Li Feng¹, Hersh Chandarana¹, Davide Piccini^2,3, Justin Ream¹, Daniel K Sodickson¹, and Ricardo Otazo^1
1Center for Advanced Imaging Innovation and Research (CAI2R), Department of Radiology, New York University School of Medicine, New York, NY, United States, ²Advanced Clinical Imaging Technology, Siemens Healthcare IM BM PI, Lausanne, Switzerland, ³Department of Radiology, University Hospital (CHUV) and University of Lausanne (UNIL) / Center for Biomedical Imaging (CIBM), Lausanne, Switzerland

This work proposes a novel framework for free-breathing 3D golden-angle radial dynamic contrast-enhanced MRI that employs respiratory motion sorting instead of explicit motion correction. The continuously acquired k-space data are sorted into different contrast-enhancement phases at multiple respiratory states using the self-navigation properties of radial imaging. The undersampled five-dimensional dataset (x-y-z-contrast-respiration) is reconstructed using a multidimensional compressed sensing approach that exploits sparsity along both contrast-enhancement and respiratory motion dimensions. The performance of the proposed approach is demonstrated for abdominal imaging using two types of 3D golden-angle radial sampling schemes that are based on stack-of-stars and spiral phyllotaxis patterns.

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

Dynamic MRI can provide quantitative assessment on the anatomy and dynamics of the articulators in real time, but usually suffers from limited spa-tiotemporal resolution and poor spatial coverage. This work presents full-vocal-tract 3D dynamic MRI at a frame rate of 150 fps with a 2.0 mm × 2.0 mm × 5.0 mm spatial resolution. It is performed by incorporating an accelerated 3D acquisition 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.

Matthias Schloegl¹, Martin Holler², Kristian Bredies², Karl Kunisch², and Rudolf Stollberger^1
1Institute of Medical Engineering, Graz University of Technology, Graz, Styria, Austria, ²Department of Mathematics and Scientific Computing, University of Graz, Graz, Styria, Austria

In this work we address the problem of undersampled dynamic MR image reconstruction from the general point-of-view of appropriate regularization for image sequences, based on the total generalized variation (TGV) functional. The extension to the dynamic scenario is achieved by infimal convolution of two suitable weighted spatio-temporal TGV functionals that automatically balance the regularity between time and space in an optimal way. This poses a very general yet computational tractable and well-studied motion model for a wide range of dynamic MR applications.

Xin Miao¹, Sajan Goud Lingala², Yi Guo², Terrence Jao¹, and Krishna S. Nayak^1,2
1Biomedical Engineering, University of Southern California, Los Angeles, CA, United States, ²Electrical Engineering, University of Southern California, Los Angeles, CA, United States

It is well known that dynamic MRI performance can be improved by employing constrained reconstruction that leverages the low rank and transform sparse properties of the dynamic image matrix. In this study, we investigate the combination of two powerful temporal constraints, locally low rank (LLR) and temporal total variation (tTV), for accelerating cardiac cine imaging. We show that this com-bination provides better reconstruction accuracy in highly accelerated cases with random or Cartesian golden-angle radial sampling patterns, compared to current state-of-art constrained reconstruction methods such as k-t SLR.

Javier Royuela-del-Val¹, Muhammad Usman², Lucilio Cordero-Grande², Federico Simmross-Wattenberg¹, Marcos Martín-Fernández¹, Claudia Prieto², and Carlos Alberola-López^1
1Laboratorio de Procesado de Imagen, Universidad de Valladolid, Valladolid, Valladolid, Spain, ²Division of Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom

Multislice 2D (M2D) CINE MRI is a clinical gold standard for the assessment of ventricular volumes and cardiac function. However, this acquisition currently needs to be performed during several breath-holds, leading to slice-misalignment and long scans duration. In this work we propose a novel undersampled reconstruction approach to perform M2D whole heart CINE MRI in a single breath hold, where each slice is acquired during a single cardiac cycle. The proposed method, which we call kt-WiSE, is based on compressed sensing (CS) and a groupwise temporal registration algorithm for the estimation and compensation of the motion of the heart.

Yi Guo¹, Yinghua Zhu¹, Sajan Goud Lingala¹, R. Marc Lebel², and Krishna S. Nayak^1
1Department of Electrical Engineering, University of Southern California, Los Angeles, CA, United States, ²GE Healthcare, Calgary, Alberta, Canada

In Dynamic Contrast Enhanced (DCE) MRI, pharmaco-kinetic (PK) maps are derived from the dynamic image series and are used for diagnostic purposes. Direct estimation of PK parameter maps could enable high acceleration rate and save resources required to estimate intermediate images. Here we present a framework to directly estimate PK parameters using a forward model and sparsity constraint, and evaluate this method at very high acceleration rates up to 100x, to demonstrate feasibility.

Joshua Trzasko¹, Eric Borisch¹, Akira Kawashima¹, Adam Froemming¹, Roger Grimm¹, Armando Manduca¹, Phillip Young¹, and Stephen Riederer^1
1Mayo Clinic, Rochester, MN, United States

Dynamic 3D contrast-enhanced MRI (DCE-MRI) is increasingly used clinically for prostate cancer lesion detection, staging, treatment planning/monitoring, and recurrence detection. However, achieving high spatiotemporal resolution and SNR in this application is challenging given the target signal’s transiency and gland’s medial location. Sparsity-driven image reconstruction is an increasingly popular tool that mitigate the tradeoff between resolution and SNR (relative to conventional methods). In this work, we present an alternating direction method-of-multipliers (ADMM) optimization strategy specifically for our Cartesian acquisition protocol that enables <5 minute 4D DCE-MRI sparse reconstructions. After overviewing the mechanics of this algorithm, we show that its results were consistently preferred for diagnosis over the clinical standard (SENSE) by radiologists in 19 suspected prostate cancer patient studies.

Frank Ong¹, Tao Zhang², Joseph Cheng², Martin Uecker³, and Michael Lustig^3
1Electrical Engineering and Computer Sciences, University of California, Berkeley, Berkeley, California, United States, ²Stanford University, California, United States, ³University of California, Berkeley, California, United States

A multi-scale low rank reconstruction method is presented to exploit spatio-temporal correlations of dynamic contrast enhanced (DCE) images across multiple scales. The proposed method separates different scales of contrast dynamics with different sizes of low rank matrices and provides a more compact representation of DCE images than conventional low rank methods. Results from multi-scale low rank reconstruction are compared to locally low rank and low rank plus sparse modeling.

Suhyung Park¹ and Jaeseok Park^2
1Center for Neuroscience Imaging Research, Institute for Basic Science (IBS), Sungkyunkwan University, Suwon, Gyeong Gi-Do, Korea, ²Biomedical Imaging and Engineering Lab., Department of Global Biomedical Engineering, Sungkyunkwan University, Suwon, Gyeong Gi-Do, Korea

Dynamic parallel magnetic resonance imaging (PMRI) has been widely used in a variety of fast imaging applications to accelerate the data acquisition without any compromise of the spatial-temporal resolution. An accurate calibration is the key for successful dynamic PMRI. However, the calibration quality typically decreases with both small amount of calibrating signals and motion-induced temporally varying coil sensitivity. In this work, we propose a new, dynamic PMRI exploiting sparse Kalman smoother (k-t SPARKS) for robust calibration and reconstruction in the presence of time-varying coil sensitivity, in which the proposed method incorporates the Kalman smoother calibration and the sparse signal recovery into a single optimization problem, leading to joint estimation of time-varying convolution kernel and full k-space. Simulation and experiments were performed using both the proposed and conventional methods in the free-breathing cardiac cine applications for comparison.

Ukash Nakarmi¹, Yanhua Wang¹, Jingyuan Lyu¹, Jie Zheng², and Leslie Ying^1,3
1Dept. of Electrical Engineering, State University of New York at Buffalo, Buffalo, NY, United States, ²Dept. of Radiology, Washington University, School of Medicine, MO, United States, ³Dept. of Biomedical Engineering, State University of New York at Buffalo, NY, United States

In this abstract, we introduce a nonlinear polynomial-kernel-based model to represent the dynamic MR images sparsely. Based on the model, a novel compressed-sensing dMRI method with self-learned nonlinear dictionary (NL-D) is proposed. Simulation results show that the proposed method outperforms the conventional CS dMRI methods with linear transforms.