Reproducibility of Localized 2D Correlated MRSpectroscopy

Nader Binesh,1 Kenneth Yue,1 Lynn Fairbanks,2 and M. Albert Thomas1,2*

The test-retest reliability of two-dimensional (2D) correlatedspectroscopy (COSY) was studied on a whole-body 1.5T MRIscanner. Single-voxel localized 2D proton spectra were re-corded in vitro as well as in vivo using a recently implementedlocalized chemical shift correlated spectroscopic (L-COSY) se-quence. A total of 40 in vitro and 40 human brain (10 volunteers,four times each) 2D L-COSY spectra were recorded. The coef-ficients of variation (CVs) of selected brain metabolites (rawvolume integrals) recorded in 10 healthy volunteers were lessthan 9% for creatine, choline, and N-acetyl aspartate, and lessthan 17% for myo-inositol, glutamine/glutamate, aspartate, andthreonine/lactate. The 2D metabolite ratios and the raw volumeintegrals of 2D diagonal and cross peaks in healthy human brainwere very well reproduced. The intraclass correlation coeffi-cients were greater than 0.4 (P < 0.05) for the major meta-bolites, indicating that the 2D peak volumes were stableenough within individuals to detect reliable differences be-tween normal subjects. Magn Reson Med 48:942–948, 2002.© 2002 Wiley-Liss, Inc.

Key words: reproducibility; 2D spectroscopy; L-COSY; brainmetabolites; coefficient of variation

In the past decade, MR spectroscopy (MRS) has become areliable clinical tool for the diagnosis of selected diseases(1–7). An important feature of any diagnostic tool is itsreproducibility. Many studies on the reproducibility ofone-dimensional (1D) 1H MRS have reported the ratios andabsolute concentrations of the metabolites (8–13). Al-though these studies have shown small errors (3–6%) inthe measurement of metabolites in vitro (8,9), the in vivostudies showed higher coefficients of variation (7–26%) ina maximum of only five metabolites (9–13). Severe overlapof brain metabolites is a major hindrance in 1D MRS.

Localized and nonlocalized versions of 2D MRS werereported a decade ago (14–18); however, only recentlyhave 2D proton spectra been recorded in healthy humansubjects and patients with brain tumors (19–25). Com-pared to 1D MRS, 2D MR spectra allow less ambiguousassignment of several metabolite resonances in human

brain and prostate (25–27). In a previous study (25), theresonances caused by glutamine/glutamate (Glx) wereclearly separated from the dominant singlet resulting fromN-acetyl aspartate (NAA) and myo-inositol (mI) in thehuman brain. Even though identification of “free” aspar-tate (Asp) was not possible with 1D MRS, the peaks wereseparated from NAA in 2D MRS (25). 2D cross peakscaused by phosphoethanolamine and ethanolamine (PE),phosphoryl choline (PCh), threonine and lactate (Thr/Lac),and �-aminobutyric acid (GABA) were also identified inthe 2D L-COSY spectra of the brain. However, to date therehas been no report on the reproducibility of 2D MRS. Thegoal of this work was to investigate the reproducibility of2D peak volumes and 2D metabolite ratios recorded inphantom solutions and healthy brain using the L-COSYsequence (25).

MATERIALS AND METHODS

A 1.5 T GE Horizon (5.8) MRI/MRS scanner (GE MedicalSystems, Waukesha, WI) with echo-speed gradients wasused with a body coil for transmission and a 3-inch surfacecoil for reception. The choice of this setup was necessi-tated by the higher signal-to-noise ratio (SNR) offered bythe 3-inch surface coil as compared to that of a head coil.The 2D L-COSY sequence consisted of three slice-selectiveRF pulses (90°-180°-90°) for the volume localization aswell as for the coherence transfer necessary for 2D spec-troscopy (25). A chemical shift selected (CHESS) sequencewas used for water suppression. 2D L-COSY spectra wererecorded using the following parameters: TE � 30 ms,TR � 2 s, total number of scans � 1024 (128 �t1 incre-ments and 8 NEX/�t1). The total duration for each 2D scanwith water suppression was approximately 34 min. The2D raw matrix consisted of 1024 complex points along thefirst dimension and 128 points along the second dimen-sion. Unsuppressed 2D L-COSY (water signal) with twoaverages was also acquired from the same location as theone used for suppressed L-COSY. The total duration forthe unsuppressed 2D L-COSY was 4 min with 64 t1 incre-ments.

In Vitro

The in vitro measurements were carried out using a GEMRS phantom placed directly on top of the surface coil.The GE MRS phantom consisted of N-acetyl-L-asparticacid (NAA, 12.5 mM); creatine hydrate (Cr, 10 mM); cho-line chloride (Ch, 3 mM); myo-inositol (mI, 7.5 mM); L-glutamic acid (Glu, 12.5 mM); DL-lactic acid (Lac, 5 mM);sodium azide (0.1%); potassium phosphate monobasic

1Department of Radiological Sciences, University of California–Los Angeles,Los Angeles, California.2Department of Psychiatry, University of California–Los Angeles, Los Angeles,California.Grant sponsor: National Institute of Mental Health; Grant number: MH-58284;Grant sponsor: Stanley Vada Foundation.*Correspondence to: M. Albert Thomas, Department of Radiological Sci-ences, UCLA School of Medicine, CHS BL 428, 10833 Le Conte Ave., LosAngeles, CA 90095-1721. E-mail: athomas@mednet.ucla.eduReceived 19 February 2002; revised 12 August 2002; accepted 13 August2002.DOI 10.1002/mrm.10307Published online in Wiley InterScience (www.interscience.wiley.com).

942© 2002 Wiley-Liss, Inc.

Magnetic Resonance in Medicine 48:942–948 (2002)FULL PAPERS

(KH2PO4, 50mM); and sodium hydroxide (NaOH, 56 mM).The phantom also contained 1 ml/l Gd-DPTA (Magnevist).A 3 � 3 � 3 cm3 voxel was placed in the lower half of thespherical phantom, with the center of the voxel approxi-mately an inch above the plane of the surface coil. A totalof 40 water-suppressed 2D COSY spectra were collectedon 20 days over a period of 8 months, with two consecu-tive measurements on each day.

In Vivo

The in vivo study involved 10 healthy volunteers (eightfemales and two males, 20–29 years old, mean age24.7 years) who were scanned four times in 1 month (ap-proximately one scan per week). The protocol was ap-proved by the institutional review board (IRB), and beforeeach study the subjects gave informed consent. Duringeach volunteer’s first scan, a 3 � 3 � 3 cm3 voxel wasplaced on the anterior cingulate gyrus using the anatomi-cal landmarks. The axial slice showing the most anteriorextent of the anterior margin of the genu of the corpuscallosum was chosen as the reference image on which theMRS voxel was localized. This slice was found by placinga cursor at the anterior margin of the genu in the midline,on several axial slices through the genu. This anatomicallandmark was chosen because it is an identifiable singlelocation that could allow better reproducibility betweenscans. For the subsequent scans, the voxel placement fromthe first scan, in addition to the anatomical landmarks, wasused to ensure correct placement of the MRS voxel. How-ever, no special device or headset was used in this study to

reproduce the voxel placement (8). The 3-inch surface coilwas placed directly on the forehead of the subject (whilekeeping the forehead covered with a thin towel). To min-imize operator errors, coil placement, signal acquisition,and postprocessing were carried out by the same person.

Postprocessing

Before double Fourier transformation, the raw data weretransferred to an SGI O2 workstation (Silicon Graphics Inc,San Jose, CA), zero-filled to 2048 and 256 complex pointsalong the two dimensions, and apodized with skewed-squared sine-bell filters. Both suppressed and unsup-pressed 2D L-COSY spectra were processed using identicalparameters. All of the 2D spectra were processed using aFelix-2000 package (Accelrys, San Diego, CA), and recon-structed in the magnitude mode. The peaks were identifiedusing data from the literature and our own phantom ex-periments. The size of each rectangle marking the areaused for the volume integration of a peak was determinedsuch that it just covered the peak (above the noise) and notthe neighboring peaks. After evaluation in a few spectra,optimized sizes of the rectangles (areas of integration)were fixed and applied to all the spectra.

Water Scaling

The unsuppressed 2D L-COSY spectra were used to scalethe metabolite volumes with respect to water content inthe same voxel. The volume integrals of these water peakswere averaged and a mean value was obtained. For each

FIG. 1. 2D L-COSY spectrum of a 3 � 3 �3 cm3 voxel placed in the GE brain phan-tom, with 128 increments (rows) and 8 NEX(averages per row). The peaks used in thestudy are shown with rectangles drawnaround them. The rectangular area denotesthe area of volume integration.

Reproducibility of Localized 2D COSY 943

scan, this number was divided by the respective waterpeak integral to obtain a scaling factor. The scaled value ofeach metabolite was calculated after multiplying the scal-ing factor by the integral value of each metabolite.

Statistical Analysis

The basic statistical analysis was done using the softwareOrigin 6.0. The intrasubject coefficient of variation (CV) foreach metabolite of a subject was obtained from the fourscans of the subject, and then the average of the CVs amongthe 10 subjects was determined. For the intersubject CVs,the average volume of each metabolite peak (average offour scans) was calculated first, and the CVs (for eachmetabolite) were then calculated from these averages. Theintersubject CVs were also calculated based on the firstday’s measurements, which were recorded without aver-aging; hence the errors due to intrasubject variability werealso included. The variance components were calculatedaccording to ANOVA (type I sum of squares) using SPSS10.0 variance component analysis. The intraclass correla-tion coefficient was computed as the ratio of between-subject variance to the sum of within- and between-subjectvariances.

RESULTS

In Vitro Study

Figure 1 shows a typical 2D L-COSY spectrum of a GEMRS phantom. The 2D cross peaks resulting from NAA,Glu, and Lac were clearly separated even though the t1-ridges of water were close to the cross peak of NAA belowthe diagonal. The 2D cross peaks of NAA were asymmetricdue to the effect of water suppression (25). There were twogroups of cross peaks due to mI: one group at [3.5 ppm,3.1 ppm] and a second group at [4.0 ppm, 3.5 ppm] over-lapping with the cross peaks of choline methylene protons(mICh). Nine peaks were chosen for the study: three diag-onal peaks (Cr_d (F2 � F1 � 3.0 ppm), Ch_d (F2 � F1 �3.15 ppm), and NAA_d (F2 � F1 � 2.0 ppm)), and six crosspeaks (Lac (4.1 ppm, 1.3 ppm), Glu (3.6 ppm, 2.0 ppm),NAA (4.4 ppm, 2.6 ppm), mI (3.5 ppm, 3.1 ppm), mICh(4.0 ppm, 3.5 ppm), and Cr (3.9 ppm, 3.0 ppm)) below thediagonal. The rectangles show the areas used for volumeintegration. The intraday CVs of the volume integrals of 2Dpeaks were calculated from the two scans on the same day,averaged over 20 days. These CVs are plotted in Fig. 2a.The average value of the volume integral of a peak on eachday was used to calculate the variation between days (in-terday) for 20 days, distributed over a period of 8 months.These CVs are shown in Fig. 2b. The ratios of these me-tabolites with respect to the diagonal peak of creatine(Cr_d) were also calculated. The CVs of the ratios of thediagonal peaks of NAA and choline with respect to Cr_dwere 1.6% and 2.2%, respectively. The CVs of the crosspeak ratios were: Glu/Cr_d � 8.8%, Lac/Cr_d � 5.6%,mI/Cr_d � 15.5%, mICh/Cr_d � 7.5%, NAA/Cr_d � 8.2%,and Cr/Cr_d � 9.5%.

2D L-COSY of Healthy Volunteers

A typical 2D L-COSY spectrum of a healthy human brain isshown in Fig. 3. There were many diagonal and cross

peaks in the spectrum. As the diagonal peaks suffer fromthe same overlap as the 1D spectra, we picked only threepeaks: Cr_d, Ch_d, and NAA_d. These were singlets due tothe respective methyl groups of Cr and NAA, and thetrimethyl groups of choline. The cross peaks were wellseparated, demonstrating the advantage of using 2D spec-tra. Among the 2D cross peaks, we chose 11 peaks (NAA,mI, mICh, Glx, Asp, Cr, PE, Thr/Lac, GABA, and PCh) thatwere identified below the diagonal peaks. The reportedconcentrations of NAA, Cr, mI, and Glx in human brainusing 1D MRS are �5 mM (9,28). The concentration of PEin rat brain was reported to be around 4 mM (28). Lowerconcentrations have been reported for Asp (1–1.4 mM),PCh (0.6 mM), Thr (0.3 mM), Lac (0.4 mM), and GABA(1.3–1.9 mM) (28). Figure 3 shows these peaks with rect-angles drawn around them, denoting the areas used for thevolume integration.

The intrasubject (within a subject) CVs for the in vivometabolites are shown in Fig. 4a. The CVs of the metabo-lites were calculated in each subject, and the average CVsacross 10 subjects were plotted. The intersubject (betweensubjects) variations shown in Fig. 4b were calculated fromthe mean values of the four measurements in each subject.The values reported for the metabolite peak volumes arethe raw integration numbers; however, they are the source

FIG. 2. CVs of the raw volume integrals of various metabolite peaksin 2D L-COSY spectra of a GE MRS phantom: (a) intraday, thevariation of two scans per day averaged over 20 days; (b) interday,the variation of the mean of the two scans, for 20 days.

944 Binesh et al.

values to be used in the calculation of absolute concentra-tion. Their relationship to absolute concentration is cur-rently being investigated. We also calculated the metabo-lite ratios with respect to the diagonal peak of creatine.These metabolite ratios for both inter- and intrasubjectvariations are tabulated in Table 1. There are two columnsof intersubject CVs: one was obtained from the averagevalue of the four measurements on a subject, and the otherwas calculated from the first-day measurements of thesubjects.

Figure 5 shows the CVs of the water-scaled metabolitevolumes: the in vitro interday variation (a) and the in vivointersubject variation (b). Here the raw volume integrationvalues of the peaks were used in order to calculate thewater-scaled values. This scaling is similar to taking theratio with respect to water, with the added advantage thatone does not need to deal with very small ratios. There-fore, water scaling helped to minimize the effect of thehardware fluctuations, in the same way that the ratios withrespect to the creatine diagonal peak did (Table 1).

In order to assess within-individual consistency, theintraclass correlation was also calculated. Table 2 summa-rizes these results with the respective P-values. An intra-class correlation of more than 0.3 (which is associatedwith a P-value of �0.05) indicates that the measurementsof the metabolites are significantly related to other mea-surements on the same subject, and differ from the valuesobtained from other subjects. Table 2 indicates that theratios of all the major metabolites show an intraclass cor-relation of �0.3 or more (P-values � 0.05). However, the

lower concentration metabolite ratios with respect to thecreatine diagonal peak had lower correlation coefficients,showing that the subjects cannot be distinguished by look-ing at those measurements.

DISCUSSION

In this study, we investigated the reproducibility of 2DL-COSY peaks in a brain phantom (in vitro) as well as inthe human brain (in vivo). A 2D MRS spectrum has betterresolution than the conventional 1D MRS, mainly becauseof the added second spectral dimension. It contains wellresolved cross peaks that can be used to measure themetabolic changes in different pathologies. The simulatedand experimental 2D L-COSY spectra of selected metabo-lites have shown that the 2D spectra are asymmetric due tothe influence of CHESS suppression (29), hence we did notuse the 2D cross peaks above the diagonal.

The in vitro intraday plot (Fig. 2a) indicates CVs � 2%for diagonal peaks and � 6% for cross peaks. The in vitro2D spectra were repeated for a second time on the sameday without moving the phantom; hence the errors couldhave been caused by 1) scanner temporal instability (shimand water suppression), or 2) data processing and mea-surement of volume integrals. Since the area of the volumeintegration and the processing parameters were fixed, thelatter error source was minimized, leaving the scannertemporal instability as the main source of error. On theother hand, the interday measurements (Fig. 2b) showedCVs of 7–15% (3–15% for the water-scaled peak volumes).

FIG. 3. 2D L-COSY spectrum of a 3 � 3 �3 cm3 voxel placed in the frontal medial graymatter of a 25-year-old volunteer, with128 increments (rows) and 8 NEX (averagesper row). The peaks used in the study areshown with rectangles drawn around them.The rectangular area denotes the area ofvolume integration.

Reproducibility of Localized 2D COSY 945

The increase in the CVs may be caused by placement of thevoxel, changes in receiver gain, or day-to-day fluctuationsin RF power and magnetic field (B0) inhomogeneity. Thehigher CV of mI may have been a result of its proximity tothe diagonal.

In addition to the above-mentioned sources of error, thefollowing factors also may have affected the in vivo repro-ducibility study: coil positioning, subject movement, andvariation of brain metabolites of each subject during the 4weekly scans (1 month). The total duration of the entire invivo study was only 4 months (half the time of the in vitro

FIG. 4. The CVs of the raw volume integrals of various metabolitepeaks in 2D L-COSY spectra: (a) intrasubject, the average CVsamong 10 subjects, each CV calculated from the four scans of asubject; (b) intersubject, the CVs calculated from the average vol-ume integral of each peak (from the four scans of a subject).

Table 1The Coefficients of Variation (SD/Mean in Percentage) of theMetabolite Ratios With Respect to the Diagonal Peak of Creatine,in Human Brain*

Metabolitesratio

IntrasubjectIntersubject,

averageIntersubject,

first-day

NAA_d/Cr_d 3.8 5.2 8.1Ch_d/Cr_d 3.1 4.6 6.1NAA/Cr_d 4.4 5.6 8.2Glx/Cr_d 5.5 6.8 9.1mICh/Cr_d 5.3 10.0 12.8mI/Cr_d 9.8 10.2 15.0Asp/Cr_d 11.4 8.4 15.7Cr/Cr_d 15.4 11.1 18.5PE/Cr_d 16.6 14.3 21.3(Thr/Lac)/Cr_d 15.5 9.4 11.4GABA/Cr_d 21.8 17.3 22.7PCh/Cr_d 26.5 14.8 26.1

*Ten volunteers, four scans each.

FIG. 5. The CVs of the raw volume integrals of various metabolitepeaks, scaled with respect to the water signal: (a) in vitro interdaystudy; (b) in vivo intersubject study.

Table 2The Intra-Class Correlation and P-values for the MeasuredMetabolite Volumes Ratios With Respect to the Diagonal Peak ofCreatine, in Human Brain*

Metabolitesratio

Intraclass correlation P-value

NAA_d/Cr_d 0.566 �.001Ch_d/Cr_d 0.603 �.001NAA/Cr_d 0.519 �.001Glx/Cr_d 0.501 �.001mICh/Cr_d 0.737 �.001mI/Cr_d 0.424 �.01Asp/Cr_d 0.170 not significantCr/Cr_d 0.144 not significantPE/Cr_d 0.299 �.05(Thr/Lac)/Cr_d 0.079 not significantGABA/Cr_d 0.186 not significantPCh/Cr_d 0.012 not significant

*Ten subjects.

946 Binesh et al.

study) and we had four averages for a total of 10 subjects,as compared to two intraday averages for a total of 20 scansof a brain phantom. Hence the fluctuations in RF powerand magnetic field (B0) inhomogeneity were expected to besmaller, i.e., better averaged than the in vitro study.

Because time is limited for in vivo scans, sensitivitybecomes an important issue. It is worth mentioning thatthe L-COSY sequence has a theoretical 50% efficiencycompared to the original point-resolved spectroscopy(PRESS) sequence (25). To improve the SNR, we used a3-inch surface coil and a 27-ml voxel for the 2D studies, asopposed to the more conventional head coil and8 ml-voxel in 1D MRS. When compared to the head coil,the 3-inch coil improves the SNR by approximately two-fold, but limits the parts of the brain that can be studied toeither the frontal or the occipital lobes. The use of a 27-mlvoxel has the disadvantages of not being very target-spe-cific and making the partial volume effect large. Ernst andcoworkers (30) showed that similar sensitivity can beachieved in 2D and 1D MRS experiments provided that: 1)T2 relaxation and inhomogeneous decay during the evolu-tion time (t1) are neglected; 2) the MRI scanner is stable, soany source of t1 noise can be neglected during t1; 3) thetotal number of averages are the same for both; and 4) thenumber of peaks are identical in the two experiments.Conditions 1 and 3 were not fully satisfied in our experi-mental setup, as the T2 of the metabolites is comparablewith the evolution time t1, and most of the 1D MRS studiesused a total NEX of 128 as opposed to the 1024 (128 � 8)used in the current work.

In light of the differences between the parameters usedin 2D L-COSY and the conventional 1D acquisition param-eters, caution should be used in comparing the two studiesof MR spectral reproducibility. However, the work pre-sented here summarizes the prospective reliability of the2D MRS technique. Only the metabolite ratios with respectto creatine (Table 1) are compared between the currentwork and the previously reported 1D MRS findings.Schirmer et al. (9) and Marshall et al. (11) both reported anintersubject CV of 8.5% for NAA/Cr as opposed to 8.1% in2D MRS. The CVs of Ch/Cr and Glx/Cr were 16.6% and23.5%, respectively, as reported by Schirmer et al. (9). Incontrast, the current study showed CVs of 6.1% and 9.1%for Ch/Cr and Glx/Cr, respectively. Here, we used the firstday’s values for intersubject CVs; if the 4-day averageswere used, the CVs would be even smaller, as shown in thesecond column of Table 1. Although the low-concentra-tion metabolites were not reported in the 1D reproducibil-ity studies of in vivo spectroscopy in humans, it is evidentthat the 2D L-COSY spectra would be far more reliable inresolving the peaks. The CVs of low-concentration metab-olite ratios were lower than 20%, with the exception ofGABA and PCh (Table 1). One possible reason for thehigher CVs of GABA could be the location of its 2D crosspeak on the t1 and t2 ridges of Cr and NAA. Future effortsshould focus on the optimization of selected metabolitecross peaks by careful manipulation of t1 (31). On the otherhand, the higher CV of PCh may be due to its short T2

value.To further test the reproducibility of the in vivo volume

integrals and metabolite ratios, the intraclass correlationwas calculated and tabulated in Table 2. It is interesting to

note that the major metabolites show significant intraclasscorrelations but the lower concentration metabolites donot. This is in agreement with the intrasubject vs. inter-subject CVs. As the peaks due to lower concentrationmetabolites are weaker (lower SNR), the difference fromone subject to another is not significant when compared toany two scans of a subject. Hence, one cannot distinguishone subject from another by considering the ratios of thelow-concentration metabolites, but one can differentiatebetween subjects by examining the ratios of the majormetabolites.

CONCLUSIONS

We have shown that the 2D L-COSY spectra are veryreliable, and that the metabolite raw volume integrals andthe ratios measured in the human brain are reproduciblewith CVs � 13% for the major metabolites (NAA, Cr, Ch,Glx, and mI). The low-concentration metabolites (such asAsp, PE, Thr/Lac, GABA, and PCh) that were not previ-ously reported in 1D reproducibility studies in humanbrain were reproduced with CVs � 26%. The CVs werefurther reduced by a proper water scaling. Free aspartatewas reproduced with a CV of 12% (intrasubject), indicat-ing the power of an added second dimension to study themetabolites at low physiological concentrations. Better RFcoil design is expected to increase the SNR of the 2D crosspeaks, which would further decrease the CVs.

ACKNOWLEDGMENTS

The authors thank Dr. Amir Huda, Dr. Nathaniel Wyckoff,Prof. Anil Kumar, Dr. Pablo Davanzo, and Dr. Anand Ku-mar for stimulating discussions.

REFERENCES

1. Burtscher IM, Holtas S. Proton MR spectroscopy in clinical routine.Magn Reson Med 2001;13:560–567.

2. Leuzzi V, Bianchi MC, Tosetti M, Carducci CL, Carducci CA, AntonozziI. Clinical significance of brain phenylalanine concentration assessedby in vivo proton magnetic resonance spectroscopy in phenylketonuria.J Inherit Metab Dis 2000;23:563–570.

3. Neubauer S. Cardiac magnetic resonance spectroscopy: potential clin-ical applications. Herz 2000;25:452–460.

4. Lin A, Bluml S, Mamelak AN. Efficacy of proton magnetic resonancespectroscopy in clinical decision making for patients with suspectedmalignant brain tumors. J Neurooncol 1999;45:69–81.

5. Shevell MI, Ashwal S, Novotny E. Proton magnetic resonancespectroscopy: clinical applications in children with nervous systemdiseases. Semin Pediatr Neurol 1999;6:68–77.

6. Salibi N, Brown MA. Clinical MR spectroscopy. New York: Wiley-Liss;1998.

7. Ross BD, Danielsen ER, Bluml S. Proton magnetic resonancespectroscopy: the new gold standard for diagnosis of clinical and sub-clinical hepatic encephalopathy? Dig Dis 1996;14:30–39.

8. Simmons A, Small M, Moore E, Williams SCR. Serial precision ofmetabolite peak area ratios and water referenced metabolite peak areasin proton MR spectroscopy of the human brain. Magn Reson Imaging1998;16:319–330.

9. Schirmer T, Auer DP. On the reliability of quantitative clinical mag-netic resonance spectroscopy of the human brain. NMR Biomed 2000;13:28–36.

10. Brooks WM, Friedman SD, Stidley CA. Reproducibility of 1H-MRSin-vivo. Magn Reson Med 1999;41:193–197.

Reproducibility of Localized 2D COSY 947

11. Marshall I, Wardlaw J, Cannon J, Slattery J, Sellar RJ. Reproducibility ofmetabolite peak areas in 1H MRS of brain. Magn Reson Imaging 1996;14:281–292.

12. Duc CO, Weber OM, Trabesinger AH , Meier D, Bosiger P. Quantitative1H MRS of the human brain in vivo based on the simulation phantomcalibration strategy. Magn Reson Med 1998;39:491–496.

13. Wardlaw JM. Letter to the editor. Magn Reson Imaging 1998;17:483–487.

14. Haase A, Schuff N, Norris D, Leibfritz D. Localized COSY NMR spec-troscopy. In: Proceedings of the 7th Annual Meeting of SMRM, NewYork, 1987. p 1051.

15. Blackband SJ, McGovern KA, McLennan IJ. Spatially localized two-dimensional spectroscopy. SLO-COSY and SLO-NOESY. J Magn Reson1988;79:184–189.

16. Berkowitz BA, Wolff SD, Balaban RS. Detection of metabolites in vivousing 2D proton homonuclear correlated spectroscopy. J Magn Reson1988;79:547–553.

17. Behar KL, Ogino T. Assignment of resonances in the 1H spectrum of ratbrain by two dimensional shift correlated and J-resolved NMR spec-troscopy. Magn Reson Med 1991;17:285–303.

18. Cohen Y, Chang LH, Litt L, James TL. Spatially localized COSY spectrafrom a surface coil using phase-encoding magnetic field gradients. JMagn Reson 1989;85:203–208.

19. Brereton IM, Galloway GJ, Rose SE, Doddrell DM. Localized two-dimensional shift correlated spectroscopy in humans at 2Tesla. MagnReson Med 1994;32:251–257.

20. Ryner LN, Sorenson JA, Thomas MA. 3D localized 2D NMR spectros-copy on an MRI scanner. J Magn Reson 1995;107:126–137.

21. Thomas MA, Ryner LN, Mehta M, Turski P, Sorenson JA. Localized 2DJ-resolved 1H MR spectroscopy of human brain tumors in vivo. J MagnReson Imaging 1996;6:453–459.

22. Kreis R, Boesch C. Spatially localized, one- and two-dimensional NMRspectroscopy and in vivo application to human muscle. J Magn Reson1996;B113:103–118.

23. Hurd RE, Gurr D, Sailasuta N. Proton spectroscopy without watersuppression: the oversampled J-resolved experiment. Magn Reson Med1998;40:343–347.

24. Dreher W, Leibfritz D. Detection of homonuclear decoupled in vivoproton NMR spectra using constant time chemical shift encoding:CT-PRESS. Magn Reson Imaging 1999;17:141–150.

25. Thomas MA, Yue K, Binesh N, Davanzo P, Kumar A, Siegel B, Frye M,Curran J, Lufkin R, Martin P, Guze B. Localized two-dimensional shiftcorrelated MR spectroscopy of human brain. Magn Reson Med 2001;46:58–67.

26. Yue K, Marumoto A, Binesh N, Thomas MA. 2D JPRESS of humanprostates using an endorectal receiver coil. Magn Reson Med 2002;47:1059–1064.

27. Swanson MG, Vigneron DB, Sailasuta N, Hurd RE, Kurhanewicz J.Single-voxel 2D J-resolved spectroscopy of the in situ human prostate.In: Proceedings of the 9th Annual Meeting of ISMRM, Glasgow, Scot-land, 2001. p 272.

28. Govindaraju V, Young K, Maudsley AA. Proton NMR chemical shiftsand coupling constants for brain metabolites. NMR Biomed 2000;13:129–153.

29. Banakar S, Venkatraman T, Yue K, Binesh N, Thomas MA. Asymmetryof localized two-dimensional shift-correlated MR spectroscopy. Inter-national Conference on Mathematics and Engineering Techniques inMedicine and Biological Sciences, Las Vegas, 2002. p 500–504.

30. Ernst RR, Bodenhausen G, Wokaun A. Principles of NMR spectroscopyin one and two dimensions. Oxford: Oxford Publications; 1987. p 353.

31. Ziegler A, Izquierdo M, Remy C, Decorps M. Optimization of homo-nuclear two-dimensional correlation methods for in vivo and ex-vivoNMR. J Magn Reson 1995;B107:10–18.

948 Binesh et al.