Comparison of ROI averaging and Spectral Localization by Imaging (SLIM) Using High Resolution 3D Echo Planar Spectroscopic Imaging (EPSI)
Sean Edmund Ellis1,2, Peter Adany1, Phil Lee1,2,3, and In-Young Choi1,2,3,4

1Hoglund Brain Imaging Center, University of Kansas Medical Center, Kansas City, KS, United States, 2Department of Bioengineering, University of Kansas, Lawrence, KS, United States, 3Department of Molecular & Integrative Physiology, University of Kansas Medical Center, Kansas City, KS, United States, 4Department of Neurology, University of Kansas Medical Center, Kansas City, KS, United States


Conventional spectroscopic imaging methods have limitations that make acquiring metabolic information for complexly-shaped brain regions a challenge. The following study compares two methods for acquiring the regional metabolic spectra for a complex compartment shape: spectral estimation via the spatial-averaging of voxels, and Spectral Localization by Imaging (SLIM). Both techniques used the original data sets acquired from 3D Echo Planar Spectroscopic Imaging sequences. The two methods were compared, with the results showing that SLIM could provide comparable compartment spectra using fewer voxel acquisitions without a significant drop in spectral quality.


Echo Planar Spectroscopic Imaging (EPSI) offers high spatial resolution compared to conventional MRSI. However, acquisition times for EPSI can exceed 20 minutes per scan. Currently, quantifying metabolic concentrations in brain regions is difficult due to restricted resolution and the need to average voxels inside regions of interest. Spectral Localization by Imaging (SLIM) offers to improve upon voxel-based post processing by utilizing high resolution magnetic resonance image (MRI) to create compartment maps which are used to estimate the spectra for the region. SLIM utilizes a non-Fourier over-determined linear reconstruction, meaning that SLIM can acquire the same compartmental data using fewer k-space acquisitions to achieve performance comparable with voxel-averaging techniques. The purpose of this study was to evaluate MRSI reconstructions acquired by 3D EPSI, using both full k-space voxel averaging and SLIM with various reduced k-space sizes.

Background and Theory

The theory of SLIM allows for the reconstruction of k-space data into precise binary-valued compartments corresponding to homogeneous anatomical regions [1]. SLIM is expressible by a matrix equation s = G*c, where s is a vector of phase encoded signals, G is a geometry matrix derived from the compartment information and phase encoding vectors, and c is a vector of the compartment coefficients. SLIM requires a greater number of k-space points than the number of compartments. With its flexible k-space sampling theory, SLIM offers to significantly reduce the number of phase encoding acquistions. For example, often-used GRAPPA acquisition in EPSI reduces the number of acquired k-space lines. Because EPSI offers a large k-space coverage, our aim was to evaluate the performance of SLIM with different k-space sizes and to compare these results to the more traditional voxel averaging technique.


Twenty healthy control subjects (35+/-10) were consented according to institutional review board approved protocols. Scans were performed on a 3 T scanner (Skyra, Siemens) with a 16-channel head receiver coil. 1H MRSI was acquired using a 3D EPSI sequence [2] (TE/TR=3980/200000 ms). Parcellation of gray and white matter was obtained from MPRAGE images using FreeSurfer [3] to provide regional anatomical compartments for MIDAS and SLIM. The FreeSurfer parcellation map was modified to generate uniform compartment shapes that would operate well in SLIM and voxel averaging packages. EPSI reconstruction and compartment-based voxel averaging was performed by Metabolite Imaging and Data Analysis Software (MIDAS). SLIM reconstructions were done using in-house programs written in Matlab, and the MIDAS [2] software (MINT) was used for ROI averaging. Spectral fitting was done using LCModel [4] with water-referenced absolute quantification.

Results and Discussion

SLIM showed consistent metabolite results among all k-space tests ranging from a size of 50x50x12 to 16x16x12, showing that a k-space 1/10 the size of the original k-space provided sufficient information to acquire the same results. The tCho and tNAA ratios to creatine were each consistent across the four tested k-space sizes. The creatine ratios of metabolites found by SLIM matched those obtained through MINT, demonstrating that the localization of anatomical compartment spectra using SLIM can be achieved using fewer phase encoding acquisitions. SLIM and related localization techniques may hence allow for shorter 3D EPSI acquisitions using non-Fourier algebraic reconstruction for improved quantitative MRSI in the human brain.


The Hoglund Brain Imaging Center is supported by the NIH (S10RR029577) and the Hoglund Family Foundation.


1. Hu X, et al., SLIM: spectral localization by imaging. Magn Reson Med, 8:314-322, 1988

2. Maudsley A.A., et al., Comprehensive processing, display, and analysis for in vivo MR spectroscopic imaging. NMR Biomed. 19:492-503, 2006

3. Dale, A.M., et al., Cortical surface-based analysis. I. Segmentation and surface reconstruction. Neuroimage 9:179-194, 1999

4. Provencher S.W., et al., Estimation of metabolite concentrations from localized in vivo proton NMR spectra. Magn. Reson. Med. 30:672–679, 1993


Figure 1. Automatic tissue parcellation of the brain obtained using FreeSurfer. FreeSurfer results were regrouped into suitable compartments for SLIM and ROI averaging.

Figure 2. SLIM reconstruction with multiple k-space sizes compared with ROI averaging. Several truncated k-space sizes were used in SLIM. ROI averaging used the full 50x50x18 k-space size.

Table 1. Comparison of SLIM and ROI averaging of voxels. The ratios of SLIM and ROI averaged values of the metabolite creatine ratios were tabulated. A value of 1.0 corresponds to agreement of the two methods.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)