fourier technique named MagAFI (Magnitude-based Asymmetric Fourier Imaging ) where only magnitude image with 0-filling is used and also it is possible to combine with POCS technique. MagAFI introduced smaller artifacts than standard homodyne method and the POCS combination sepecially in the portion of large and spatially high-frequency phase, reflecting the difference of residual phase errors after phase correction. Our proposed MagAFI is practically useful algorithm from the views of balancing image quality and simplicity. If allowing us to use phase information and to take processing time, MagAFI combining POCS is further better to improve image quality and robustness.

Dan Xu¹, Joseph K. Maier¹, Kevin F. King¹, and R. Scott Hinks^1
1GE Healthcare, Waukesha, WI, United States

Uncompensated high order eddy currents (HOEC) can cause various image artifacts, including image distortion in echo planar imaging (EPI) and phase errors in phase contrast imaging. While a recently proposed method is capable of measuring and correcting long term HOEC to minimize distortion in EPI, it cannot provide accurate characterization of short term HOEC, which is believed to be one of the main sources of phase errors in phase contrast imaging. In this paper, we propose to extend the Duyn’s method to 3D to measure and characterize short term HOEC with a phantom.

Aiming Lu¹, Ian C. Atkinson¹, and Keith R. Thulborn^1
1Ctr Magnetic Resonance Research, University of Illinois, Chicago, IL, United States

Quantitative sodium MR imaging with flexible twisted projection imaging (flexTPI) provides valuable information about regional changes in tissue sodium concentration (TSC) in human brain tumors responsive to fractionated radiation treatment. The acquisition time of 10 minutes is at the upper limits for which a patient can be expected to remain stationary. Stimulations show that quantification of TSC is compromised by even small head motions without appreciable degradation of the visual impression of image quality. Navigators incorporated within the flexTPI sequence have been investigated as a means of detecting head motion during acquisition and then correcting the quantification of TSC.

Fa-Hsuan Lin^1,2, Yi-Cheng Hsu³, Panu Vesanen², Jaako O. Nieminen², Koos C. J. Zevenhoven², Juhani Dabek², Lauri T. Parkkonen², and Risto J. Ilmoniemi^2
1Institute of Biomedical Engineering, National Taiwan University, Taipei, Taiwan, ²Department of Biomedical Engineering and Computational Science, Aalto University, Espoo, Finland, ³Department of Mathematics, Nnational Taiwan University, Taipei, Taiwan

To mitigate the challenge of low SNR in ultra-low-field (ULF) MRI (typically B0 ~ 100 T), we exploit the simultaneous measurements from multiple superconductive quantum interference device (SQUID) sensors to enforce a linear dependency among local k-space data points across coils. Experimental data using 47 SQUID sensors demonstrate that this data consistency (DC) constraint can improve the ULF MRI peak SNR by 2 fold.

Murat Aksoy¹, Melvyn B. Ooi¹, Anh Tu Van¹, and Roland Bammer^1
1Center for Quantitative Neuroimaging, Department of Radiology, Stanford University, Stanford, CA, United States

A method is described to retrospectively correct rigid body motion artifacts in interleaved slice-selective acquisitions. Instead of the conventional slice-by-slice reconstruction, the proposed method relies on reconstructing the 3D volume using all the slices simultaneously using an iterative technique. Simulations demonstrate that, if the motion information is known, it is possible to correct through-plane motion in an interleaved slice-selective acquisition.

Christoph Kolbitsch¹, Claudia Prieto¹, Charalampos Tsoumpas¹, Christian Buerger¹, and Tobias Schaeffter^1
1Division of Imaging Sciences & Biomedical Engineering, King's College London, London, United Kingdom

Simultaneous PET-MR acquisition has recently been proposed to improve cancer diagnosis by combining metabolic information from PET-acquisitions with high-resolution MR-images. Since PET data is usually acquired in multi-stations, 3D MR-images need to be obtained during the PET scan. Bulk motion during acquisition can severely impair image quality of both techniques. Here we present an MR acquisition which acquires T1- and T2-weighted 3D images and allows for the estimation of bulk motion from the data itself. The motion information can be used to correct MR and PET-images. Our approach was successfully tested in healthy volunteers in real MR-acquisitions and PET simulations.

Ali Gholipour¹, Judy A. Estroff¹, and Simon K. Warfield^1
1Department of Radiology, Boston Children's Hospital and Harvard Medical School, Boston, MA, United States

This educational E-Poster provides an overview of the recently developed motion-robust super-resolution volumetric MRI reconstruction methods with exemplary application in fetal MRI. The mathematical details of the methods will be discussed and illustrated with visual examples.

Christoph Kolbitsch¹, Claudia Prieto¹, and Tobias Schaeffter^1
1Division of Imaging Sciences & Biomedical Engineering, King's College London, London, United Kingdom

ECG-gated cine MRI provides accurate functional information on the heart, however reliable ECG signals are not always available. We propose a novel framework to assess heart function based on physiological information obtained from MR images without an ECG. A Golden-Angle radial acquisition is used to obtain quantitative cardiac gating signals from blood volume changes in real-time images. This is used for reconstruction of retrospectively gated 2D multi-slice cine images. Finally, information on heart valve closure is used for slice synchronisation. Functional assessment of the left ventricle in five volunteers showed deviations less than 5% compared to standard ECG-gated Cartesian approach.

Claudio Santelli^1,2, Johannes F.M. Schmidt², 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

Scan acceleration and motion compensating techniques are invaluable to reduce acquisition times and breathing artifacts in cardiac imaging. Respiratory motion compensation with navigator based gating or breathholding is widely used, but may be limited in non-compliant patients. In this work, the SPIRiT reconstruction framework is extended by incorporating linear motion operators into the signal model allowing for scanning during the entire breathing cycle in cardiac imaging. Simulation and in-vivo phase-contrast data demonstrate the benefits of the technique: Results show reduction in blurring artifacts for magnitude and velocity induced phase.

Robert Grimm¹, Simon Bauer², Berthold Kiefer², Joachim Hornegger¹, and Kai Tobias Block^3
1Pattern Recognition Lab, FAU Erlangen-Nuremberg, Erlangen, Germany, ²Siemens Healthcare, Erlangen, Germany, ³Department of Radiology, NYU Langone Medical Center, New York City, NY, United States

In most respiratory self-gating techniques, the self-gating signal can be derived from all acquired channels. However, only coil elements close to the diaphragm deliver a reliable signal. Therefore, previous approaches usually required manual selection of an appropriate coil element or fixed scan protocols, both of which is infeasible for clinical routine applications. This work proposes a fully automatic method for this task, based on a score to characterize how well the self-gating signal shape resembles physiological respiration.

Wei Lin¹, Qin Qin^2,3, Stewart H. Mostofsky^4,5, and George Randy Duensing^1
1Invivo Corporation, Philips Healthcare, Gainesville, Florida, United States, ²Radiology, The Johns Hopkins University, Baltimore, Maryland, United States, ³F.M. Kirby Research Center, Kennedy Krieger Institute, Baltimore, Maryland, United States, ⁴Neurology and Psychiatry, The Johns Hopkins University, Baltimore, Maryland, United States, ⁵Kennedy Krieger Institute, Baltimore, Maryland, United States

A fast, robust and self-navigated 3D imaging method is proposed, based on sampling k-space with pairs of orthogonal view planes forming a cylinder. Rotation around its main axis is corrected in real-time, while other motions are corrected retrospectively at a temporal resolution of every 1-2 second. The method achieves the same off-resonance behavior and reconstruction speed as the conventional 3D Cartesian imaging. When array coil is used, scan time can be reduced significantly by acquiring less view planes. The efficacy of the proposed technique is demonstrated in healthy volunteers using a T1-weighted MP-RAGE sequence and a T2-weighted FLAIR sequence.

Wei Lin¹, Tim Nielsen², Qin Qin^3,4, Stewart H. Mostofsky^5,6, and George Randy Duensing^1
1Invivo Corporation, Philips Healthcare, Gainesville, Florida, United States, ²Philips Research Europe, Hamburg, Germany, ³Radiology, The Johns Hopkins University, Baltimore, Maryland, United States, ⁴F.M. Kirby Research Center, Kennedy Krieger Institute, Baltimore, Maryland, United States, ⁵Neurology and Psychiatry, The Johns Hopkins University, Baltimore, Maryland, United States, ⁶Kennedy Krieger Institute, Baltimore, Maryland, United States

A real-time strategy for 3D rigid-body motion correction in 2D multi-slice multi-shot MRI scans is proposed. Two through-plane navigator (tNav) echoes are collected on each imaging slice to derive two tNav projection images within each TR. Rotation/translation within each tNav image plane is detected using an image correlation measure. Rotation within the imaging plane is detected by replacing tNav with an orbital navigator on one slice. The proposed method does not introduce any additional 3D RF excitation, and does not require any additional hardware. The efficacy of the proposed method is demonstrated with in vivo brain studies.

Gregory R. Lee^1,2, Yong Chen^3,4, Nicole Seiberlich^4,5, Mark A. Griswold^3,4, and Vikas Gulani^4,6
1Radiology, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, United States, ²Radiology, University of Cincinnati, Cincinnati, OH, United States,³Radiology, Case Western Reserve University, Cleveland, OH, United States, ⁴Radiology, University Hospitals of Cleveland, Cleveland, OH, United States,⁵Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, ⁶Radiology, University Hospitals Case Medical Center, Cleveland, OH, United States

Dynamic contrast enhanced imaging of the abdomen is complicated by respiratory motion and often requires multiple breath-holds to be completed. In this work, the properties of a recently developed multi-echo 3D radial acquisition are leveraged to develop a method of self-navigation and subvolume registration. Self-navigation was performed by reconstructing low resolution images every 0.5 seconds to allow the extraction of a pencil-beam navigator. Nonlinear registration of extended duration images corresponding to each respiratory position was performed. Subsequent combination of undersampled images using the corresponding registration parameters allowed rapid, free breathing dynamic abdominal exams to be performed.

Mei-Lan Chu^1,2 and Nan-Kuei Chen^3
1Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei, Taiwan, ²Brain Imaging and Analysis Center, Duke University Medical Center, Durham, NC, United States, ³Brain Imaging and Analysis Center, Duke University, Durham, NC, United States

A robust motion-immune MRI acquisition and reconstruction strategy is presented in this study. The experimental results indicate that the proposed REKAM method can effectively suppress motion-related artifacts even for highly challenging cases, such as continual head tremor during scans. As compared with the conventional SPGR that is highly susceptible to intra-scan motion, the REKAM method can produce artifact-free images of the same spatial resolution within the same scan time. Even though the REKAM method is demonstrated here with T2*-weighted imaging, the developed technique can be applied to structural MRI pulse sequences of other weightings and contrasts.

Muhammad Usman¹, Paul Aljabar¹, Tobias Schaeffter¹, and Claudia Prieto^1,2
1King's College London, London, United Kingdom, ²Pontificia Universidad Catolica de Chile, Santiago, Chile

Manifold learning approaches can be applied in MRI to extract meaningful patterns of variation from the high-dimensional set of images. An example is respiratory self-gating where the low dimensional respiratory signal can be extracted from a set of free breathing images. In this work, we propose Compressive Manifold learning for respiratory self-gated MRI. This approach estimates the respiratory signal directly from undersampled k-space data, without the need for image reconstruction. Results from simulated free-breathing liver MR data show that the respiratory signal can be accurately extracted from highly undersampled k-space samples using the proposed method. Free-breathing acquisitions, performed prospectively in 3 volunteers, also demonstrate the feasibility of CML respiratory self-gating in k-space.

Cris Lovell-Smith¹, Julian R. Maclaren², Michael Herbst¹, Ilja Kadashevich³, K.A. Danishad³, Oliver Speck³, and Maxim Zaitsev^1
1Department of Radiology, University Medical Center Freiburg, Freiburg, Germany, ²Department of Radiology, Stanford University, Stanford, California, United States, ³Faculty of Natural Sciences, Institute of Experimental Physics (IEP), Magdeburg, Germany

Motion-induced artefacts in MR images degrade quality. The use of an optical tracking system to prospectively compensate for head movement during brain imaging has been shown to reduce these artefacts. To successfully apply motion correction, the cross-calibration matrix defining the transform between tracking system coordinates and scanner coordinates must be accurately known. We previously reported a method to determine this transform, however it required two experienced MR technicians upward of 20 minutes to perform. We present two newly developed methods that accurately calculate the cross-calibration within two minutes with one operator, thereby reducing the system setup time considerably. This is a major step in bringing prospective motion correction to clinical routine.

Murat Aksoy¹, Melvyn B. Ooi¹, Julian R. Maclaren¹, and Roland Bammer^1
1Center for Quantitative Neuroimaging, Department of Radiology, Stanford University, Stanford, CA, United States

Cross-calibration of the optical tracking system and the scanner is an essential step in prospective motion correction. Errors in cross-calibration cause inaccuracies in motion correction and subsequent motion artifacts in the image. In this study, we assessed the accuracy requirement of cross-calibration for robust motion tracking and correction. Simulations show that for a maximum of 0.5mm pixel position error and in the presence of 10° head rotation, the cross-calibration has to be accurate up to 0.3° and 3 mm. The required accuracy of cross-calibration depends on the maximum motion that needs to be corrected and the baseline position of the camera in the scanner.

Axel Hartwig¹, Magnus Mårtensson², and Stefan Skare^1
1Neuroradiology, Karolinska University Hospital, Stockholm, Sweden, ²EMEA Research and Collaboration, Applied Science Laboratory, GE Healthcare, Stockholm, Sweden

A new image based navigation sequence is proposed to overcome nodding motion in MRI, where a thick coronal excitation plane over the patient’s nose followed by a sagittal EPI readout tracks the nose in real time - NoseNav. This sequence module will be used in a real-time setting and update the prescribed slices over the brain so that they follow the nodding motion of the patient, which also is the direction that is hardest to restrain in practice.

Trong-Kha Truong¹ and Arnaud Guidon^1
1Brain Imaging and Analysis Center, Duke University, Durham, NC, United States

In multi-shot spiral diffusion tensor imaging, subject motion causes phase errors among different shots, leading to signal loss and aliasing artifacts. A novel method is proposed to inherently correct for these errors without requiring a variable-density spiral trajectory or a navigator echo. This method uses a sensitivity encoding reconstruction algorithm to estimate the phase error for each shot and a conjugate gradient algorithm to correct for them. In vivo experiments were performed to demonstrate that it can inherently and efficiently correct for phase errors caused by rigid and nonrigid motion, without increasing the scan time or reducing the signal-to-noise ratio.

Jessica Schulz¹, Thomas Siegert¹, Pierre-Louis Bazin¹, Julian R. Maclaren², Michael Herbst³, Maxim Zaitsev³, and Robert Turner^1
1Neurophysics, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, ²Stanford University, Stanford, CA, United States, ³Radiology, University Medical Center Freiburg, Freiburg, Germany

This study compared the rate of false positive activations in an fMRI group study with and without slice-by-slice prospective motion correction using an optical tracking system. The paradigm involved strong task-correlated motion - a major cause of false positive activations in fMRI. The number of voxels with T-values higher than 5 outside of grey matter – i.e. activations that are obviously false positive - was strongly reduced using prospective motion correction. Additionally, the statistical power of the analysis increased. Therefore, we suggest that prospective correction techniques should be used whenever available.

Ashley Gould Anderson III¹, Christopher J. François², and Oliver Wieben^1,2
1Medical Physics, University of Wisconsin, Madison, Wisconsin, United States, ²Radiology, University of Wisconsin, Madison, Wisconsin, United States

A detailed analysis of the motion of 11 segments of major abdominal vessels (renal, hepatic, mesenteric, splenic arteries, aorta) during free breathing, as well as during inspiration and expiration breath holds was conducted. The results form a comprehensive reference for motion of the abdominal vasculature that extends the knowledge from previous motion studies. These data should be valuable for those interested in imaging the abdominal vasculature using respiratory-gated or breath-held acquisitions.

Shujing Cao¹, Feng Huang², Rui Li¹, and Chun Yuan^3
1Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, Beijing, China, ²Philips Healthcare, Beijing, Beijing, China, ³Department of Radiology, University of Washington, Seattle, WA, United States

A new retrospective motion compensation method named as GeneRAlized Multiple Averages (GRAMA) is proposed. GRAMA synthesizes two copies of original k-space with relative consistency using a couple of optimized convolution kernel. Based on the interleaved data acquisition manner, error in different k-space copies is inherent. Instead of average with original k-space like that in conventional multiple averages method, GRAMA directly use the average of two synthetic k-space to reconstruct motion corrected image. Volunteer experiments indicate GRAMA effectively balances SNR preservation and artifact reduction.

Riad Ababneh¹, Thomas Benkert², Martin Blaimer², and Felix A. Breuer^2
1Physics, Yarmouk University, Irbid, Jordan, ²Research Center Magnetic Resonance Bavaria, Würzburg, Bavaria, Germany

We present a robust approach to separate fat and water signals in the abdomen during free breathing. The approach combined with the k-space-weighted image contrast (KWIC) technique of different respiratory phases in free-breathing. In this study a radial TrueFISP sequence was modified, wherein TE was made to vary between subsequent readouts. Good separation without streaking artifacts or blurring due to respiratory motion was obtained in all studied cases.

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

In this abstract, we report the use of NMR probes for prospective motion corrections of head motion in high-resolution gradient echo (GRE) imaging at high field of 7 Tesla. 3 NMR probes mounted on a volunteer’s head are used in conjunction with a linear navigator sequence to measure and correct for head motion in real time. The system is evaluated with 1 mm³ isotropic voxel GRE scans with a series of motion patterns on normal volunteers. Effective real time compensation is demonstrated for each type of motion tested.

Tobias Kober^1,2, Davide Piccini^1,2, Christoph Forman^3,4, Thorsten Feiweier⁵, and Gunnar Krueger^1,2
1Advanced Clinical Imaging Technology, Siemens Healthcare IM S AW, Lausanne, Switzerland, ²CIBM - AIT, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, ³Pattern Recognition Lab, Friedrich-Alexander-University Erlangen-Nürnberg, Erlangen, Germany, ⁴Erlangen Graduate School in Advanced Optical Technologies, Friedrich-Alexander-University Erlangen-Nürnberg, Erlangen, Germany, ⁵Siemens AG, Erlangen, Germany

The Magnetisation-Prepared Rapid Acquisition Gradient Echo (MP-RAGE) sequence is widely used in clinical routine and research. However, its comparably long acquisition times render MP-RAGEs susceptible to motion artefacts. The present work explores the feasibility of mitigating the motion susceptibility of MP-RAGEs. While a free-induction-decay navigator is used for motion detection, the self-navigating properties of a hybrid acquisition trajectory are exploited to perform motion correction directly from portions of the imaging data. The motion detection and correction performance was tested in five volunteers proving the feasibility of this concept.

Shayan Guhaniyogi¹, Allen Song¹, Jeffrey Petrella², and Nan-Kuei Chen^1
1Brain Imaging and Analsyis Center, Duke University, Durham, NC, United States, ²Department of Radiology, Duke University, Durham, NC, United States

Multishot EPI in diffusion weighted imaging allows improved spatial resolution and reduced geometric distortions compared to single-shot acquisitions. However, multishot EPI acquisitions are very sensitive to patient motion. To overcome this limitation we present a technique which inherently corrects motion-induced artifacts and blurring in multishot diffusion-weighted EPI images using only the originally acquired data, without needing navigator echoes. The technique is shown to greatly reduce artifacts and blurring, enabling visual identification of small structures such as the optic nerve fibers. This method is anticipated to be valuable in clinical and neuroscience investigations requiring high resolution images with detailed microstructure information.

Michael Herbst¹, Cris Lovell-Smith¹, Benjamin Haeublein¹, Rebecca Sostheim¹, Julian R. Maclaren², Jan G. Korvink³, and Maxim Zaitsev^1
1Department of Radiology, University Medical Center Freiburg, Freiburg, Germany, ²Department of Radiology, Stanford University, Stanford, California, United States, ³Laboratory for Simulation, IMTEK - Institute of Microsystem Technology, Freiburg, Germany

Despite the recent advances in prospective motion correction (PMC) for the correction of rigid body motion, the application of this technique to clinical routine still lacks behind. External tracking and motion correction has been shown to be applicable to a wide range of sequences and to provide sufficient accuracy and precision to improve even data from cooperative volunteers. However, when used in patients, problems such as marker fixation and involuntary skin motion limit the abilities of PMC and can even lead to additional motion artifacts. The aim of this work was to address this last key barrier to help bring PMC to clinical routine.

Patrick J. Bolan¹, Xiufeng Li¹, and Gregory J. Metzger^1
1Radiology, CMRR, University of Minnesota, Minneapolis, MN, United States

While strategies for handling cardiac and respiratory motion are well established for cardiac and abdominal imaging, the aperiodic and non-rigid deformations from bowel contractions can limit the ability to perform high-resolution imaging in the lower abdomen and pelvis. This work assesses the feasibility of using a retrospective, non-rigid motion-corrected averaging approach to reduce the impact of involuntary motion on anatomical MRI of the pelvis.

Anthony N. Price¹, Shaihan J. Malik¹, and Joseph V. Hajnal^1
1Division of Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom

Balanced SSFP provides excellent contrast and SNR efficiency, however, its familiar off-resonance characteristics result in black-band artefacts, which are particularly problematic at higher field strengths. Less well-known, but also significant to image quality, are the unwanted signal fluctuations arising from out-off-slice contributions, which are folded back into the imaging plane. Here we design a new RF pulse specifically for 3D SSFP that has very low stopband ripple. When compared to the standard truncated sinc pulse typically used, it yields effective suppression of off-resonance signal ripple whilst achieving a more homogeneous slab profile.

Kaveh Vahedipour^1,2 and Nadim Jon Shah^1,3
1Juelich Research Centre, Juelich, Germany, ²Maastricht University, Maastricht, Netherlands, ³RWTH Aachen University, Aachen, Germany

The spatio-temporal duality of the Fourier transform, the time-reveral parity of Bloch equation without T1 and T2* relaxation as well as the principle of reciprocity can be used to rewind the physical process of gradient encoding along the time axis to obtain images from signals acquired along arbitrary k-space trajectories. In this work we show advantages of such a reconstruction scheme when compared to the state of the art non-uniform FFT solutions. The algorithm can be implemented in a very efficient way with respect to memory as well as processing complexity. It is discussed why this algorithm behaves favourably particularly for larger number of receive elements and large image matrices.

Myung-Ho In¹, Oleg Posnansky¹, Erik B. Beall², Mark J. Lowe³, and Oliver Speck^1
1Biomedical Magnetic Resonance, Otto-von-Guericke University, Magdeburg, Germany, ²Radiology, Cleveland Clinic, Cleveland, OH, United States,³Radiology, Imaging Institute, Cleveland Clinic, Cleveland, OH, United States

The geometric distortion correction quality is important to generate a distortion free diffusion weighted echo planar image (DW-EPI) without loss of spatial information. To recover spatial information in strongly compressed areas, two sets of DWIs with opposite phase-encoding (PE) polarities have been measured and combined after distortion correction. The point spread function (PSF) mapping method was used for correction due to its very high accuracy and robustness. To reduce PSF scan time, the PSF method was extended to correct distortions in both DW-EPIs with opposite PE polarities from a single PSF reference.

Mario Zeller¹, Alexander Müller¹, Dietbert Hahn¹, and Herbert Köstler^1
1Institute of Radiology, University of Würzburg, Würzburg, Germany

Artifacts induced by field inhomogeneities in non-Cartesian EPI acquisitions typically cannot be corrected by simple pixel-shifting. Instead, conjugate phase based methods can be applied. These methods require the acquisition of a separate gradient-echo field map. The acquisition of such a map is relatively time-consuming, as multiple radiofrequency pulses are applied. In this work, a method is described, which allows deriving the field map from a phase-labeled reference EPI scan comprising of only two fast EPI acquisitions. With this map, geometric distortions arising in non-Cartesian EPI can be successfully corrected.

Benedikt A. Poser¹, Soeren Johst², Weiran Deng¹, Stephan Orzada², Mark E. Ladd^2,3, and Victor Andrew Stenger^1
1John A Burns School of Medicine, University of Hawaii, Honolulu, Hawaii, United States, ²Erwin L Hahn Institute for Magnetic Resonance Imaging, University Duisburg-Essen, Essen, Germany, ³Department of Diagnostic and Interventional Radiology and Neuroradiology, University Hospital Essen, Essen, Germany

Gradient-echo BOLD fMRI suffers signal voids in regions of strong susceptibility variation e.g. near ear canals and frontal sinuses. “z-shimming” can correct this by applying a z-gradient moment to counteract the local background gradient. The global correction affects all spins and dephases originally intact signal, hence an intermediate compromise is required. Timeshifting an RF pulse equivalently creates a shim. This gives “pTX z-shimming” some spatial selectivity following each channel’s sensitivity profile. pTX z-shimming is explored at 7T, using EPI with slice and channel specific corrections and breathold fMRI. Signal recovery was seen near ear canals, but not the frontal cortex.

Rahul Dewal¹ and Qing X. Yang^1
1Radiology, Pennsylvania State University, Hershey, PA, United States

Calculation of B₀ field inhomogeneity maps from susceptibility models via convolution with dipole kernels is limited by computer memory (RAM) and calculation time requirements in the case of large data matrices. In some applications, only small portions of the full matrix may be of interest as a susceptibility perturbation source or as a volume of interest (VOI). A VOI-based approach is here proposed that can reduce memory usage or calculation time in this sparse case. The method can be expected to accelerate various applications, such as passive shimming simulations and patient-specific B₀ field calculations derived from anatomic scans.

Ek T. Tan¹, Christopher J. Hardy¹, and John F. Schenck^1
1Diagnostics and Biomedical Technologies, GE Global Research, Niskayuna, NY, United States

The correction of residual phase intrinsic to phase-contrast velocity measurements remains a challenging problem, and affects accurate flow quantitation in imaging of congenital heart disease. A high degree of correlation between the residual phase and the concomitant field motivated an investigation into an improved concomitant field correction that accounts for gradient nonlinearity. The nonlinear correction resulted in a residual phase that was more quadratic in shape, but did not affect the residual phase near the vessels of interest. However, the incorporation of nonlinear concomitant fields into a hybrid-fitting method showed an improvement in the fitted result over the linear correction.

Simon P. Robinson¹, Horst Schödl¹, and Siegfried Trattnig^1
1Department of Radiology, Medical University of Vienna, Vienna, Vienna, Austria

We describe a temporal phase unwrapping method applicable to multi-echo data called UMPIRE. This uses unequal spacings between echoes and an assessment of phase evolution in the time difference between the echo spacings to allow - uniquely amongst temporal unwrapping approaches - a number of wraps between echoes to be resolved. UMPIRE successfully resolved wraps in complicated simulated shapes which could not be unwrapped with prior temporal methods or the spatial methods PRELUDE and PHUN. UMPIRE also outperformed spatial methods in unwrapping in-vivo data and was faster than spatial methods.

Amir Moussavi^1,2, Markus Untenberger¹, Martin Uecker³, and Jens Frahm^1,2
1Biomedizinische NMR Forschungs GmbH, am Max-Planck-Institut für biophysikalische Chemie, Göttingen, Niedersachsen, Germany, ²DFG Research Center for Molecular Physiology of the Brain (CMPB), Göttingen, Niedersachsen, Germany, ³Dept. of Electrical Engineering and Computer Sciences, University of California, Berkeley, California, United States

Gradient-induced eddy currents affect the MRI data acquisition by gradient delays and phase errors that may lead to severe image artifacts. This work introduces a robust method for quantifying and correcting phase errors in radial MRI without the need of complementary measurements. The proposed method yields a specific phase error per gradient that can be used for correcting the raw data. It is also able to distinguish between and separately correct for phase errors due to eddy currents caused by the in-plane gradients and other phase alterations which are constant for an image and that may carry relevant physiologic information.

Kevin F. King¹, Graeme C. McKinnon¹, Xu Dan¹, and Joseph K. Maier^1
1GE Healthcare, Waukesha, WI, United States

Eddy currents are typically measured using small samples with a special fixture. As part of an image-based tool without a special fixture, we propose to measure eddy currents with time constants of a few hundred microseconds or less with ultrashort TE imaging. A spherical phantom was scanned with eddy current generating preparation gradients followed by readouts at increasing delays. The image phase was fit to spatially linear terms which were then fit to sums of decaying exponentials. Reasonable agreement with fixture-based decay curves and comparable correction for ghosting in single-shot EPI and spectral-spatial RF pulse banding was found.

Yuval Zur^1
1GE Healthcare, Haifa, Israel

Spectral spatial (spsp) RF pulses are used for simultaneous fat suppression and slice selective excitation. However, spsp pulses are very sensitive to RF gradient delay and eddy currents. We present a correction method, such that spsp pulses are, in practice, not sensitive to system imperfections. This correction enables practical use of spsp RF pulses with slice width of 1.5 mm at high field strength (3T).

Md. Shahadat Hossain Akram¹ and Katsumi Kose^1
1Institute of Applied Physics, University of Tsukuba, Ibaraki, Tsukuba, Japan

Transient response of eddy currents generated by planar z gradient coil in the local Rf shielding box of a 0.3T permanent magnet MRI system are simulated by applying a novel approach in the network analysis of electromagnetic system. Magnetic scalar potential of Biot Savart’s law is implemented in the calculation of mutual inductances between the z gradient coil and the conducting structure in interest, which is always the most crucial part in network analysis method. This very unique and efficient method would be possible to implement in network analysis of eddy current generated by any type of planar gradient coil.

Adrienne E. Campbell-Washburn^1,2, David Atkinson³, Oliver Josephs⁴, Mark F. Lythgoe⁵, Roger J. Ordidge⁶, and David L. Thomas^7
1Centre for Advnaced Biomedical Imaging, University College London, London, London, United Kingdom, ²Department of Medical Physics and Bioengineering, University College London, London, United Kingdom, ³Centre for Medical Imaging and Centre for Medical Image Computing, University College London, London, United Kingdom, ⁴University College London and Birkbeck College, London, United Kingdom, ⁵Centre for Advanced Biomedical Imaging, Division of Medicine and Institute of Child Health, University College London, London, United Kingdom, ⁶Centre for Neuroscience, University of Melbourne, Melbourne, Victoria, Australia, ⁷Department of Brain Repair and Rehabilitation, UCL Institute of Neurology, University College London, London, United Kingdom

RF spike noise, caused by hardware problems, can lead to striping artefacts in MR images. These artefacts affect the image quality and quantitative information from the MRI data, and often must be removed in post-processing. This abstract presents an algorithm for semi-automated detection and correction of RF spike noise based on Robust Principal Component Analysis (RPCA). RPCA is used to decompose the measured k-space into low-rank (artefact-free) and sparse (RF spike) matrix components, including an automatic correction for the misidentification of the central k-space cluster. This algorithm is demonstrated to efficiently and effectively recover artefact-free data and regain quantitative information.

Paul Kyu Han¹, Jong Chul Ye¹, HyunWook Park², and Sung-Hong Park^1
1Bio and Brain Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Daejeon, Korea, ²Electrical Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Daejeon, Korea

We proposed a new technique that allows for the correction of DC artifact for any phase-cycled pulse sequence. The proposed method includes receiver phase unification and its reversal before and after the application of a usual DC artifact removal technique in the specifically phase-cycled pulse sequence. Results show complete removal of the DC artifact for phase cycling angles of 0¡Æ, 90¡Æ, 180¡Æ, 270¡Æ in the phase-cycled bSSFP sequence, and provides useful insight for the application of DC artifact removal in other phase-cycled pulse sequences.

Horst Schödl¹, Siegfried Trattnig¹, and Simon Robinson^1
1Department of Radiology, Medical University Vienna, Vienna, Vienna, Austria

Phase images can be unwrapped with spatial or temporal approaches. A new and fast temporal approach, suited to highly wrapped phase images at ultra-high field, is UMPIRE. It is based on a multi-echo acquisition with unequal echo spaces, and uses the difference between phase difference images (essentially a wrap-free estimation of B0) to unwrap phase images. UMPIRE was implemented in C using parallelization. The performance of UMPIRE was compared with region growing methods PHUN and PRELUDE using simulated data with complicated shapes and in-vivo images. UMPIRE was dramatically faster than the spatial methods PHUN and PRELUDE in all tests.

Wentao Liu¹, Xin Tang¹, Yajun Ma¹, and Jia-Hong Gao^1,2
1MRI Research Center and Beijing City Key Lab for Medical Physics and Engineering, Peking University, Beijing, China, ²Brain Research Imaging Center and Department of Radiology, The University of Chicago, Chicago, Illinois, United States

Multi-shot echo-planar imaging (msEPI) has advantages of high spatial resolution and less distortion when comparing to single-shot EPI (ssEPI). However, in the application of diffusion imaging, msEPI often suffers ghosting artifacts due to phase shifts from shot to shot. A recently proposed phase cycled reconstruction (PCR) method can correct 2D linear phase errors in msEPI efficiently using a search algorithm, but phase shifts caused by the pulsing of cerebrospinal fluid (CSF) are nonlinear and hard to be compensated by PCR method directly. In this study, an efficient parallel imaging estimated phase cycled reconstruction (PIPCR) method is introduced. The inter-shot phase maps are estimated from the parallel imaging (PI) reconstruction of the subset k-space of msEPI data. After a smoothing process, these phase maps are utilized as nonlinear phase compensation for PCR msEPI reconstruction. The proposed PIPCR is a fast and robust method to correct arbitrary inter-shot phase errors of diffusion weighted msEPI and the reconstruction result is demonstrated in the report.

Dan Xu¹, Kenichi Kanda¹, Kevin F. King¹, Zhenghui Zhang¹, and Robert D. Peters^1
1GE Healthcare, Waukesha, WI, United States

Conventional echo planar imaging (EPI) phase correction can introduce a channel dependent phase which could conflict with parallel imaging unaliasing especially for high channel count coils, leading to banding or residual aliasing artifacts in the final images. In this paper we propose a phase matching method to remove the incompatibility by introducing a channel dependent phase while retaining the odd-even phase correction capability, thereby producing images free of banding and aliasing artifacts, in addition to minimizing EPI Nyquist ghost.

Martin Krämer¹, Karl-Heinz Herrmann¹, and Jürgen R. Reichenbach^1
1Medical Physics Group, Institute of Diagnostic and Interventional Radiology I, Jena University Hospital - Friedrich Schiller University Jena, Jena, Germany

Radial center-out trajectories are extremely sensitive towards imperfections in the gradient and data sampling hardware. Even the smallest jitter between data sampling and gradient onset must be compensated during image reconstruction in order to obtain images free of delay artifacts. We propose to estimate such delays with a template scan using a very short additional dephaser. The delay is given from the difference between the expected k-space center position along the template readouts and the actual center crossing. We shot that a very robust estimation of the true k-space center crossing can be made by analyzing k-space phase directly.

Robert Grimm¹, Christoph Forman², Jana Hutter¹, Berthold Kiefer³, Joachim Hornegger¹, and Kai Tobias Block^4
1Pattern Recognition Lab, FAU Erlangen-Nuremberg, Erlangen, Germany, ²Pattern Recognition Lab, Friedrich-Alexander-University Erlangen-Nürnberg, Erlangen, Germany, ³Siemens Healthcare, Erlangen, Germany, ⁴Department of Radiology, NYU Langone Medical Center, New York City, NY, United States

In radial stack-of-stars GRE imaging, distortions of the magnetic B₀ field and gradient fields distant from the isocenter can cause significant streaking artifacts. The image quality can be severely degraded by coils that contribute mostly artifacts but little image information. Here, an efficient, k-space-based score is proposed as an indicator for channels that are corrupted by streaking artifacts. A k-means clustering algorithm is employed to detect affected channels in order to exclude them from reconstruction. The method requires no manual training and, therefore, can be applied easily to different scan protocols.

Ikuko Uwano¹, Kohsuke Kudo¹, Fumio Yamashita¹, Jonathan Goodwin¹, Tsuyoshi Metoki¹, Satomi Higuchi¹, Kenji Ito¹, and Makoto Sasaki^1
1Division of Ultrahigh Field MRI, Institute for Biomedical Sciences, Iwate Medical University, Morioka, Iwate, Japan

In ultra-high field 7 Tesla (T) MRI, signal intensity variation or inhomogeneity are caused by main magnetic field (B0) and RF field (B1) inhomogeneity, susceptibility effects, and inhomogeneous coil sensitivity. We tested a post-processing technique available with SPM8, for signal intensity correction of various scan sequences on 7 T MRI. We found that by using SPM8, signal inhomogeneity was successfully corrected and a reduction in signal non-uniformity between the subcortical and deep white matter was observed.

Sven Jaeschke^1,2, Robin Martin Heidemann², and Aleksandar Petrovic^2
1HAW Hamburg, Hamburg, Germany, ²Healthcare Sector, Siemens AG, Erlangen, Germany

Ultra high field (UHF) MRI offers improved image contrast and higher SNR enabling isotropic, high-resolution in vivo anatomical imaging. However, image quality is affected by bias fields, which can be a hindrance for tissue segmentation or classification. In this study, we propose a novel, fast and fully automated image inhomogeneity correction method, which is well suited for UHF applications. In comparison to former methods, we employ a bounded Nelder-Mead simplex optimizer and use a joint intensity-gradient histogram to calculate entropy. We show that our approach outperforms other methods in UHF-MRI in terms of processing speed and promises better inhomogeneity correction.

Sven Jaeschke^1,2, Alexander Hubert³, Andreas J. Bartsch⁴, Antje Kickhefel², and Aleksandar Petrovic^2
1HAW Hamburg, Hamburg, Germany, ²Healthcare Sector, Siemens AG, Erlangen, Germany, ³Neuroradiology, Universitätsklinikum Heidelberg, Heidelberg, Germany, ⁴Neuroradiology, University of Heidelberg, Heidelberg, Germany

Intensity inhomogeneities remain an issue in high-field MRI and can degrade clinical interpretability of images. To improve image quality in a fast, robust and fully automated fashion, we developed and clinically validated a novel image inhomogeneity correction method, based on entropy minimization. A wide range of different MR-contrasts/sequences in healthy-control and images with pathological findings were assessed. Two trained radiologists evaluated each image in a blind fashion and both agreed on each rating. We show that the combination of the 3D implementation of our method and the current built-in Siemens solution significantly improves the clinical impression of 3T neuroradiological MRI.

Sumeeth Vijay Jonathan¹, Parmede Vakil¹, Yong Jeong¹, and Timothy J. Carroll^1,2
1Biomedical Engineering, Northwestern University, Chicago, IL, United States, ²Radiology, Northwestern University, Chicago, IL, United States

We characterize and introduce a phase correction scheme for N/2 ghosting artifacts in 3D RAZIR, a new radial 3D GRE-EPI pulse sequence. Phase modulations between alternating echoes in our radial 3D EPI acquisition are dependent on radial view. Our phase correction scheme measures phase modulations for each radial view without the need for a separate reference scan. N/2 ghosting artifacts that occur in radial 3D multiecho acquisitions like 3D RAZIR can be corrected with our technique.

Paul Wighton¹, Matthew Dylan Tisdall¹, and André J. W. van der Kouwe^1
1Department of Radiology, MGH, Martinos Center for Biomedical Imaging, Charlestown, Massachusetts, United States

In this abstract, we explore the suppression of eye-movement artifacts using a subject-specific, spatially-selective RF-excitation pulse designed to excite only the subject's brain, ensuring no signal (and therefore no artifact) originates from the eyes. A 3D-encoded FLASH sequence was augmented with a 2D-selective RF-excitation pulse. A subject specific 2D brain-excitation shape was generated using a FreeSurfer segmentation of the subject acquired in a previous scan. Compared to a conventional FLASH scan, eye-motion artifacts were significantly suppressed with our modified spatially-selective RF-excitation scan.

Mark Chiew¹, Karla L. Miller¹, Peter J. Koopmans², Elizabeth M. Tunnicliffe³, Stephen M. Smith¹, and Thomas Blumensath^4
1FMRIB Centre, University of Oxford, Oxford, United Kingdom, ²Donders Institute for Brain, Cognition and Behaviour, Radboud University Nijmegen, Nijmegen, Netherlands, ³AVIC, Nuffield Department of Clinical Medicine, University of Oxford, Oxford, United Kingdom, ⁴ISVR, University of Southampton, Southampton, Hampshire, United Kingdom

In matrices that are low rank or approximately so, matrix completion strategies can be used to recover data in the presence of undersampling. Here we present a novel algorithm, iterative hard thresholding + matrix shrinkage (IHT+MS), for the recovery of low rank approximations to k-t undersampled MRI data. Performance of the IHT+MS algorithm is compared to other matrix completion techniques in retrospectively undersampled cardiac cine and fMRI data. Results indicate that good reconstruction fidelity is observed in both cardiac and fMRI data, even at high undersampling factors, and that IHT+MS produces the best results in many cases.

KyungHyun Sung¹ and Brian Andrew Hargreaves^2
1Radiological Sciences, UCLA, Los Angeles, California, United States, ²Radiology, Stanford University, Stanford, CA, United States

High computational complexity is one major issue for compressed sensing (CS) reconstruction. We present a new way to reduce the computational complexity for the CS reconstruction by directly transforming between k-space and wavelet domains. This replaces FIR filtering in the image domain with a multiplication in k-space and can reduce computational complexity. This efficient computation can benefit almost all CS methods that exploit the wavelet sparsity, and we have shown the actual CS reconstruction time can be reduced by 46% on MATLAB.

Evan Levine^1,2, Kyunghyun Sung³, Manojkumar Saranathan², Bruce Daniel², and Brian Andrew Hargreaves^2
1Electrical Engineering, Stanford University, Palo Alto, CA, United States, ²Radiology, Stanford University, Palo Alto, CA, United States, ³Radiological Sciences, University of California, Los Angeles, Los Angeles, CA, United States

Iterative Soft Thresholding (IST) and a more recent method, Approximate Message Passing (AMP) are promising techniques for fast compressed sensing (CS) reconstruction. Combining CS, parallel imaging, partial-Fourier, and other techniques is still being investigated, and the reconstruction time is critical. The AMP method, as is currently implemented on a single coil basis, does not exploit known dependencies among coil images. We have developed a novel reconstruction method, AMP-SENSE, that incorporates coil sensitivity information into AMP to address the combined CS and parallel imaging problem, improving both performance and computational efficiency. We demonstrate an application to dynamic contrast-enhanced (DCE) breast MRI.

Peng Lai¹, Shreyas S. Vasanawala², Michael Lustig³, Kang Wang⁴, and Anja C.S Brau^5
1MR Applications & Workflow, GE Healthcare, Menlo Park, CA, United States, ²Radiology, Stanford University, Stanford, CA, United States, ³Electrical Engineering & Computer Science, University of California, Berkeley, CA, United States, ⁴MR Applications & Workflow, GE Healthcare, Madison, WI, United States, ⁵MR Applications & Workflow, GE Healthcare, Garching, Munchen, Germany

Compressed sensing reconstruction requires many iterations to converge. This work developed an auto-calibrating parallel imaging method that can efficiently reconstruct coil-combined k-space data from random k-space sampling. Our preliminary results show that the proposed method can provide similar reconstruction accuracy with much faster computation compared to conventional auto-calibrating parallel imaging and can significantly improve the initial condition and convergence of compressed sensing reconstruction.

Jing Liu¹ and David A. Saloner^1
1University of California San Francisco, San Francisco, CA, United States

Compressed sensing (CS) and parallel imaging (PI) have been exploited to reduce scan time by undersampling k-space data, which is highly desirable for 3D applications that usually involve unreasonably long scan times. This study proposes a novel method for generating undersampling patterns for 3D Cartesian acquisition, providing easy implementation, flexible sampling patterns, and high accuracy of image reconstruction with CS&PI.

Zhitao Li¹, Benjamin Paul Berman², Maria I. Altbach³, Jean-Philippe Galons³, Diego R. Martin³, Bin Dong⁴, Puneet Sharma³, Natarajan Raghunand³, and Ali Bilgin^1,5
1Electrical and Computer Engineering, University of Arizona, Tucson, Arizona, United States, ²Applied Mathematics, University of Arizona, Tucson, Arizona, United States, ³Medical Imaging, University of Arizona, Tucson, Arizona, United States, ⁴Mathematics, University of Arizona, Tucson, Arizona, United States, ⁵Biomedical Engineering, University of Arizona, Tucson, Arizona, United States

High spatial and temporal resolution in DCE are desirable. High temporal resolution is needed for accurate kinetic data analysis and high spatial resolution contributes to small structures identification. Radial DCE techniques have been combined with CS to accelerate DCE. Radial trajectories provide desirable attributes: Oversampling of the center k-space, incoherent artifacts from undersampled trajectories can be exploited in CS. Radial trajectories are also less sensitive to motion. A dynamic MRI technique using 3D stack-of-stars was proposed for free breathing 3D liver perfusion MRI. We propose a highly accelerated 3D dynamic MRI technique which uses 3D stack-of-stars with non-uniform kz sampling.

Kaveh Vahedipour^1,2 and Nadim Jon Shah^1,3
1Juelich Research Centre, Juelich, Germany, ²Maastricht University, Maastricht, Netherlands, ³RWTH Aachen University, Aachen, Germany

Combination of non-Cartesian data sampling and Compressed Sensing is an obvious way to go for approaching to the best SNR per unit time. Here, the non-Cartesian sampling strategy must incorporate the random sampling paradigm in the most efficient way in multiple facets. In general, one would like to acquire as much signal amplitude as possible per image while achieving as much randomness as possible of the sampling footprint and thus as little coherences in the artifact domain as achievable. However, a third aspect, which has thus far remained unconsidered, is that of induction of Eddy-currents and their effect on the image quality.

Youngseob Seo^1,2, Jonathan M. Chia³, Nancy K. Rollins^2,4, and Zhiyue J. Wang^2,4
1Division of Convergence Technology, Korea Research Institute of Standards and Science, Daejeon, Korea, ²Radiology, University of Texas Southwestern Medical Center at Dallas, Dallas, Texas, United States, ³Philips Healthcare, Cleveland, Ohio, United States, ⁴Radiology, Children's Medical Center Dallas, Dallas, Texas, United States

Using compressed sensing reconstruction method for fast cardiac MRI, we shows how the number of central k-space lines with varying k-space sampling schemes affects sparsely sampled reconstructed image quality.

Suyash P. Awate¹ and Edward V.R. DiBella^2
1Scientific Computing and Imaging (SCI) Institute, University of Utah, Salt Lake City, Utah, United States, ²Utah Center for Advanced Imaging Research (UCAIR), University of Utah, Salt Lake City, Utah, United States

We propose a novel unified framework for compressed-sensing reconstruction of multidimensional magnetic resonance imaging (MRI) including dynamic MRI and high angular resolution diffusion imaging (HARDI). This brand-new framework incorporates a novel formulation for the compressed-sensing reconstruction problem which makes it very flexible with regards to (i) the kinds of imaging or undersampling strategies that can be exploited as well as (ii) the kinds of sparse models that need to be enforced on the data, allowing a variety of wavelet-frame models, total-variation models, and novel dictionary models.

Amaresha Sridhar Konar¹, Steen Moeller², Julianna Czum³, Barjor Gimi³, and Sairam Geethanath^1
1Biomedical Research Center, Dayananda Sagar Institutions, Bangalore, Karnataka, India, ²Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, Minnesota, United States, ³Dept. of Radiology, Giesel School of Medicine at Dartmouth, Lebanon, New Hampshire, United States

Compressed sensing (CS) performance depends significantly on sparsity of the image data.The current work aims at providing additional sparsity regardless of the transform chosen to achieve increased acceleration than the conventional CS approach, usinga novel technique called “Region of Interest Compressed Sensing” (ROICS). ROICS allows for enhanced sparsity by decreasing the number of non-zero coefficients to be estimated by restricting the CS reconstruction to a ROI. This work demonstrates that ROICS outperforms CS at higher acceleration factors, quantified through reduced normalized root mean square error, as applied to cardiac MRI frames.

Sajan Goud Lingala¹ and Mathews Jacob^1
1The University of Iowa, Iowa city, IA, United States

We propose an sparse blind compressive algorithm to learn dictionary atoms that are constrained to be sparse for accelerated dynamic MRI reconstruction. The sparsity promoting norm on the dictionary atoms penalizes the learning of noisy basis functions. We demonstrate through examples on free breathing cardiac data, that the proposed scheme results in superior image quality in comparison to the conventional blind CS scheme and methods with fixed dictionaries.

Huisu Yoon¹, Daniel Kim², Kyung Sang Kim¹, and Jong Chul Ye^1
1KAIST, Daejeon, Korea, ²The University of Utah, Salt Lake City, Utah, United States

Recently, many compressed sensing (CS) based algorithms have been developed for dynamic MR imaging applications by exploiting sparsity in temporal transform domain. For example, in k-t FOCUSS with motion estimation and compensation (ME/MC), when a high resolution reference frame is available, ME/MC is shown a quite effective sparsifying transforms. However, one of the limitations of ME/MC is that the energy of the residual measurement after motion compensation is significantly reduced compared to the original k-space measurement. Hence, a new reconstruction algorithm for motion residual is required that judiciously reconstructs geometrically meaningful features. One of main contributions of this paper is a novel patch-based signal processing algorithm for motion residual reconstruction that overcomes the limitation of the existing k-t FOCUSS with ME/MC. More specifically, we impose a non-convex patch-based low-rank penalty that exploits self-similarities within the residual images. This penalty is shown to favor capturing geometric features such as edges rather than reconstructing the background noises. To solve the resulting non-convex optimization problem, we propose a globally convergent concave-convex procedure (CCCP)2 using convex conjugate, which has closed form solution at each sub-iteration.

Yanhua Wang¹ and Leslie Ying^1
1Department of Biomedical Engineering, Department of Electrical Engineering, The State University of New York at Buffalo, Buffalo, New York, United States

We propose a patch-based dictionary learning model for dynamic MRI reconstruction. The image sequence is divided into overlapping patches along both spatial and temporal directions. A set of temporal dependent dictionaries with three-dimensional atoms are adopted to provide sparse representations for compressed sensing reconstruction. This model adapts to specific local spatial-temporal features. Results on cardiac cine dataset demonstrate that the proposed method is capable of preserving both spatial structures and temporal variations.

Qiu Wang¹, Jun Liu¹, Nirmal Janardhanan¹, and Mariappan S. Nadar^1
1Imaging and Computer Vision, Siemens Corporation, Corporate Technology, Princeton, NJ, United States

Dynamic cardiovascular MRI facilitates the assessment of the structure and function of the cardiovascular system. To fit the data acquisition time, the data must be highly undersampled. Compressed sensing or sparsity based MR reconstruction takes advantage of the fact that the image is compressible in some transform domain, and enables reconstruction based on under-sampled k-space data thereby reducing the acquisition time. The design of such transform is a key to the success of the reconstruction. In this paper, we propose to use tight frame learning for computing data-driven transforms. Empirical results demonstrate improvement over the transform associated with the redundant Haar Wavelets.

Jose Caballero¹, Anthony Price², Daniel Rueckert¹, and Joseph V. Hajnal^2
1Department of Computing, Imperial College London, London, United Kingdom, ²Division of Imaging Sciences and Biomedical Engineering Department, King's College London, London, United Kingdom

The reconstruction of MR data from undersampled observations has been studied as a solution to the acceleration of MR acquisition and shown to have great potential. Nonetheless, exploration of sparsity models has been somewhat limited, particularly in the case of dynamic MRI. Here we propose a combination of dictionary learning techniques and temporal gradient sparsity for the reconstruction of cardiac cine sequences. A comparison with an established method enforcing x-f support sparsity shows the benefits of carefully choosing a model. The technique presented is able to reconstruct the full complex data with an independent treatment of real and imaginary components.

Jerome Yerly^1,2, Mari Elyse Boesen^2,3, Michel Louis Lauzon^2,4, and Richard Frayne^2,4
1Electrical and Computer Engineering, University of Calgary, Calgary, AB, Canada, ²Seaman Family MR Research Centre, Foothills Medical Centre, Calgary, AB, Canada, ³Physics and Astronomy, University of Calgary, Calgary, AB, Canada, ⁴Radiology and Clinical Neurosciences, University of Calgary, Calgary, AB, Canada

Prospectively gated FSE sequence for time-resolved imaging of physiological motion is often prohibitively lengthy. In this study, we propose to use a sparse SENSE approach with spatial and temporal constraints to reconstruct dynamic retrospectively gated undersampled FSE carotid images. Our spatiotemporal sparse SENSE reconstruction yielded readily interpretable and quantifiable vessel wall motion. We measured a maximum change in diameter of approximately 0.5 mm over a cardiac cycle. Our approach enables time-resolved characterization of the motion of the carotid vessel wall. To measure small changes more precisely, however, an increase in spatial resolution is necessary.

Xun Jia¹, Yifei Lou², and Jiang Du^3
1Radiation Medicine and Applied Sciences, University of California, San Diego, La Jolla, CA, United States, ²Mathematics, University of California, Irvine, Irvine, CA, United States, ³Radiology, University of California, San Diego, San Diego, CA, United States

Regular two-dimensional (2D) UTE imaging sequences employ half pulse excitations followed by radial ramp sampling. The 2D UTE sequence requires the summation of two acquisitions which adds a significant penalty to the total scan time. 3D UTE imaging employs a short rectangular pulse excitation followed by 3D radial ramp sampling, which is even more time consuming. Both 2D and 3D UTE sequences typically require undersampling to reduce scan time, but this produces streak artifacts. Recent advances in compressive sensing (CS) permits data recovery from extremely under-sampled measurements. In this study we aimed to reconstruct UTE image from the understampled data using a tight-frame (TF) regularized compressive sensing (TF-CS) technique.

Kai Tobias Block¹, Robert Grimm², Li Feng³, Ricardo Otazo¹, Hersh Chandarana¹, Mary Bruno¹, Christian Geppert⁴, and Daniel K. Sodickson^1
1Center for Biomedical Imaging, NYU Langone Medical Center, New York, NY, United States, ²Pattern Recognition Lab, University of Erlangen-Nuremberg, Erlangen, Germany, ³Center for Biomedical Imaging, New York University Langone Medical Center, New York, NY, United States, ⁴Siemens Medical Systems, New York, NY, United States

Compressed sensing is a promising technique for MRI reconstruction from incomplete data, but the feasibility and reliability for clinical routine applications have not been verified. Here, we present a prototypic setup that allows evaluating a recently proposed CS approach for motion-robust dynamic T1-weighted imaging in daily patient exams. It consists of a 3D stack-of-stars GRE sequence, a fully-automatic service for raw-data transfer to a multi-core server, and a highly parallelized implementation of the reconstruction algorithm. Images are saved in DICOM format and forwarded to the PACS archive, which enables our radiologists to read the images along with the conventional exams.

Joshua D. Trzasko¹, Yunhong Shu¹, Armando Manduca¹, John Huston III¹, and Matthew A. Bernstein^1
1Mayo Clinic, Rochester, MN, United States

In this work, we describe a sparsity-driven reconstruction framework for single-phase (non-Cartesian) SWIRLS 3D CE-MRA that incorporates both sensitivity encoding and off-resonance correction. As demonstrated, the proposed framework substantially reduces both noise amplification and geometric distortion that are routinely present in, and can compromise the diagnostic utility of, standard gridding reconstruction images.

Jana Hutter^1,2, Andreas Greiser³, Robert Grimm¹, Christoph Forman^2,4, Joachim Hornegger^1,2, and Peter Schmitt^5
1Pattern Recognition Lab, Friedrich-Alexander-Universitaet Erlangen-Nuernberg, Erlangen, Germany, ²School of Advanced Optical Technologies, Erlangen, Germany, ³Siemens AG, Erlangen, Germany, ⁴Friedrich-Alexander-University Erlangen-Nürnberg, Erlangen, Germany, ⁵MR Application & Workflow Development, Siemens Healthcare, Erlangen, Germany

The proposed compressed sensing based algorithm for velocity-encoded phase contrast MRA combines an incoherent interleaved undersampling pattern in both velocity-encoding and temporal dimension with a flow-adapted temporal resolution. High undersampling factors are supported while preserving the contrast and temporal resolution. Results are compared with state-of-the-art Grappa and Sense methods and the fully sampled reference.

Qiu Wang¹, Jun Liu¹, Zhili Yang¹, Nicolas Chesneau¹, Michael O. Zenge², Michaela Schmidt², 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

High temporal resolution is often desired in Cardiac Magnetic Resonance Imaging (CMRI). Compressed sensing has enables the reconstruction with a reduced the number of acquired frequencies, hence accelerating the acquisition. Spatial-temporal regularization has been proven effective for enforcing the smoothness. However, the fine details of the heart such as valve leaflets can be eliminated by strong regularization. In this work, a new approach was proposed for dynamic cardiac MRI reconstruction by setting the motion-dependent L1 regularization. Experiments conducted on CMRI data demonstrate the effectiveness of the proposed approach in preserving fine details of the heart.

Samuel T. Ting¹, Rizwan Ahmad¹, Yu Ding¹, Hui Xue², Lee C. Potter¹, and Orlando P. Simonetti^1
1The Ohio State University, Columbus, Ohio, United States, ²Siemens Corporate Research, Princeton, New Jersey, United States

We propose shrinkage SPIRiT (S-SPIRiT), an application of the fast iterative shrinkage-thresholding algorithm (FISTA) to SPIRiT that results in an L1-regularized implementation of SPIRiT that is more efficient than typical nonlinear conjugate gradient (NLCG) approaches and exhibits robustness to suboptimal parameter tuning and presence of noise. This approach may be especially applicable to cardiac magnetic resonance imaging, where kernel mismatch due to breathing motion can impact image quality.

Fan Lam^1,2, Bo Zhao^1,2, Michael Weiner^3,4, Norbert Schuff^3,4, and Zhi-Pei Liang^1,2
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 propose a new method to jointly denoise a sequence of noisy images typically acquired in quantitative imaging experiments. The proposed method uses a penalized maximum likelihood estimation formalism, integrating two modeling constraints: a low-rank model that captures any correlation in the edge structures from one frame to another and a penalty function that promotes sparse edge structures. A computationally efficient algorithm is developed to solve the associated optimization problem. Representative results from a parametric mapping experiment are presented to demonstrate the performance of the proposed method.

Yeyang Yu¹, Jin Jin¹, Feng Liu¹, Stuart Crozier¹, and Mingjian Hong^2
1The School of Information Technology and Electrical and Engineering, The University of Queensland, Brisbane, QLD, Australia, ²School of Software Engineering, Chongqing Univeristy, Chongqing, Chongqing, China

In this work, we introduce the conception of tensor sparsity for Compressed Sensing dynamic MRI. Conventionally, the spatial and temporal information were then sparsified independently and sequentially. Therefore the spatial-temporal correlation may not be sufficiently exploited. This work applys the Tucker model based Higher-order Singular Value Decomposition (HOSVD) in the Compressed Sensing dynamic MRI framework. Instead of treating the 3D/4D data as series of 2D images, HOSVD inherits the high-dimensional data format, leading to significantly improved dMRI reconstructions compared with those well-established CS-dMRI methods. The advantages of the tensor sparsity in terms of reconstruction accuracy have been demonstrated in a given cardiac dynamic MRI study.

Samuel T. Ting¹, Yu Ding¹, Rizwan Ahmad¹, Hui Xue², and Orlando P. Simonetti^1
1The Ohio State University, Columbus, Ohio, United States, ²Siemens Corporate Research, Princeton, New Jersey, United States

We seek to address the ill-conditioned properties of the SPIRiT reconstruction method when applied in scenarios requiring highly accelerated acquisitions, as is the case in real-time cardiac imaging under free-breathing conditions. We show that the use of a suitable data initialization in the POCS implementation of SPIRiT can help alleviate the issues resulting from the ill-conditioned properties of SPIRiT.

Ying-Hua Chu¹, Shang-Yueh Tsai², Yi-Cheng Hsu³, Wen-Jui Kuo⁴, and Fa-Hsuan Lin^1,5
1Institute of Biomedical Engineering, National Taiwan University, Taipei, Taiwan, ²Graduate Institute of Applied Physics, National Cheng-Chi University, Taipei, Taiwan, ³Department of Mathematics, Nnational Taiwan University, Taipei, Taiwan, ⁴Institute of Neuroscience, National Yang-Ming University, Taipei, Taiwan, ⁵Department of Biomedical Engineering and Computational Science, Aalto University, Espoo, Finland

We exploit the redundancy among channels of a receiver coil array to improve the SNR of DTI. Our method uses a universal kernel to enforce the data-consistency (DC) among k-space data across receiver coils. This DC constraint was then applied to all diffusion-weighted images to suppress noise disturbing the data consistency required by the parallel MRI theory. Experimental results at 3T with b = 4,000 s/mm2 demonstrate that the SNR can be improved by approximately 40% by applying this constraint to DTI reconstructions.

Alessandro Sbrizzi¹, Alexander Raaijmakers¹, Cornelis A.T. van den Berg¹, Jan J.W. Lagendijk¹, Peter R. Luijten¹, and Hans Hoogduin^1
1Imaging Division, UMC Utrecht, Utrecht, Utrecht, Netherlands

It is shown how the fields derived upon singular value decomposition are related to the true receive and transmit fields maps of a pTx coil array. TRIPLET produces all necessary RF field maps for a parallel transmit and parallel imaging experiment in a very short time and with use of minimal RF power. In addition, the methods results in receive sensitivity maps per channel without the need for a homogeneous receive reference. The method is generic and will also work in case of a single transmit channel combined with several receive channels.

Mingyan Li¹, Jin Jin¹, Feng Liu¹, Adnan Trakic¹, and Stuart Crozier^1
1School of Information Technology and Electrical Engineering, The University of Queensland, Brisbane, Queensland, Australia

With the ability of encoding larger number of sensitivity profiles, the novel 4-element rotating radiofrequency coil array (RRFCA) has better imaging acceleration performance than stationary phased-array coils (PACs). However, with more time-varying sensitivity encoding, the system matrix of RRFCA is more complex and leads to a longer reconstruction time. In this work, while maintaining the same acceleration capability, the fast hybrid image reconstruction strategy reduces the number of sensitivity encodings and obtains the fast initial estimation to improve the efficiency and accuracy of accelerated image reconstruction with RRFCA.

Kang Wang¹, Tao Zhang², Philip J. Beatty³, Dan W. Rettmann⁴, Ersin Bayram⁵, and James H. Holmes^1
1Global Applied Science Laboratory, GE Healthcare, Madison, WI, United States, ²Electrical Engineering, Stanford University, Stanford, CA, United States,³Sunnybrook Research Institute, Toronto, ON, Canada, ⁴Global Applied Science Laboratory, GE Healthcare, Rochester, Minnesota, United States, ⁵GE Healthcare, Waukesha, WI, United States

Auto-calibrating parallel imaging (acPI) methods have advantages over physically-modeled methods in reduced FOV applications or when it is difficult to accurately measure coil sensitivity maps, such as breath-hold exams. However, for challenging clinical protocols that use large channel counts, big matrix sizes and high parallel imaging factors, conventional channel-by-channel acPI methods may still have long reconstruction latency. To address this issue, Coil Compression (CC) and Direct Virtual Coil (DVC) techniques have been proposed independently. This work is to demonstrate the feasibility of combining the two techniques to achieve even higher reduction in computation without compromise in image quality.

Bernd A. Jung¹, Michael Markl², Pegah Entezaril², Riti J. Mahadevia², and Susanne Schnell^2
1Dept. of Radiology, Medical Physics, University Medical Center, Freiburg, Germany, ²Dept. of Radiology, Feinberg School of Medicine, Northwestern University, Chicago, Illinois, United States

The purpose of this study was to evaluate the utility of k-t parallel imaging for the acceleration of aortic 4D-flow MRI by systematically investigating the impact of different acceleration factors R and coil elements on quantification of hemodynamics parameters such as blood flow and wall shear stress. K-t accelerated 4D-flow MRI was performed in 10 healthy volunteers with R=3, 5 and 8 using a 12-channel and a 32-channel receiver coil and compared to conventional parallel imaging (GRAPPA, R=2). Data analysis included 3D blood flow visualization and quantification of peak velocities, flow rates and wall shear stress in different aortic locations.

Derya Gol¹ and Lee C. Potter^1
1Electrical & Computer Engineering and Davis Heart & Lung Institute, The Ohio State University, Columbus, OH, United States

Interpolation approaches in parallel MRI exhibit a noise amplification effect that may be mitigated via regularization techniques which are computationally expensive. In this study, we propose a pre-processing technique based on structured low-rank matrix approximation via truncated singular value decomposition (TSVD), which is able to suppress noise and ghost artifacts efficiently. TSVD method has been previously used in parallel MRI to improve the conditioning of the system matrix and to reconstruct k-space via matrix completion. In contrast to previous work, here rank properties are used to denoise acquired data in a computationally simple preprocessing for GRAPPA reconstruction.

Thomas A. Depew¹ and Qing-San Xiang^1,2
1Physics & Astronomy, University of BC, Vancouver, BC, Canada, ²Radiology, University of BC, Vancouver, BC, Canada

Partial Parallel Imaging (PPI) is a popular choice for performing accelerated acquisitions in MRI. PPI algorithms that use calibration and fitting to recreate the omitted k-space signals are candidates for using the Differential Energy (E_D) as a metric for optimizing reconstruction parameters. E_D provides a measure of the accuracy of the reconstruction and can be calculated without a fully sampled reference image. The E_D reliably predicted optimal RW for a regionally tuned GRAPPA algorithm (r-GRAPPA).

Felix A. Breuer¹, Simon Bauer², and Martin Blaimer^1
1Research Center Magnetic Resonance Bavaria, Würzburg, Bavaria, Germany, ²Siemens Healthcare, Erlangen, Bavaria, Germany

In this work, the concept of CAIPIRINHA is applied to 3D stack-of-stars (SOS) VIBE liver imaging. It is demonstrated that by rotating the stars by Δφ with respect to each other significantly improved parallel imaging performance is achieved compared to standard acceleration in only the radial direction. pMRI reconstruction was performed using a novel self-calibrating segmented radial GRAPPA algorithm tailored to the individual CAIPIRINHA undersampling schemes eliminating the need of an extra calibration scan. This allows for robust high resolution 3D liver imaging within a 12s breath-hold.

Arjun Arunachalam^1
1Electrical Engineering, Indian Institute of Technology Bombay, Mumbai, Maharashtra, India

A method for restoring aliased, sub-sampled k-space for a multiplicative increase in scan acceleration is presented. Recently, the RATE method for accelerating dynamic MRI by acquiring and then restoring aliased k-space through a Fourier transformation was presented. In this work, it’s shown that for normal and dynamic MRI, receiver sensitivities can be used to restore aliased k-space. Also, if aliased k-space is sub-sampled, Parallel Imaging is used first to synthesize the un-acquired samples. Next, the coil sensitivities are used again with the acquired, synthesized samples in an iterative reconstruction to enable acceleration factors as high as 12 for 2D scans.

Yu Ding¹, Rizwan Ahmad¹, Hui Xue², Samuel T. Ting¹, Ning Jin³, and Orlando P. Simonetti^1
1The Ohio State University, Columbus, OH, United States, ²Corporate Research, Siemens Corporation, Princeton, NJ, United States, ³Siemens Healthcare, Columbus, OH, United States

SENSE is a widely used parallel imaging technique. The so-called g-factor represents how noise in the raw data affects the noise in the reconstructed image. However, the g-factor calculation does not take into account the noise in the channel sensitivity map. In this abstract, we present a noise transfer model in SENSE reconstruction that takes into account the noise in both the raw data and the channel sensitivity map. A phantom study showed that the model has satisfactory accuracy. We observed that a large portion (> 35%) of the image noise originates from the noise in the sensitivity map.

Stefan Kroboth¹, Frederik Testud², Kristian Bredies³, Kelvin J. Layton⁴, Daniel Gallichan⁵, Chris A. Cocosco², Gerrit Schultz², Florian Knoll¹, Christoph Barmet^6,7, Klaas P. Pruessmann⁸, Maxim Zaitsev², and Rudolf Stollberger^1
1Institute of Medical Engineering, Graz University of Technology, Graz, Austria, ²Medical Physics, University Medical Center Freiburg, Freiburg, Germany,³Institute for Mathematics and Scientific Computing, University of Graz, Graz, Austria, ⁴Department of Electrical and Electronic Engineering, University of Melbourne, Parkville, Victoria, Australia, ⁵LIFMET, Ecole Polytéchnique Fédérale de Lausanne, Lausanne, Switzerland, ⁶Institute for Biomedical Engineering, University and ETH Zurich, Zürich, Switzerland, ⁷Skope Magnetic Resonance Technologies, Zürich, Switzerland, ⁸Institute for Biomedical Engineering, University of Zürich and ETH Zürich, Zürich, Switzerland

This work investigates the reconstruction of single shot North-West Echo Planar Imaging (NW-EPI) data with estimated trajectories using field cameras. NW-EPI is a multidimensional trajectory designed to improve the resolution in a selected region by exploiting the spatially varying resolution characteristic of nonlinear PatLoc fields. Reconstruction with Total Generalized Variation (TGV) was successfully applied to PatLoc imaging with 2D trajectories to reduce noise, Gibbs ringing and undersampling artifacts. However, for multidimensional trajectories, TGV converges very slowly. We therefore present a new concept of numerically solving the inverse problem, called Total Generalized Variation - Conjugate Gradient (TGV-CG) to increase convergence speed.

Holden H. Wu^1,2 and Dwight G. Nishimura^3
1Radiological Sciences, UCLA, Los Angeles, CA, United States, ²Cardiovascular Medicine, Stanford University, Stanford, CA, United States, ³Electrical Engineering, Stanford University, Stanford, CA, United States

A new real-time cardiac MRI technique combining the advantages of spiral imaging with flexible golden-ratio acquisition ordering and acceleration from self-consistent parallel imaging is presented in this work. The scan acceleration achieved by the proposed technique is used to acquire free-breathing real-time images at a relatively high in-plane resolution of 1.6 mm and expanded field of view of 40 cm. Adequate temporal resolution is maintained to visualize cardiac and respiratory motion.

Benjamin Paul Berman¹, Zhitao Li², Maria I. Altbach³, Jean-Philippe Galons³, Diego R. Martin³, Bin Dong⁴, Puneet Sharma³, Bobby T. Kalb³, and Ali Bilgin^2,5
1Applied Mathematics, University of Arizona, Tucson, Arizona, United States, ²Electrical and Computer Engineering, University of Arizona, Tucson, Arizona, United States, ³Medical Imaging, University of Arizona, Tucson, Arizona, United States, ⁴Mathematics, University of Arizona, Tucson, Arizona, United States, ⁵Biomedical Engineering, University of Arizona, Tucson, Arizona, United States

There is constant demand for high quality images and short data acquisition times for MRI. Traditionally, a choice is made for one or the other, but due to the development of CS theory, it is often possible to have both. For 3D MR imaging, various trajectories are used to undersample in Fourier space including Cartesian, radial, and cylindrical. The cylindrical trajectory – or stack-of-stars – is used for dynamic and motion sensitive 3D imaging. For CS it is crucial that each measurement contain as much information as possible. We show that the 3D stack-of-stars trajectory benefits from a center-dense sampling scheme.

Daniel Kopeinigg¹, Murat Aksoy¹, Samantha J. Holdsworth¹, Rafael O'Halloran¹, and Roland Bammer^1
1Center for Quantitative Neuroimaging, Department of Radiology, Stanford University, Stanford, CA, United States

An iterative POCS algorithm for reconstructing arbitrary 3D k-space data is introduced and applied to undersampled 3D Cones trajectories. Our results indicate that the algorithm can greatly reduce aliasing artifacts after only 3-5 iterations.

Ali Ersoz¹, Volkan Emre Arpinar², and L. Tugan Muftuler^2,3
1Department of Biophysics, Medical College of Wisconsin, Milwaukee, WI, United States, ²Department of Neurosurgery, Medical College of Wisconsin, Milwaukee, WI, United States, ³Center for Imaging Research, Medical College of Wisconsin, Milwaukee, WI, United States

We developed a new technique, named Highly Accelerated Projection Imaging (HAPI) with coil sensitivity encoding, which is capable of reconstructing a 2D image using fewer projections than the previous reports. The essence of this new technique is to sample each spoke more densely than the conventional scheme. The feasibility of this new technique was investigated with realistic simulations and experimental phantom studies. Simulation results show that 1-2 projections might be sufficient to reconstruct a 2D image. Experimental results demonstrated that HAPI is a promising new technique for fast imaging.

Kang Wang¹, Ann Shimakawa², Kevin M. Johnson³, Philip J. Beatty⁴, Oliver Wieben^3,5, and James H. Holmes^1
1Global Applied Science Laboratory, GE Healthcare, Madison, WI, United States, ²Global Applied Science Laboratory, GE Healthcare, Menlo Park, CA, United States, ³Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, ⁴Sunnybrook Research Institute, Toronto, ON, Canada, ⁵Radiology, University of Wisconsin-Madison, Madison, WI, United States

Multi-channel non-Cartesian phase contrast imaging has been demonstrated with great capability for various 4D flow MRI applications. However, reconstruction latency has been challenging, as well as the need for a robust phase preserving coil combination algorithm. Among several recently reported algorithms, the Direct Virtual Coil (DVC) technique provides great potential to address both challenges simultaneously. This work is to demonstrate the usage of the DVC technique for non-Cartesian phase-contrast imaging, specifically, PC-VIPR.

Jing Li¹, Lin Chen¹, Congbo Cai², Shuhui Cai¹, and Zhong Chen^1
1Department of Electronic Science, Xiamen University, Xiamen, Fujian, China, ²Department of Communication Engineering, Xiamen University, Xiamen, Fujian, China

Last decades have witnessed a continuous growth of single-scan MRI, both in scientific research and clinical application. Spatiotemporally encoded (SPEN) MRI has an intrinsic immunity to field inhomogeneity. However, its phase encoding dimension is always sub-Nyquist sampling for ultrafast acquisition, which leads to aliasing artifacts. In this study, we propose a spiral sampling method for two-dimensional SPEN single-scan MRI. It can not only maintain the advantages of SPEN method, but also eliminate the aliasing artifacts due to undersampling. In combination with a super-resolved reconstruction based on the singular value decomposition, images with good quality can be obtained.

Xu Han¹, Katherine L. Wright¹, Vikas Gulani^2,3, and Nicole Seiberlich^1
1Dept. of Biomedical Engineering, Case Western Reserve University, Cleveland, Ohio, United States, ²Dept. of Radiology, Case Western Reserve University, Cleveland, Ohio, United States, ³Dept. of Radiology, University Hospitals, Cleveland, Ohio, United States

The goal of this work is to demonstrate that undersampled golden angle radial data can be reconstructed using through-time radial GRAPPA. The golden-angle trajectory is advantageous when it is not clear what acceleration factor is desired, as any subset of temporally adjacent k-space lines can be used for the reconstruction. This abstract describes two approaches for through-time radial GRAPPA weight calibration: using an additional sequentially ordered golden angle calibration dataset and a self-calibrating golden angle technique. Using these formulations, real-time free-breathing cardiac images can be reconstructed at retrospectively selected temporal resolutions from the same golden angle dataset.

Johannes F.M. Schmidt¹, Lukas Wissmann¹, Robert Manka^1,2, and Sebastian Kozerke^1,3
1Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland, ²University Hospital Zurich, Zurich, Switzerland, ³Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom

In this work, an iterative k-t PCA algorithm is proposed where an additional spatial transformation is used to further sparsify the data. Training data based regularization is performed in a motion corrected x-pc domain where each time frame is warped to a reference respiratory position. Spatial transformations are derived from frame-by-frame composite images using atlas-based image registration. Using 3D perfusion data acquired in vivo it is demonstrated that this approach successfully corrects for incomplete unfolding due to respiratory bulk motion.

Jesse I. Hamilton¹, Katherine L. Wright¹, Mark A. Griswold², and Nicole Seiberlich^1
1Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, ²Radiology, Case Western Reserve University, Cleveland, OH, United States

Despite the high acceleration factors that are achievable with through-time non-Cartesian GRAPPA, the major limitation of this method is the need for a separate fully-sampled calibration scan. Here we present a self-calibrating method for through-time non-Cartesian GRAPPA using interleaved trajectories. In free breathing, non-gated cardiac scans, the self-calibrating reconstruction with both interleaved radial and interleaved variable density spiral trajectories produced images with higher quality and less blurring compared to view-sharing without parallel imaging reconstruction. Due to its shorter temporal footprint, less blurring was observed with self-calibrating through-time non-Cartesian GRAPPA using an undersampled spiral trajectory than the radial trajectory.

Stefan Wundrak^1,2, Jan Paul¹, Johannes Ulrici², Erich Hell², and Volker Rasche^1
1Klinik für Innere Medizin II, Ulm University, Ulm, Germany, ²Sirona Dental Systems, Bensheim, Germany

We introduce a retrospective self-gated MRI reconstruction method for the imaging of the temporomandibular joint dynamics. Compared to real-time imaging the proposed technique shows an improved tempo-spatial resolution and allows for the first time the imaging of the TMJ dynamics under a continuous mastication.

Haris Saybasili¹, Daniel A. Herzka², Kestutis Barkauskas³, Nicole Seiberlich³, and Mark A. Griswold^1
1Radiology, Case Western Reserve University, Cleveland, OH, United States, ²Biomedical Engineering, Johns Hopkins University, Baltimore, MD, United States, ³Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States

In the recent years, many research oriented, customizable, external MR image reconstruction frameworks have been presented. To the best of our knowledge, none of these frameworks provided fully automated, remotely and locally distributed (multi-node, and multi-GPU) image reconstruction capabilities. Additionally, these frameworks may depend on high-level software libraries that make it difficult to maintain, debug and update the existing code. In this work, we present a highly customizable, automatically distributed, multi-threaded image reconstruction environment, built using only low-level system libraries for improved performance and portability. Our framework utilizes multiple GPUs and multiple workstations (nodes) by transparently distributing reconstruction tasks.  

Hidenori Takeshima¹, Shuhei Nitta¹, Taichiro Shiodera¹, Tomoyuki Takeguchi¹, Masao Yui², and Shigehide Kuhara^2
1Corporate Research & Development Center, Toshiba Corporation, Kawasaki, Kanagawa, Japan, ²MRI Systems Division, Toshiba Medical Systems Corporation, Otawara-shi, Tochigi, Japan

The number of calibration samples is relatively high when the number of phase encodes is small or when the reduction factor is high. To reduce the acquisition time, we have developed a novel method for estimating k-t coil sensitivity maps from only the k-t data to be reconstructed. The images reconstructed by the proposed method are as clear as those reconstructed by k-t SENSE. The processing time of the proposed method is about 11.6 seconds for data scanned with reduction factor R = 4, 256 readout encoding steps, 96/R phase encoding steps, 32 coils, and 96 frames.

Xu Li^1,2, Issel Anne L. Lim^1,2, Craig K. Jones^1,2, and Peter C.M. van Zijl^1,2
1F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, Maryland, United States, ²Radiology, Johns Hopkins University School of Medicine, Baltimore, Maryland, United States

We propose regularization strategies for susceptibility tensor imaging (STI) based on a combination of structural information and the values of mean magnetic susceptibility (MMS) and magnetic susceptibility anisotropy (MSA). The goal is to generate color maps of fiber tracts using MR phase data from a small number of orientations. Human brain imaging data acquired at 7 Tesla using six head orientations with rotation angles of absolute magnitude 5° to 15° show that the principle eigenvector (PEV) map obtained from the regularized STI method resembles the PEV map from DTI, whereas the PEV map from the non-regularized STI method does not.

Hongchen Wang¹, Xavier Maître¹, Jean-Pierre Ruaud¹, Luc Darrasse¹, and Ludovic De Rochefort^2
1Imagerie par Résonance Magnétique Médicale et Multi-Modalités (UMR8081) IR4M, CNRS, Univ. Paris-Sud, Orsay, France, ²Univ Paris-Sud, Orsay, France

Apparent transverse relaxation rate R2* and magnetic susceptibility χ are distinct measurable parameters that can be used to quantify paramagnetic and superparamagnetic materials. R2* mapping requires several echoes times in a gradient-echo scan and is based on the analysis of signal amplitude decrease. Magnetic Susceptibility mapping uses field mapping that can be measured from the same multi-echo dataset using phase images. Here, phantom experiments are done to compare the precision of R2* and susceptibility methods for quantifying contrast agent concentration.

Junmin Liu¹, David A. Rudko^1,2, Joseph S. Gati¹, Ravi S. Menon^1,2, and Maria Drangova^1,2
1Imaging Research Laboratories, Robarts Research Institute, Shulich School of Medicine & Dentistry, University of Western Ontario, London, ONtario, Canada, ²Department of Medical Biophysics, Shulich School of Medicine & Dentistry, University of Western Ontario, London, Ontario, Canada

When a frequency mask (FM) is used for generating multi-echo susceptibility-weighted images, a cutoff frequency value (fth) must be defined for the frequency image at each echo. In this work, we present a new FM generation method which determines fth by using the criterion 2π×|fth|×TE =π. The performance of the proposed technique is evaluated with a set of volunteer brain images acquired at 7 T, and the results are compared with the standard frequency normalization approach, which uses the lowest frequency value in the image as fth. The proposed method produces a more robust FM and higher contrast SW images.

Amanda C. L. Ng^1,2, David K. Wright^3,4, Parnesh Raniga^2,5, Stephen Moore⁶, Gary F. Egan², and Leigh A. Johnston^4,7
1Dept of Electrical & Electronic Engineering, The University of Melbourne, Melbourne, VIC, Australia, ²Monash Biomedical Imaging, Monash University, Melbourne, VIC, Australia, ³Centre for Neuroscience, The University of Melbourne, Melbourne, VIC, Australia, ⁴Florey Institute of Neuroscience and Mental Health, The University of Melbourne, VIC, Australia, ⁵The Australian e-Health Research Centre-BioMedIA The Australian e-Health Research Centre-BioMedIA, CSIRO Preventative Health National Research Flagship ICTC, Herston, QLD, Australia, ⁶IBM Research Collaboratory for Life Sciences-Melbourne, Victorian Life Sciences Computing Initiative, The University of Melbourne, VIC, Australia, ⁷NeuroEngineering Laboratory, Dept. Electrical & Electronic Engineering, The University of Melbourne, Melbourne, VIC, Australia

Quantitative Susceptibility Mapping (QSM) aims to derive reliable estimates of the magnetic susceptibility of voxels from phase data arising from 3D gradient-echo MRI acquisitions. Current methods compute the contribution of all tissue types using a spherical model, however research has shown that white matter is better modelled as cylinders. We present a method for deriving susceptibility maps from MRI phase data that combines both spherical and cylindrical models. We use MR diffusion data to identify voxels best modelled as cylinders and to determine the orientation of those cylinders. Our results demonstrated improved accuracy and robustness compared to conventional methods.

Wei Li¹, Bing Wu², and Chunlei Liu^1,3
1Brain Imaging & Analysis Center, Duke University, Durham, North Carolina, United States, ²GE Healthcare China, Beijing, China, ³Radiology, Duke University, Durham, North Carolina, United States

The dependence of R2* on fiber orientation was evaluated using multi-orientation brain images. The significant derivation from sine squared relationship between R2* and fiber orientation indicated the contribution of anisotropic magnetic susceptibility. Assuming a constant ratio of -0.4 between isotropic and anisotropic susceptibility contributions, R2* anisotropy were estimated on a voxel-by-voxel basis. R2* anisotropy due to anisotropic susceptibility accounted for approximately 10~30% of the total R2* of white matter, and is highly correlated with diffusion anisotropy. These results suggested the importance of anisotropic magnetic susceptibility induced R2* variations and its value for the study of white matter microstructure and composition.

Hongchen Wang¹, Jean-Sébastien Raynaud², and Ludovic De Rochefort^3
1Univ. Paris Sud - CNRS, UMR 8081, IR4M, Orsay, France, ²Experimental Imaging, MRI unit, Research Division, Guerbet, Aulnay-sous-bois, France, ³Univ Paris-Sud, Orsay, France

Quantitative susceptibility mapping (QSM) provides a way to quantify bulk magnetic susceptibility distribution from field inhomogeneity images. It involves phase unwrapping, background filtering and source reconstruction. QSM reconstruction with spatial priors can be performed to solve this ill-posed problem. However, the iterative procedure is limited by a long calculation time, and multiple iterations are needed to choose adequate regularization parameters. Here we accelerate QSM using a rapid estimation in k-space and introduce a general and efficient parameter search method.

Eugene Kim¹, B. Douglas Ward², and Arvind P. Pathak^3
1Department of Biomedical Engineering, Johns Hopkins University, Baltimore, MD, United States, ²Department of Biophysics, Medical College of Wisconsin, Milwaukee, WI, United States, ³Russell H. Morgan Department of Radiology and Radiological Science, The Johns Hopkins University School of Medicine, Baltimore, MD, United States

The finite perturber method (FPM) and Monte Carlo simulations were used to compute steady-state susceptibility contrast (SSC)-MRI biomarkers of fractional blood volume (FBV), vessel size (VSI), and vessel density (N) for tumor vasculature from μCT data and for an ensemble of randomly oriented cylinders (RC). For the μCT data, the correlations between simulated and true biomarker values were lower and the median errors greater compared to the RC data, indicating that vascular morphology significantly affects the accuracy of these biomarkers. The FPM can be used to elucidate how various biophysical factors affect the SSC-MRI signal and help develop more accurate imaging biomarkers of angiogenesis.

Bo Xu¹, Tian Liu², Pascal Spincemaille², Nanda Deepa Thimmappa², Martin R. Prince², and Yi Wang^1
1Biomedical Engineering, Cornell University, Ithaca, New York, United States, ²Weill Cornell Medical College, New York, New York, United States

A dynamic quantitative susceptibility imaging method is presented for 3D imaging of Gadolinium concentration at sub-second frame rate. A multi-echo flow compensated spiral gradient echo is used for data acquisition. Temporal resolution acceleration with constrained evolution reconstruction is used to generate high temporal frame rate. A morphology enabled dipole inversion with nonlinear formulation is used to generate quantitative susceptibility mapping from complex frame images.

Diego Hernando¹ and Scott B. Reeder^1,2
1Radiology, University of Wisconsin-Madison, Madison, WI, United States, ²Medical Physics, University of Wisconsin-Madison, Madison, WI, United States

MR-based quantification of liver magnetic susceptibility may enable field strength-independent quantification of liver iron concentration (LIC), for diagnosis, staging and treatment monitoring of hepatic iron overload. However, susceptibility measurement is challenging, due to the non-local effect of the susceptibility distribution on the measured B0. The purpose of this work is to demonstrate feasibility of MR-based LIC quantification using a fat-referenced approach to estimate liver susceptibility from the measured B0 field map. The proposed method is validated at 1.5T and 3T using an R2-based LIC measurement (Ferriscan) as the reference standard.

Meng-Chi Hsieh^1,2 and Jyh-Horng Chen^1,2
1Institute of Biomedical Electronic and Bioinformatics, National Taiwan University, Taipei, Taiwan, ²Interdisciplinary MRI/MRS Lab, Department of Electrical Engineering, National Taiwan University, Taipei, Taiwan

Functional MRI measures brain activity and relative blood flow using blood-oxygen-level-dependent (BOLD) contrast. However, the mechanism of BOLD contrast is caused by intrinsic susceptibility change. A recent MRI approach, referred to as quantitative susceptibility mapping (QSM), has been proposed to have the potential to quantify the susceptibility within tissues, which may be also capable of improving the measurement of oxygenation-dependent susceptibility changes. To understand the relationship between tissue oxygenation and resulting susceptibility changes, in this study, we aimed to employ QSM technique on rats with different oxygenation levels and investigate the susceptibility changes within different regions of the brain.

Meng-Chi Hsieh^1,2, Jyh-Horng Chen^1,2, and Hon-Man Liu^3
1Institute of Biomedical Electronic and Bioinformatics, National Taiwan University, Taipei, Taiwan, ²Interdisciplinary MRI/MRS Lab, Department of Electrical Engineering, National Taiwan University, Taipei, Taiwan, ³Department of Medical Imaging, National Taiwan University Hospital and College of Medicine, Taipei, Taiwan

Cerebral hemorrhage is a common cerebrovascular disease in elderly people and associated with hypertension cerebral amyloid angiopathy. However, the amount of interacerebral microbleeding is strongly and dependently relate to incidence of cerebral hemorrhage and its effect on treatment decision. In this study, we accessed the quantity of microbleed using quantitative susceptibility mapping technique to provide a possible quantified approach for medical diagnosis.

Sung-Min Gho¹, Yoonho Nam¹, Dongyeob Han¹, and Dong-Hyun Kim^1
1Electrical and Electronic Engineering, Yonsei University, Seodaemun-gu, Seoul, Korea

Quantitative susceptibility mapping (QSM) methods determined from the image phase have been developed for description of tissue anatomy, structure and susceptibility. MR image and its phase value, however, represent the combined signal of sub-voxel components. In addition, image voxel signal at an echo times (TE) represents signal contributions of various T2 and/or T2* value components. For example, myelin, one of major sub-voxel components in white matter regions, is hard to obtain at long TEs because of its very short T2 relaxation times. A QSM, if determined at different TEs, therefore, would represent different characters even in the same voxels. We estimated the temporal variations of QSM images at different TEs using multi-echo gradient echo data.

Dongyeob Han¹, Yoonho Nam¹, Sung-Min Gho¹, and Dong-Hyun Kim^1
1Electrical & Electronic Engineering, Yonsei University, Seoul, Korea

Quantitative Susceptibility Mapping (QSM) based on phase data of gradient echo (GRE) is a novel technique for measuring susceptibility differences of the tissue. However, the macroscopic B0 inhomogeneity due to air/tissue boundary causes artifacts which is a common drawback of the GRE1. Furthermore, the effect of the macroscopic B0 inhomogeneity become worse as the echo time increases, and this is critical to QSM studies which have long echo times. To overcome this problem, we propose a 3D z-shim multi-echo GRE pulse sequence which can generate B0 inhomogeneity compensated QSM and a simple robust algorithm for B0 inhomogeneity compensated field map.

Shuai Wang¹, Tian Liu², Weiwei Chen³, Cynthia Wisnieff², Pascal Spincemaille², Apostolos John Tsiouris², and Yi Wang^2
1School of Electronic and Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan, China, ²Department of Radiology, Weill Cornell Medical College, New York, New York, United States, ³Department of Radiology, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, Hubei, China

To help understand various QSM methods, we propose the following categorization: Non-Bayesian approach with alteration of the dipole kernel or approximation of the dipole kernel to overcome ill condition, and Bayesian approach using a general mathematical prior or a specific physical structure prior. We also examine the noise modeling error in QSM methods.

Wolfgang Stefan¹, Erol Yeniaras¹, Ken-Pin Hwang², Linda Chi³, Ed Jackson¹, John D. Hazle¹, and R. Jason Stafford^1
1Imaging Physics, The University of Texas MD Anderson Cancer Center, Houston, Texas, United States, ²Applied Science Laboratory, Healthcare Technologies, WI, Hosuton, Texas, United States, ³Diagnostic & Radiology, The University of Texas MD Anderson Cancer Center, Houston, Texas, United States

We have developed a new background suppression technique that can be used to suppress the background in MR phase images and PRF measurements. In contrast to the most common approach of high pass filtering, our approach leaves the low frequency components of the tissue intact, which enables us to make quantitative measurements of the PRF. The method separates sparse structures in the PRF from the non-sprase background by means of non-convex optimization similar to compressed sensing.

Eo-Jin Hwang¹, Min-Ji Kim², Hyug-gi Kim³, Kyung-Mi Lee⁴, Ji-Seon Park⁵, Wook Jin², Dal-Mo Yang², and Geon-Ho Jahng^2
1Kyung Hee University Hospital at Gangdong, Seoul, Seoul, Korea, ²KyungHee University Hospital at GangDong, Seoul, Seoul, Korea, ³KyungHee University, YoungIn, Gyeonggi-do, Korea, ⁴Department of Radiology, Kyung Hee University Hospital, Seoul, Seoul, Korea, ⁵Kyung Hee University Hospital, Seoul, Seoul, Korea

The objective of this study was to apply voxel-based analyses to compare the susceptibility changes among three different groups, cognitive normal (CN), mild cognitive impairment (MCI), and Alzheimer’s disease (AD) using a quantitative susceptibility mapping (QSM) technique. The QSM images were produced by implementing a Morphology Enabled Dipole Inversion (MEDI) method, and the direct comparisons of the whole brains between CN and MCI, CN and AD, and MCI and AD subjects were performed. Throughout the study, the brain regions where different susceptibility effects existed were identified, and the effects of both paramagnetic and diamagnetic substances were visible in all comparisons.

Ryan Topfer^1,2, Ferdinand Schweser³, Andreas Deistung⁴, Alan H. Wilman^1,2, and Jürgen R. Reichenbach^3
1Department of Physics, University of Alberta, Edmonton, AB, Canada, ²Department of Biomedical Engineering, University of Alberta, Edmonton, AB, Canada, ³Medical Physics Group, Institute of Diagnostic and Interventional Radiology I, Jena University Hospital - Friedrich Schiller University, Jena, Thuringia, Germany, ⁴Medical Physics Group, Institute of Diagnostic and Interventional Radiology I, Jena University Hospital - Friedrich Schiller University Jena, Jena, Thuringia, Germany

SHARP and PDF, two recently proposed techniques for filtering B0 field (phase) maps, share a common pitfall: failure to accurately filter background field near the edges of the brain (e.g. cortex). This study presents an adaptation to conventional SHARP whereby the analyticity of the harmonic background field is exploited to recover local phase contrast (that which pertains to tissue magnetic susceptibility) of the hitherto lost edge voxels. The method is quantitatively assessed with a “field-forward” experiment and qualitatively demonstrated using in vivo data.

Yan Wen^1,2, Tian Liu^1,3, and Yi Wang^1,3
1Department of Radiology, Weill Cornell Medical College, New York, NY, United States, ²Department of Physics and Astronomy, SUNY at Stony Brook, Stony Brook, NY, United States, ³Department of Biomedical Engineering, Cornell University, Ithaca, NY, United States

A background removal procedure removes background fields to isolate local fields, which is essential for susceptibility calculation. A previous method based on spherical mean value (SMV) principle, Sophisticated Harmonic Artifact Reduction on Phase Data (SHARP), has shown promising results but its result depends on the radius of the sphere and requires a subjective selection of truncation threshold for the deconvolution procedure. Here, we present an infinite SMV algorithm to reduce the dependence of radius and eliminates the need of deconvolution.

Debra E. Horng^1,2, Diego Hernando¹, Samir D. Sharma¹, and Scott B. Reeder^1,2
1Radiology, University of Wisconsin-Madison, Madison, WI, United States, ²Medical Physics, University of Wisconsin-Madison, Madison, WI, United States

Quantitative Susceptibility Mapping (QSM), which has been employed in brain applications to measure iron, also has the potential to quantify liver iron overload. However, unlike in the brain, the presence of abdominal fat provides a unique opportunity to constrain the ill-posed QSM estimation problem. In this work, we constrain the QSM estimation by assuming that the magnetic susceptibility of fat (and air) is constant in the presence of iron overload. Comparison of a conventional QSM constraint (ℓ1-regularization) to the proposed fat/air-constraint shows that the latter is more accurate even in the presence of background field variation and noise.