Room 250 BCEF

10:30 - 12:30

Moderators:

Justin P. Haldar, Lei Leslie Ying

Thomas Gaass¹, Guillaume Potdevin², Grzegorz Bauman^3,4, Peter Noël⁵, and Axel Haase^1
1Zentralinstitut für Medizintechnik, Technische Universität München, Garching, Germany, ²Department of Physics, Technische Universität München, Garching, Germany, ³Department of Medical Physics, German Cancer Research Center, Heidelberg, Germany, ⁴Department of Medical Physics, University of Wisconsin, Madison, Wisconsin, United States, ⁵Department of Diagnostic and Interventional Radiology, Technische Universität München, Munich, Germany

We introduce a novel reconstruction technique (RHiCA), based on the specific signature of incoherent undersampling artifacts in histogram space and relying on a minimization algorithm. Utilizing the histogram of a low resolution reference image we propose to correct the impaired image via its altered histogram. The performance of the Reference Histogram Constrained Artifact (RHiCA) reduction is successfully presented on a numerical phantom and a 3T head MRI. We successfully illustrate, that applying RHiCA leads to an effective suppression of undersampling artifacts in radial MRI acquisitions up to undersampling factors of 6.

Joshua D. Trzasko¹ and Armando Manduca^1
1Mayo Clinic, Rochester, MN, United States

Previously, low-rank matrix regression methods have been used to enable "calibrationless" parallel and "training-free" dynamic MRI reconstruction. In this work, we present a novel low n-rank tensor regression framework for calibrationless reconstruction of dynamic and multi-channel MRI data, and demonstrate that previously image-domain strategies arise as instances of this unifying model.

Sagar Mandava¹, Jean-Philippe Galons², and Ali Bilgin^1,3
1Electrical and Computer Engineering, University of Arizona, Tucson, AZ, United States, ²Medical Imaging, University of Arizona, Tucson, AZ, United States, ³Biomedical Engineering, University of Arizona, Tucson, AZ, United States

Simultaneous Multislice Acquisition (SIMA) by Hadamard-encoded excitation has been proposed as an alternative to 3D volume imaging (3DFT) when acquiring fewer than 64 slices to avoid ringing and leakage artifacts. As slices are excited simultaneously, SIMA enjoys SNR benefits over slice-by-slice imaging similar to 3DFT. In this work, we investigate the use of compressive sampling strategies within the SIMA framework. In addition to Hadamard and complex Hadamard encoding, we introduce the use of Noiselet encoding in SIMA.

Yihang Zhou^1,2, Yanhua Wang^1,2, Jie Zheng³, and Leslie Ying^1,2
1Department of Electrical Engineering, University at Buffalo, Buffalo, NY, United States, ²Department of Biomedical Engineering, University at Buffalo, Buffalo, NY, United States, ³Department of Radiology, Washington University, St. Louis, MO, United States

In this study, a broader family of nonlinear transforms is investigated for sparse representation of dynamic images in compressed sensing (CS). We propose a novel kernel-based CS method that implicitly and adaptively sparsifies the dynamic image series of interest using nonlinear transforms. The proposed method is evaluated using accelerated arterial spin labeled perfusion data. It is shown to be able to better preserve the spatial and temporal information than the conventional CS method with linear transforms.

Li Feng^1,2, Jing Liu³, Kai Tobias Block⁴, Jian Xu⁵, Leon Axel^1,2, Daniel K. Sodickson^1,2, and Ricardo Otazo^1,2
1Center for Biomedical Imaging, New York University, School of Medicine, New York, New York, United States, ²Sackler Institute of Graduate Biomedical Sciences, New York University, School of Medicine, New York, New York, United States, ³Radiology and Biomedical Imaging, University of California, San Francisco, San Francisco, CA, United States, ⁴Center for Biomedical Imaging, NYU Langone Medical Center, New York, New York, United States, ⁵Siemens Medical Solutions, New York, New York, United States

Respiratory motion reduces the temporal sparsity and thus the performance of compressed sensing reconstruction. In this work, we propose a respiratory motion compensation method for compressed sensing reconstruction using golden-angle radial sampling by creating an extra respiratory-phase dimension estimated from the acquired data with self-gating. The additional respiratory-phase dimension improves the performance of compressed sensing for free-breathing imaging due to (a) additional correlation and thus increased overall multidimensional sparsity and (b) higher incoherence, since the dimension is formed by sorting complementary golden-angle radial data. We demonstrate the feasibility of the technique for accelerated free breathing cardiac cine and liver imaging.

Maojing Fu^1,2, Bo Zhao^1,2, Joseph Holtrop^2,3, Jamie Perry⁴, David Kuehn⁵, Zhi-Pei Liang^1,2, and Bradley P. Sutton^2,3
1Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States, ²Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States, ³Department of Bioengineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States, ⁴Department of Communication Sciences and Disorders, East Carolina University, Greenville, North Carolina, United States, ⁵Department of Speech and Hearing Science, University of Illinois at Urbana-Champaign, Urbana, Illinois, United States

Dynamic MRI can provide quantitative assessment on the anatomy and dynamics of the vocal tract, but usually suffers from poor spatial or temporal resolution or poor spatial coverage. This work presents full-vocal-tract dynamic imaging at a frame rate of 102.2 fps with a 2.2 mm × 2.2 mm × 6.5 mm spatial resolution. It is performed by incorporating an optimized 3D volumetric navigation scheme into a Partial Separability (PS) model-based imaging method. Subtle temporal behaviors of the articulator motion and fine 3D anatomy of the vocal tract are well captured and analyzed.

Manoj K. Sarma¹, M. Albert Thomas¹, Peng Hu², Daniel B. Ennis², and Ke K. Sheng^3
1Radiological Sciences, UCLA School of Medicine, Los Angeles, CA, United States, ²Radiological Sciences, University of California Los Angeles, Los Angeles, CA, United States, ³Radiation Oncology, UCLA School of Medicine, Los Angeles, CA, United States

Radiotherapy guided by MRI has afforded the hardware potential to treat a moving tumor more accurately but existing imaging speed is inadequate for 3D real-time lung and lung tumor imaging. By exploiting the intrinsic coherence of the patient anatomy during time, we adapted a k-t SLR compressed sensing method to dramatically reduce the amount of data that is needed to update a new dynamic imaging without losing details. We were able to accurately track moving tumors of nine patients based on images reconstructed with very high data under-sampling ratio up to 5% of the original data.

Muhammad Usman¹, Mariya Doneva², Tobias Schaeffter¹, and Claudia Prieto^1,3
1King's College London, London, United Kingdom, ²Philips Research Europe, Hamburg, Germany, ³Pontificia Universidad Catolica de Chile, Santiago, Chile

 

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

This work compares several sparsity models for dynamic MRI with the focus on diffusion MRI. The low-rank model, a global sparsification of diffusion images via SVD, generally was found to be the best model, while the rank-sparsity decomposition was shown to be the best when the diffusion images are non-low-rank.

Jiangsheng Yu¹, Yiqun Xue², and Hee Kwon Song^2
1Toshiba Medical Research Institute USA, Cleveland, Ohio, United States, ²University of Pennsylvania, Philadelphia, Pennsylvania, United States

Synopsis: Temporally constrained reconstruction based on compressed sensing (CS) has recently been developed for dynamic MR imaging to obtain high temporal resolution without losing image quality. The intensive computation overhead in CS reconstruction has limited the application for clinical data processing where large data sets are generated from multi-slice and multi-channel acquisition. The current work presents a parallelized GPU implementation to accelerate the CS-based image reconstruction in DCE-MRI. The forward and backward gridding operations, which are the most-time consuming part of the conjugate gradient searching, is addressed with a radial-point driven parallelization approach by assigning a thread for each radial point operation. A comparison with the C++ sequential implementation shows an acceleration factor of ~15 was achieved on a moderately GPU-powered laptop computer.