Reconstruction & Processing Methods for Quantitative Susceptibility Mapping

Wednesday 14 May 2014

Tian Liu¹, Dong Zhou², Pascal Spincemaille², and Yi Wang^2
1MedImageMetric LLC, New York, New York, United States, ²Radiology, Weill Cornell Medical College, New York, New York, United States

High quality background field removal is a critical yet challenging step in quantitative susceptibility mapping (QSM), as it involves phase unwrapping, skull stripping, and knowledge of the boundary conditions. In this work, we provide a theoretical analysis of the background field removal and propose to bypass all the steps using a differential equation approach.

Wei Li^1,2, Alexandru V Avram^1,3, Bing Wu^1,4, Xue Xiao^1,5, and Chunlei Liu^1,6
1Brain Imaging and Analysis Center, Duke University, Durham, NC, United States, ²Research Imaging Institute, University of Texas Health Science Center at San Antonio, San Antonio, TX, United States, ³National Institute of Health, Bethesda, DC, United States, ⁴GE Healthcare, Beijing, China,⁵Tsinghua University, Beijing, China, ⁶Radiology, Duke University, Durham, NC, United States

3D phase unwrapping and background phase removal are typically two separated preprocessing steps for quantitative susceptibility mapping (QSM). We developed a novel method for integrated phase unwrapping and harmonic background phase removal using Laplacian, named as HARPERELLA. HARPERELLA is fast, robust, easily to implement and yields local tissue phase with similar accuracy to that using the well-known SHARP and PDF methods. To facilitate evaluation and dissemination, we combined HARPERELLA, QSM, susceptibility tensor imaging (STI) algorithms and related graphical user interfaces into a software package, named as “STI Suite”, which is available online for free academic use.

Ferdinand Schweser¹, Pιnar Senay Özbay^2,3, Andreas Deistung¹, Edsel Daniel Peres Gomez¹, Xiang Feng¹, Daniel Nanz², and Jürgen R Reichenbach^1
1Medical Physics Group, Institute of Diagnostic and Interventional Radiology I, Jena University Hospital - Friedrich Schiller University Jena, Jena, Germany, ²Department of Radiology, University Hospital Zurich, Zurich, Switzerland, ³Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland

Sophisticated harmonic artifact reduction for phase data (SHARP) is regarded as one of the most important techniques in the area of background correction for QSM. The influence of the SHARP parameters has not been investigated in detail so far and they are usually chosen empirically, which may have serious consequences for the comparability of studies relying on SHARP processed phase images. We performed a comprehensive analysis of the impact of different parameter choices on the resulting images and propose a universal definition of the regularization parameter that does not depend on the spherical radius or on the image resolution.

Dong Zhou¹, Tian Liu², Pascal Spincemaille¹, and Yi Wang^1,3
1Radiology Department, Weill Cornell Medical College, New York, NY, United States, ²Medimagemetric LLC, NY, United States, ³Biomedical Engineering Department, Cornell University, Ithaca, NY, United States

We assume simple boundary conditions and remove the background field by explicitly solving the boundary value problems of Laplace’s or Poisson’s equation. The proposed Laplacian boundary value (LBV) method for background field removal retains data near the boundary and is computationally efficient. Tests on a numerical phantom, an experimental phantom and in in vivo data sets showed that LBV was more effective than the SHARP and PDF methods.

Berkin Bilgic¹, Audrey Fan², Cornelius Eichner¹, Stephen Cauley¹, Jonathan Polimeni¹, Marta Bianciardi¹, Elfar Adalsteinsson², Lawrence Wald¹, and Kawin Setsompop^1
1Martinos Center for Biomedical Imaging, Charlestown, MA, United States, ²MIT, Cambridge, MA, United States

A high-resolution whole brain QSM reconstruction can take up to 20 min on a workstation, which poses a limit on QSM usability in clinical and research settings. Herein, we introduce an improved Split-Bregman (SB) L1-regularized dipole inversion algorithm that offers 20× faster reconstruction relative to the standard nonlinear conjugate gradient (NCG) solver. Additionally, we extend SB L1-regularization to admit magnitude-weighting that prevents smoothing across edges identified on the magnitude signal, and solve this more complicated problem 5× faster than the NCG approach. Further, we extend the previously proposed closed-form L2-based inversion to admit magnitude-weighting, and demonstrate 15× acceleration relative to NCG by employing a preconditioner that leads to faster convergence. Utility of the proposed methods is demonstrated in high-resolution (0.6 mm isotropic) 3D GRE data at 3T, as well as multi-echo Simultaneous Multi-Slice (SMS) EPI time-series at 7T.

Diana Khabipova¹, Rolf Gruetter^1,2, and Marques P. José^3
1LIFMET, EPFL, Lausanne, Vaud, Switzerland, ²Radiology, University of Geneva, Geneva, Geneva, Switzerland, ³Radiology, University Lausanne, Lausanne, Vaud, Switzerland

Quantitative susceptibility mapping (QSM) has shown the potential delivering iron deposition estimations in deep grey matter structures. Compared to multiple orientation acquisitions, many proposed regularization methods remain time consuming, rely on careful regularization selection and provide inferior results. A recently proposed closed-form solution including an l2-regularization allows fast reconstructions. Our methodology uses additional k-space position dependent modulation, thus avoiding regularizing not ill-posed regions and being effective without showing signs of over-smoothing QSM maps. Optimal modulation enables largely independence from regularization parameters, . Being comparable for  values an order of magnitude greater than optimum makes it ideal for unsupervised usage.

Fei Cong¹, Bo Wang¹, Xiaohong Joe Zhou², Yan Zhuo¹, and Yongquan Ye^3
1Institute of Biophysics, Chinese Academy of Sciences, Chaoyang District, Beijing, China, ²Department of Radiology and Center for MR Research, University of Illinois Medical Center, Chicago, IL, United States, ³Department of Radiology, School of Medicine, Wayne State University, Detroit, MI, United States

A new edge-preserving method for quantitative susceptibility mapping (QSM) is proposed, which uses 3D slab segmented Wiener filtered phase data as input and the convex half-quadratic regularizing algorithm to suppress the artifacts when solving the inverse problem by using ill-conditioned Green¡¯s function kernel. Compared to current QSM methods, this edge-preserved method reserves the high spatial frequency details for veins in the final susceptibility maps. The comparisons between quadratic regularized and edge-preserved methods of the whole brain are presented, and the feasibility and future work are discussed.

Kristian Bredies¹, Stefan Ropele², Benedikt A Poser³, Markus Barth⁴, and Christian Langkammer^2
1Institute of Mathematics and Scientific Computing, University of Graz, Graz, Austria, ²Department of Neurology, Medical University of Graz, Graz, Austria, ³Faculty of Psychology and Neuroscience, Maastricht University, Maastricht, Netherlands, ⁴Donders Institute for Brain, Cognition and Behaviour, Radboud University Nijmegen, Nijmegen, Netherlands

Combined 3D EPI acquisition combined with single-step TGV reconstruction yielded reliably QSM images of the entire brain with 1mm isotropic resolution in 1 minute acquisition time.

Ferdinand Schweser¹, Andreas Deistung¹, Xiang Feng¹, Edsel Daniel Peres Gomez², and Jürgen R Reichenbach^1
1Medical Physics Group, Institute of Diagnostic and Interventional Radiology I, Jena University Hospital - Friedrich Schiller University Jena, Jena, Germany, ²Medical Physics Group, Institute of Diagnostic and Interventional Radiology I, Jena University Hospital - Friedrich Schiller University Jena, Germany

Quantitative susceptibility mapping (QSM) is a novel post-processing technique for gradient-echo phase data. Schweser et al. have recently presented that inverse filtering with extreme thresholding of the unit dipole response (SDI) yields within seconds susceptibility maps without noise amplification and with a low level of streaking artifacts. Bilgic et al. recently presented a closed-form solution for rapid L2-regularized QSM with a gradient-based penalty (CF-L2) that provides images with even reduced noise level compared to SDI . In this contribution, we show that combining CF-L2 with SDI can considerably reduce reconstruction artifacts.

Samir D. Sharma¹, Diego Hernando¹, Debra E. Horng¹, and Scott B. Reeder^1,2
1Radiology, University of Wisconsin - Madison, Madison, WI, United States, ²Medical Physics, University of Wisconsin - Madison, Madison, Wisconsin, United States

Noninvasive quantification of liver iron concentration is important for the detection and staging of iron overload as well as for effective longitudinal monitoring during treatment. Magnetic susceptibility is a fundamental property of tissue that has a well-defined relationship to iron concentration. Quantitative susceptibility mapping (QSM) methods have been developed in the brain to quantify the concentrations of localized iron deposits. The purpose of this work was to develop a QSM technique for the liver and to demonstrate its feasibility for measuring susceptibility in the liver, which may serve as an imaging biomarker of liver iron overload.