Fast Spectroscopic Imaging Using Online Optimal Sparse k-Space Acquisition and Projections Onto Convex Sets Reconstruction Yun Gao, 1,2 Stephen M. Strakowski, 1–3 Stanley J. Reeves, 4 Hoby P. Hetherington, 5–7 Wen-Jang Chu, 1,2 and Jing-Huei Lee 1,3 * Long acquisition times, low resolution, and voxel contamination are major difficulties in the application of magnetic resonance spectroscopic imaging (MRSI). To overcome these difficulties, an online-optimized acquisition of k-space, termed sequential forward array selection (SFAS), was developed to reduce ac- quisition time without sacrificing spatial resolution. A 2D proton MRSI region of interest (ROI) was defined from a scout image and used to create a region of support (ROS) image. The ROS was then used to optimize and obtain a subset of k-space (i.e., a subset of nonuniform phase encodings) and hence reduce the acquisition time for MRSI. Reconstruction and processing soft- ware was developed in-house to process and reconstruct MRSI using the projections onto convex sets method. Phantom and in vivo studies showed that good-quality MRS images are obtain- able with an approximately 80% reduction of data acquisition time. The reduction of the acquisition time depends on the area ratio of ROS to FOV (i.e., the smaller the ratio, the greater the time reduction). It is also possible to obtain higher-resolution MRS images within a reasonable time using this approach. MRSI with a resolution of 64  64 is possible with the acquisi- tion time of the same as 24  24 using the traditional full k-space method. Magn Reson Med 55:1265–1271, 2006. © 2006 Wiley-Liss, Inc. Key words: fast spectroscopic imaging; sequential forward ar- ray selection; projections onto convex sets reconstruction; par- tial k-space sampling; region of support Magnetic resonance spectroscopy (MRS) was first intro- duced in the early 1970s to measure in vivo tissue metab- olism in intact biological structures (1,2). Since then, MRS has been utilized to measure the metabolic status of almost every organ system in the body, and in particular is an established tool for studying neurochemical and metabolic abnormalities in the human brain. However, because they require relatively long acquisition times and have a low sensitivity (particularly on low-field MRI systems), MRS studies are frequently limited to single-voxel acquisitions, which may not capture information from the most impor- tant pathologic regions. MRS imaging (MRSI), also known as chemical shift imaging (CSI), is a combination of MRS and MR imaging (MRI). It is a completely noninvasive, multivoxel technique that can acquire information that is representative of both anatomy and regional metabolic states. In addition, MRSI often provides higher-spatial- resolution in vivo biochemical information than the sin- gle-voxel approach. MRSI has been used for basic physio- logical research and clinical imaging of metabolites (3). Proton ( 1 H) MRSI studies have identified both focal and global neuronal metabolic changes in a variety of diseases, including brain tumor (4), subacute and acute cerebral infarction (5), multiple sclerosis (6), AIDS dementia (7), Alzheimer’s disease (8), degenerative ataxia (9), epilepsy (10), and psychiatric disorders (11). Most of these diseases present challenges to neuronal viability, which particu- larly relate to a reduction in the N-acetyl-L-aspartic acid (NAA) concentration. Proton MRSI is also capable of re- vealing the accumulation of lipids (12) and lactate (13) in ischemic myocardium. In addition, applications of phos- phorus MRS and MRSI have focused on energy metabo- lism in the human brain, skeletal muscle, and cardiac muscle (10,14). Despite its potential, MRSI has been used largely for preclinical research because it requires a relatively long acquisition time to obtain images with sufficient spatial information. Consequently, it is difficult to use on human subjects. Furthermore, physicians are interested in using multiple approaches in order to make accurate, timely diagnoses. These approaches might require MRSI data in addition to other imaging information, such as that ob- tained by diffusion tensor imaging (DTI), perfusion, and functional MRI (fMRI), in one study session so that more information is available. In order for MRSI to reach its full potential for clinical applications, the acquisition time must be reduced. A few methods have been proposed to reduce MRSI acquisition time (15). These include echo-planar spectro- scopic imaging (EPSI) (16 –19), spiral spectroscopic imag- ing (20,21), parallel spectroscopic imaging (22,23), chem- ical shift encoding (24,25), and partial k-space sampling (26 –35). EPSI and spiral spectroscopic imaging use a sin- gle-shot technique in which the polarity of the gradients is rapidly switched during data acquisition. Although this 1 Center for Imaging Research, University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. 2 Department of Psychiatry, University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. 3 Department of Biochemical Engineering, University of Cincinnati College of Medicine, Cincinnati, Ohio, USA. 4 Department of Electrical Engineering, Auburn University, Auburn, Alabama, USA. 5 Department of Radiology, Albert Einstein College of Medicine, Bronx, New York, USA. 6 Department of Physiology, Albert Einstein College of Medicine, Bronx, New York, USA. 7 Department of Biophysics, Albert Einstein College of Medicine, Bronx, New York, USA. *Correspondence to: Jing-Huei Lee, Center for Imaging Research, University of Cincinnati, 231 Albert Sabin Way, Room 6153A, Medical Science Building, Cincinnati, OH 45267-0586. E-mail: jing-huei.lee@uc.edu Received 1 August 2005; revised 16 February 2006; accepted 21 February 2006. DOI 10.1002/mrm.20905 Published online 5 May 2006 in Wiley InterScience (www.interscience.wiley. com). Magnetic Resonance in Medicine 55:1265–1271 (2006) © 2006 Wiley-Liss, Inc. 1265