A pilot validation of compressed sensing four dimensional multi-echo based echo-planar J-resolved spectroscopic imaging in human brain
Manoj Kumar Sarma1, Zohaib Iqbal1, Rajakumar Nagarajan1, and M. Albert Thomas1

1Radiological Sciences, UCLA School of Medicine, Los angeles, Los Angeles, CA, United States


Phase encoded MR spectroscopic imaging (MRSI) is a time-consuming protocol because there is a large number of phase encodings to be collected. Non-uniform undersampling (NUS) and Compressed Sensing (CS), which has been widely used in MRI/MRS can be used to speed up the acquisition further. In this study, we have implemented a novel CS based multi -echo echo planar J-Resolved spectroscopic imaging (ME-EP-JRESI) acquisition and CS reconstruction in in human brain at 3T. We were able to detect and quantify several metabolites using a modified ProFit algorithm. The CS reconstructed spectra were of high quality, with metabolite ratios matching the fully encoded data closely.


Localized J-resolved spectroscopy (JPRESS), which resolve overlapping metabolites better than 1D MRS taking advantage of J-coupling interactions between protons of metabolites and an extra spectral dimension, has previously been shown to be a powerful tool in the study of metabolism in human brain in vivo1, 2. To increase the spatial coverage, J-resolved spectroscopy sequence was modified with an echo-planar spectroscopic imaging (EPSI)3-5 readout to yield a novel four-dimensional (4D) spectroscopic imaging (SI) (2 spectral and 2 spatial) called echo-planar J –resolved spectroscopic imaging (EP-JRESI)6 that allows collection of 2D JPRESS spectra from multiple regions in a single experiment. One of the limitations of EP-JRESI to use in human brain is the long scan time. A significant acceleration was achieved for the EP-JRESI using a multi-echo (ME) based acquisition scheme, called Multi-Echo based Echo-Planar J-resolved Spectroscopic Imaging (ME-EP-JRESI)7, which uses two bipolar echo-planar imaging (EPI) read-out trains to collect dual phase encoded lines within a single TR2. But acquisition duration still remains a limitation for a routine clinical use in brain. Non-uniform undersampling (NUS) and Compressed Sensing (CS)8, which has been widely used in MRI/MRS can be used to speed up ME-EP-JRESI acquisition further. In this study, we have implemented a novel CS based ME-EP-JRESI acquisition and CS reconstruction in human brain at 3T. We further quantify the metabolites from prospectively undersampled brain phantom data using prior knowledge fitting (ProFit) algorithm9 to determine the integrity and reproducibility of CS reconstruction.

Materials and Methods:

The basic 4D ME-EP-JRESI sequence (Fig. 1) which uses a 90°–180°-Δt1-180° scheme for localization was modified by imposing non-uniform sampling (2X) along the t1 direction using a exponential sampling density scheme. To examine the performance and reliability, fully encoded phantom data were collected together with prospectively undersampled phantom scans. The sequence was tested further in the brain of five healthy volunteers (age=23-58 years). All data were collected on a 3T Prisma MRI scanner with a 16-channel head ‘receive’ coils. The following parameters were used for both fully sampled and NUS-based ME-EP-COSI phantom data: TR/TE=2s/30ms, 1x1x2cm3 voxel for VOI localization, 100 Δt1 increments, 256 bipolar echo pair, FOV=16x16cm2, F1 and F2 bandwidths of 1000 Hz and 1190 Hz respectively. A non-water-suppressed ME-EP-JRESI data with t1=1 were also recorded. In-vivo data were collected using the same parameters as the phantom scans except that the TR was decreased to 1.5s and t1 increments to 64 for a scan time ~7min. The undersampled data was reconstructed using a modified Split Bregman algorithm10 which solves the unconstrained optimization problem, $$$\arg min_{u} \|\triangledown\ u\|_1 +\frac{\lambda}{2} \|F_{p}u - d\|_2^2$$$ where u is the reconstructed data, ∇ is the gradient operator, Fp is the undersampled Fourier transform, d is the under-sampled data, λ is a regularization parameter, and ||x||n is the ln norm. Acquired data were post-processed with a custom MATLAB-based library and then reconstructed using CS along the t1 dimension. Modified version of the Profit algorithm was used for quantitation and metabolite ratios were calculated with respect to the 3.0 ppm creatine peak (S/SCr).

Results and Discussion:

Figure 2(a) shows the PRESS VOI localization volume on a T1-weighted axial MRI of a 38-year-old healthy subject brain. Spatial distribution of Cr/Cho map from the same subjects is shown in Figure 2(b) and representative 2D J-resolved spectra extracted from the left hippocampus and thalamus regions following CS reconstruction are shown in Figure 2(c) and 2(d) respectively. No visible aliasing is seen after the reconstruction. All major metabolites: mI, creatine Cr, Cho, Glx, NAA are visible. Figure 3 shows the metabolite map of the 2D diagonal peaks of Cr/Cho, and NAA from fully sampled and CS reconstructed ME-EP-JRESI phantom data. The reconstructed metabolite map of the diagonal peaks exhibited similar spatial profiles as that of the fully sampled for 2x accelerations showing that the CS reconstruction successfully cleans up the incoherent aliasing produced by the NUS. Figure 4 shows a comparison of metabolite ratios with respect to Cr calculated by ProFit for the fully-sampled, and prospectively 2x accelerated phantom datasets for the central 4x4 voxels. The CS reconstructed spectra were of high quality, with metabolite ratios matching the fully encoded data closely. Figure 5 shows extracted spectra from the same location for a fully sampled and CS-reconstructed ME-EP-JRESI phantom data exhibiting similar spectral profiles.


This is an exploratory, validation, and feasibility study for CS-based ME-EP-JRESI for human brain application at 3T. We have shown that NUS/CS can successfully be applied to the ME-EP-JRESI sequence providing an acceleration factor of 2x. All these bring ME-EP-JRESI closer to becoming a clinical reality.


This research was supported by National Institute of Health (NIH) grant 1R21NS08064901A1.


1. Ryner LN, Sorenson JA, Thomas MA.Localized 2D J-resolved 1H MR spectroscopy: strong coupling effects in vitro and in vivo. Magn Reson Imaging. 1995;13(6):853-69.

2. Thomas MA, Hattori N, Umeda M, et al. Evaluation of two-dimensional L-COSY and JPRESS using a 3 T MRI scanner: from phantoms to human brain in vivo.NMR Biomed. 2003;16(5):245-51.

3. Mansfield P. Spatial mapping of the chemical shift in NMR. J. Phys. D. Appl. Phys. 1983; 16:L235?L238.

4. Mansfield P. Spatial mapping of the chemical shift in NMR. Magn. Reson. Med. 1984 1(3):370?86.

5. Posse S, DeCarli C, Le-Bihan D. Three-dimensional echo-planar spectroscopic imaging at short echo times in human brain. Radiology 1994; 192(3):733-38.

6. Nagarajan R, Furuyama J, Margolis D, et al. Echo planar based J resolved and correlated spectroscopic imaging of human prostate using external coil. In: Proceedings of the 19th Annual Meeting of the International Society for Magnetic Resonance in Medicine, May 6–13, 2011; Montreal, Quebec, Canada: 2801.

7. Furuyama JK, Burns BL, Wilson NE, Thomas MA. Multi-Echo-Based Echo-Planar Spectroscopic Imaging Using a 3T MRI Scanner. Materials. 2011;4:1818-34.

8. Lustig M, Donoho D, Pauly JM. Sparse MRI: the application of compressed sensing for rapid MR imaging. Magn Reson Med 2007;58(6):1182–95.

9. Schulte RF, Boesiger P. ProFit: two-dimensional prior-knowledge fitting of J-resolved spectra. NMR Biomed. 2006;19:255-63.

10. Goldstein T, Osher S. The Split Bregman method for L1-regularized problems. Siam Journal on Imaging Sciences 2009;2:323-43.


Figure 1: A Schematic diagram of the 4D ME-EP-JRESI pulse sequence.

Figure 2: (a) T1-weighted coronal MRI of a 38-year-old healthy subject with the white box indicating the PRESS localization; (b) multivoxel spatial map of Cr/Cho at 3.0/ 3.2 ppm; 2D J-resolved spectra extracted from the (c) left hippocampus and (d) thalamus of the same subject.

Figure 3: (a) Spatial map of 2D diagonal peaks of Cr/Cho and NAA from the (a) fully sampled and (b) CS reconstruction of NUS 4D ME-EP-JRESI phantom data.

Figure 4: Table showing comparison of selected ProFit-quantified metabolite ratios of phantom scans tNAA=NAA+NAAG, Glx=Glu+Gln; tCho=Cho+GPC+PCH. NAA=N-acetylaspartate, Cho=free choline, Asp=aspartate, GABA=γ-aminobutyric acid, Glc=Glucose; Glu=glutamate, Gln=glutamine, GSH=glutathione, mI=myo-inositol, PE=phosphoethanolamine, Tau=taurine, GPC=glycerylphosphocholine, PCh=phosphorylcholine, NAAG=Nacetylaspartylglutamate. tNAA=NAA+NAAG, Glx=Glu+Gln; tCho=Cho+GPC+PCH.

Figure 5: Extracted spectra from fully sampled (left) and CS reconstructed (right) phantom ME-EP-JRESI data.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)