Measurement of Hepatic Lipid: High-Speed T2-Corrected Multiecho Acquisition at 1 H MR Spectroscopy—A Rapid and Accurate Technique 1

Measurement of Hepatic Lipid:High-Speed T2-Corrected MultiechoAcquisition at 1H MR Spectroscopy—A Rapid and Accurate Technique1

Nashiely Pineda, PhDPuneet Sharma, PhDQin Xu, PhDXiaoping Hu, PhDMiriam Vos, MDDiego R. Martin, MD, PhD

Purpose: To evaluate the feasibility, accuracy, and reproducibility ofa fast breath-hold magnetic resonance (MR) spectroscopicmethod for T2-corrected hepatic lipid measurement inphantoms and in humans.

Materials andMethods:

All experiments were institutional review board approvedand HIPAA compliant; informed consent was obtainedfrom all subjects. The 15-second breath-hold high-speedT2-corrected multiecho (HISTO) MR spectroscopic tech-nique was developed to acquire multiple echoes in a singleacquisition, which enables the quantification of water andlipid T2, and subsequently to provide a corrected measureof hepatic lipid fraction. The accuracy of T2-corrected MRspectroscopy was evaluated in eight lipid phantoms dopedwith iron to simulate variable T2 effects. The mean abso-lute error of the HISTO technique with the known lipidamounts, as well as with uncorrected MR spectroscopicmeasures, was evaluated. The HISTO sequence was per-formed in 25 male subjects (mean age, 23.0 years � 19.2[standard deviation]) to evaluate measurement bias withconventional, uncorrected MR spectroscopy. Three addi-tional male subjects (mean age, 30.0 years � 1.0) wereexamined to assess reproducibility by using analysis ofvariance testing within subject and between separate im-aging sessions.

Results: The absolute error in quantifying lipid fraction by usingiron-doped lipid phantoms was less than 11% for theHISTO technique, compared with more than 50% for un-corrected MR spectroscopy. In the 25 human subjects,hepatic lipid measured by using HISTO differed signifi-cantly from that by using uncorrected MR spectroscopicmethods by 5.1% � 2.6. Analysis of variance of threeseparate imaging sessions with the HISTO technique indi-cated no significant variance (P � .13) in three subjects.

Conclusion: HISTO is an accurate, reproducible MR spectroscopic se-quence for quantifying hepatic lipid noninvasively. Evi-dence has shown this method to be feasible in vivo forclinical use.

� RSNA, 2009

1 From the Department of Biomedical Engineering, Geor-gia Institute of Technology, Atlanta, Ga (N.P., Q.X., X.H.);Department of Radiology, Emory University School ofMedicine, 1365 Clifton Rd NE, Clinic Building A-AT622,Atlanta, GA 30322 (P.S., D.R.M.); and Department of Pe-diatric Hepatology, Children’s Healthcare of Atlanta, At-lanta, Ga (M.V.). Received November 24, 2008; revisionrequested January 21, 2009; revision received February24; accepted March 17; final version accepted March 31.Address correspondence to D.R.M. (e-mail: [email protected] ).

� RSNA, 2009

ORIG

INAL

RESE

ARCH

�TE

CHNI

CAL

DEVE

LOPM

ENTS

568 radiology.rsnajnls.org ▪ Radiology: Volume 252: Number 2—August 2009

Note: This copy is for your personal non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at www.rsna.org/rsnarights.

Obesity-related fatty liver diseasehas become a common cause ofhepatitic inflammation and fibro-

sis in developed nations and affects anincreasing fraction of young adults andchildren (1–4). We currently depend onresults of invasive biopsies for the de-tection of and monitoring of this dis-ease. Liver biopsies may be painful andmay cause bleeding and other seriouscomplications (5). Results of liver biop-sies are not generally used for measur-ing and monitoring hepatic lipids. Vari-ous magnetic resonance (MR) imagingtechniques have been introduced for thedetection or visualization of hepatic lip-ids (6–9). Hydrogen 1 (1H) MR spec-troscopy offers a means for noninva-sively determining hepatic lipid and mayachieve high sensitivity for even lowamounts of or small changes in hepaticlipid (10), whereas conventional MR im-aging with Dixon methods (11–18) maybe hampered by limitations in signal-to-noise ratio and sensitivity.

Even though it is known that T2 ef-fects detriment the accuracy of hepaticlipid measures at MR spectroscopy, priorstudies (19,20) have not fully explored oraccounted for the variability and differ-ence in measurements of T2 on liver wa-ter and lipids, because often the T2 valueswere assumed constant. Failure to ac-count and correct for T2 effects may con-tribute to measurement variability be-

tween patients or within an individual pa-tient over time. Other limitations of priorstudies include relatively long imagingtimes (19,21,22), generally requiring mul-tiple acquisitions of spectra, which makesuch approaches impractical for potentialuse in common clinical applications.

MR spectroscopy has been used toassess T2 values of water signal by usingsingle-voxel stimulated-echo acquisitionmode (STEAM) sequences at a number ofecho times (TEs) and by fitting an expo-nential decay to the echo amplitude atdifferent TEs (6). However, this T2 mea-surement strategy requires separate ac-quisitions, which further prolongs exami-nation time. For clinical utility of a T2-corrected MR spectroscopic technique, itmust be performed in a time frame com-parable with that of other body MR imag-ing methods, preferably in a breath hold.To our knowledge, no body MR spectro-scopic method has been developed to ad-dress the needs of accuracy, reproducibil-ity, and efficiency for measuring hepaticlipid.

In this article, we describe an MRspectroscopic technique that we devel-oped that allows the rapid and simulta-neous acquisition of multiple echoes toassess hepatic lipid within a singlebreath hold. The purpose of this studywas to evaluate the feasibility, accuracy,and reproducibility of a fast breath-holdMR spectroscopic method for T2-cor-rected hepatic lipid measurement inphantoms and in humans.

Materials and Methods

PhantomsLiver phantoms containing knownamounts of lipid were constructed for theassessment of the measurement accuracy

and reproducibility of our MR spectro-scopic technique. The phantoms were ad-ditionally doped with varying amounts ofparamagnetic iron to reproduce differentlevels of susceptibility effects. Twelve200-mL phantoms were formed in250-mL sterile screw-top containers. Tis-sue water was simulated with 2% agar-water gels, to which varying concentra-tions of paramagnetic iron (Feridex; Ber-lex, Montville, NJ) (0, 0.1, 0.3, and 0.5mmol/L) and varying percentages of lip-ids (vegetable oil) (0%, 10%, and 30%per volume) were added at a pH of 7.00 �0.2 (standard deviation) prior to poly-merization. The range of lipid was se-lected to represent normal to moder-ately fatty liver, and iron levels repre-sented physiologic range seen in humanlivers (5,6). Micelles were produced bythe addition of 2 g of lecithin to emulsifythe lipid and reproduce morphologicpackaging analogous to lipid vacuoleswithin hepatocytes. Furthermore, mi-celle formation was essential for agarpolymerization.

SubjectsThis study was Health Insurance Port-ability and Accountability Act compli-ant and was approved by our institu-tional review board at Emory Univer-sity School of Medicine. Informedconsent was obtained. Twenty-five male

Published online before print10.1148/radiol.2523082084

Radiology 2009; 252:568–576

Abbreviations:HISTO � high-speed T2-corrected multiechoSTEAM � stimulated-echo acquisition modeTE � echo timeTR � repetition time

Author contributions:Guarantors of integrity of entire study, N.P., P.S., X.H.,D.R.M.; study concepts/study design or data acquisitionor data analysis/interpretation, all authors; manuscriptdrafting or manuscript revision for important intellectualcontent, all authors; manuscript final version approval, allauthors; literature research, N.P., P.S., X.H., M.V., D.R.M.;clinical studies, N.P., P.S., M.V., D.R.M.; experimentalstudies, N.P., P.S., Q.X., D.R.M.; statistical analysis, N.P.,P.S., D.R.M.; and manuscript editing, N.P., P.S., X.H.,D.R.M.

Authors stated no financial relationship to disclose.

Advances in Knowledge

� We have developed an MR spec-troscopic sequence for the liverthat can accurately determine he-patic lipid levels within a 15-sec-ond acquisition and that allowsoffline lipid fraction T2 correc-tion.

� We have shown that susceptibilityeffects require separate T2 cor-rection for water and lipid spectrato produce accurate measurementof the lipid fraction.

� Our technique provides reproduc-ible calculated hepatic lipid mea-surements, which makes the tech-nique potentially useful for clinicalapplications that require monitor-ing of lipid levels over time.

Implication for Patient Care

� The development of a rapid, accu-rate, and reproducible method forquantifying hepatic lipid will aid inthe evaluation and monitoring ofpatients with conditions associ-ated with elevated hepatic lipids,such as nonalcoholic fatty liverdisease.

TECHNICAL DEVELOPMENTS: New Technique for Measurement of Hepatic Lipid Pineda et al

Radiology: Volume 252: Number 2—August 2009 ▪ radiology.rsnajnls.org 569

subjects (mean age, 23.0 years � 19.2;age range, 11–71 years) known to haveor clinically suspected of having nonal-coholic fatty liver disease were studied.Referrals to our nonalcoholic fatty liverdisease program are originally made onthe basis of elevated body mass indexand transient or persistent elevatedliver enzymes. Exclusion criteria includedcontraindications to MR imaging (claustro-phobia, metal implants, or pacemakers) orinability to suspend respiration for rele-vant acquisitions. No subjects were ex-cluded from this study and were chosensequentially on the basis of the referrals.Intrasubject variability of the techniquewas determined from repeated MR spec-troscopic acquisitions in three subjects(ages: 31, 29, and 30 years).

1H MR SpectroscopyWe developed a high-speed T2-cor-rected multiecho (HISTO) 1H MR spec-troscopic technique. The pulse se-quence design and programming wasdone with an imaging platform (Sie-mens Medical Systems, Erlangen, Ger-many) by using specially available ven-dor software (Integrated DevelopmentEnvironment for Applications [IDEA],VB15A; Siemens Medical Solutions,Malvern, Pa). Modifications were per-formed with the sequence languageANSI-C��. A conventional STEAM se-quence was chosen for modification dueto its inherently short TE (approxi-mately 20 msec). To further shorten theminimum available TE, the spoiling gra-

dient amplitude between the initial ra-diofrequency pulses was doubled from11.5 to 23 mT/m (Fig 1). Because thezeroth moment of the gradient was heldconstant, this modification effectivelyreduced the minimum allowed TE/2.From this process, a minimum TE of 12msec was achieved. Furthermore, thenumber of “dummy” preparation radio-frequency pulses and number of signalsacquired were minimized (to zero andone, respectively) to maintain a shortoverall imaging time, which allows ex-amination completion within a breathhold. Finally, with HISTO, the modifiedSTEAM sequence was repeated follow-ing a user-defined TR period. Five loopswere installed to allow for five consecu-tively acquired TEs. These TEs weremade freely variable in the design; how-ever, the number of TR loops was heldconstant (Fig 1).

A 1.5-T MR imaging system (Avanto;Siemens Medical Solutions, Malvern,Pa) with a surface phased-array coil wasused to obtain HISTO spectra in phan-toms and human subjects. Imaging pa-rameters, which were the same forphantoms and humans, included a TR of3000 msec, a mixing time between thesecond and third radiofrequency pulsesof 10 msec, and five TEs of 12, 24, 36,48, and 72 msec. (TE was 12, 15, 18,21, and 24 msec for high iron levels.) Atotal of 1024 points were acquired at abandwidth of 1200 Hz, with one signalacquired by using a voxel size of 30 �30 � 30 mm.

In the human subjects, HISTO spec-tra were obtained in two different loca-tions by using voxels placed either in theright anterior (segment VIII) or left lat-eral (segment II) liver segments. Theseregions provided a highly homogeneoussampling of liver tissue, without adversecontamination from major hepatic ves-sels. Each HISTO acquisition was com-pleted in 15 seconds during a singlebreath hold.

Selection of an adequate TR forHISTO was made on the basis of liverT1 measurements in a healthy volunteerby using a TE of 30 msec and TRs of 0.6,1, 2.5, 5, 10, and 20 seconds, whichresulted in 747.4 msec for T1 of liver.From the signal of a specific metaboliteduring the STEAM sequence (23), it canbe shown that the effect of signal satura-tion from incomplete T1 is within 3%. Inthese experiments, we held TR constantat 3000 msec for all HISTO acquisitions.This assumption was tested in vivo incomparison to a single-average STEAMsequence (in seven of 25 subjects),which had identical parameters to theHISTO sequence, except for a TR muchgreater than five T1s.

Spectrum PostprocessingThe phantom spectra were analyzedwith custom MATLAB-based software(MATLAB 7.04.365; Mathworks, Natick,Mass). The real part of the spectrumwas phased and filtered with a cutofffrequency of 240 Hz. The algorithm in-tegrated the water (from 3.6 to 5.8

Figure 1

Figure 1: Pulse sequence diagram of HISTO MR spectroscopic sequence. Each concatenation is a modified STEAM sequence (mixing time [TM] � 10 msec) with onesignal acquired, which is then combined to produce a series of acquisitions within a single breath hold. In this formulation, each STEAM module has aTE (12, 24, 36, 48, or 72 msec) and is separated by repetition time (TR) of 3000 msec, resulting in a 15-second breath hold. Gx, Gy, and Gz aregradients in each of the respective planes. RF � radiofrequency, t � time.



ppm) and lipid (methylene and methylsignals of �-CH2, (CH2)n-2, and CH3

from 0 to 3.6 ppm) signals for each TE.Postprocessing of the in vivo data

was performed by using LCModel (24)(LCModel, version 6.2– 0; StephenProvencher, Oakville, Ontario, Can-ada). The analysis of the spectrum wasperformed in the frequency domain byusing a linear combination of modelspectra. By setting the “SPTYPE” vari-able equal to “liver-1,” the software an-alyzed the spectra assuming that therewere lipid and water, possibly cholinesignals, and signals in the 3.4–3.8-ppmregion (sometimes attributed to glyco-gen and other metabolites). Commer-cial software was implemented for thein vivo data to show potential practical-ity and transferability to other centers.

Lipid Content and R2 EstimationWith spectral integrals (S) measured ateach TE by using HISTO, monoexpo-nential curve fits were performed withleast-squares approximation by usingthe equation S � M0 exp(�R2 � TE) toestimate R2 and M0 (the equilibriummagnetization) for water and lipid sig-nals individually. Subsequently, T2-cor-rected lipid content was calculated as:Percentage of lipid � [M0lipid/(M0lipid �M0water)] � 100. Previous methods cal-

culated lipid content by looking at frac-tional lipid signal for a given TE (12–14,18,25). To show that previous ap-proaches without T2 correction introduceinaccuracies, lipid content was also calcu-lated by using the integrated signal at TEsof 12 and 24 msec.

Reproducibility of R2 Measurements andLipid ContentIn two phantoms (10% lipids, 0.3mmol/L iron; 30% lipids, 0.3 mmol/Liron), we performed HISTO measure-ments three different times on the sameday with the same position, acquisitionparameters, and shimming values (in-trasession measurement) and then re-peated the measurements on two moredays (intersession measurement) toevaluate the measurement reproducibil-ity.

To ascertain the intrasubject vari-ability, HISTO was performed in threesubjects, and the measurement was re-peated three times for the assessmentof within-session measurement varia-tion. This entire procedure was re-peated at three separate imaging ses-sions to determine intersession variabil-ity. After each session, the subject wasremoved and then returned for com-plete repetition of all steps, includingrepositioning of the subject and coil and

all imager adjustments, which includedshimming and localization.

Statistical AnalysisData analysis was performed with soft-ware (Excel 2003, Microsoft, Redmond,Wash; SAS, SAS Institute, Cary, NC).All results were expressed as means �standard deviations. Accuracy of HISTOin phantoms was determined by usingmeasures of absolute errors across ironlevels. Measurement similarity betweenconventional STEAM and HISTO wasassessed by using a Pearson correlationcoefficient (r), along with 95% confi-dence intervals, without assuming nor-mal distribution, which required aFisher z transformation to covert corre-lation coefficient to a value that was nor-mally distributed. Within-subject differ-ences between percentage of lipid mea-surements using T2 correction (TE � 0msec) and that using a TE of 24 msecwere determined by using a Wilcoxonsigned rank test. For reproducibilitymeasures of R2 of water, R2 of lipid,and percentage of hepatic lipid by usingHISTO, pooled standard deviation wascalculated to estimate the differences inmeasures. In addition, within-subjectanalysis of variance testing was per-formed for each variable with signifi-cance set to P � .05.

Results

Lipid Content and R2 Assessment inPhantomsTable 1 summarizes the results of theestimated lipid content, with and with-out T2 correction, in the phantomscomposed of 10% and 30% lipid. Lipidpercentage calculated with a shortestpossible TE at STEAM (TE � 12 msec)showed a large bias that increased assusceptibility effects (related to the ironcontent) increased (�50% mean er-ror). However, HISTO, employing a T2correction, remained relatively stable(approximately 10% mean error) andprovided a consistently more accuratemeasurement of the lipid fraction for alliron concentrations. Without T2 cor-rection, the calculated lipid content ex-hibited a significant bias that increased

Table 1

Lipid Percentage in Phantoms and Accumulating Errors with Increasing TEs andSusceptibility

ParameterT2-corrected LipidPercentage

Lipid Percentage with12-msec TE

Lipid Percentage with24-msec TE

10% lipid phantoms0 mmol/L iron 11.0 10.1 10.30.1 mmol/L iron 11.6 11.9 14.60.3 mmol/L iron 9.1 18.0 37.10.5 mmol/L iron 9.4 28.8 61.2Mean � standard deviation 10.2 � 1.2 17.2 � 8.4 30.8 � 23.4Mean absolute error (%) 10.4 71.9 208.0

30% lipid phantoms0 mmol/L iron 29.7 29.2 28.70.1 mmol/L iron 27.2 32.2 38.60.3 mmol/L iron 34.4 57.4 80.60.5 mmol/L iron 33.8 64.7 83.2Mean � standard deviation 31.3 � 3.4 45.8 � 17.8 57.8 � 28.2Mean absolute error (%) 9.4 54.2 94.8

Note.—Unless otherwise indicated, data are percentages.



with an increase in TE or an increase insusceptibility effects. This trend is dueto the different and higher rate ofchange in water R2 compared with lipidR2. Water R2 correlates with iron con-centration, and a regression analysis inthis study revealed a highly linear rela-tionship: R2 � 280.1 � Fe � 14.6 (R2 �

0.98, P � .001), with iron (Fe) in unitsof millimoles per liter and R2 in sec�1.The standard error in R2 is 8.5 sec�1.The intercept (14.6) represents the av-erage R2 of agarose, without iron,across the tested range of lipid.

Intra- and intersession measure-ments for the 10% lipids and 0.3

mmol/L iron and 30% lipids and 0.3mmol/L iron phantoms showed a highdegree of reproducibility. Intersessionreproducibility had higher overall varia-tion, but standard deviation remainedless than 10% of the mean. Analysis ofvariance between individual measure-ments showed no significant differences

Figure 2

Figure 2: Spectroscopic analysis and fitting for water and lipid spectra for one subject in vivo. (a) Raw data were processed offline, and a high degree of fit correlationwas achieved for individual spectra. Each TE of the HISTO sequence generated water and lipid spectra, which were similarly processed. (b) Shown together in one subject,the processed spectra of water and lipid show TE dependency.

Figure 3

Figure 3: Signal integrals from water and lipid spectrum fits were used to measure T2 decay and estimate the equilibrium signal at TE of 0 msec. Left: Graph of T2 de-cay of water and lipid in this subject shows marked differences in decay rate, which was found for all subjects. The estimated equilibrium signal is used to correct for theseT2 differences when calculating hepatic lipid fraction. Right: Graph shows error of not performing this correction procedure. The calculated lipid fraction changes de-pending on TE used for analysis. HISTO allows acquisition of five TEs to perform necessary T2-corrected procedures. a.u. � arbitrary units.



in R2 of water and percentage of lipidestimation with HISTO (P � .1). Thepooled standard deviations of R2 of wa-ter were 0.3 and 0.6 sec�1 for the 10%lipids and 0.3 mmol/L iron phantom andthe 30% lipids and 0.3 mmol/L ironphantom, respectively, and 0.1% and1.0% for R2 of lipids, respectively.

Application of HISTO in VivoIn all in vivo experiments, high quality curvefits were obtained by using LCModel(Fig 2). The series of acquired spectrafor water and lipid at each of the fiveTEs enabled the variability of the per-centage of hepatic lipid measurementsto be evaluated (Fig 3). These measure-ments were based on the calculated wa-ter and lipid signal integrals and R2measures. In 25 subjects, we were ableto quantify the R2 of lipid, R2 of water,and percentage of hepatic lipid by usingHISTO (Table 2). The results indicateda measurable difference between T2-corrected percentage of lipid and per-

centage of lipid with TE of 24 msec. Thisintersubject bias was significantly differ-ent (P � .05), which indicates there wasa systematic overestimation of percent-age of hepatic lipid measurements ob-tained with a TE of 24 msec versusthose obtained with T2 correction. Thisresult was also found for TEs of morethan 24 msec.

Reproducibility of Measurements inHumansIn three subjects, the overall pooledstandard deviation of R2 of water, R2 oflipid, and percentage of hepatic lipidmeasurements was low (Table 3), withmaximum intra- and intersession vari-ability occurring with subject 2 for R2 oflipid calculations. The R2 of water washighly reproducible, determined by re-peating examinations in the same indi-vidual; the pooled standard deviationwas 1.02 sec�1. Similarly, percentage ofhepatic lipid measurements exhibitedhighly acceptable standard deviations

(pooled standard deviation � 0.69%).Despite the apparent sensitivity of theHISTO measurement to low lipid con-tent (subject 2), a within-subject analy-sis of variance showed no statistical sig-nificance between repeated imaging ses-sions (P � .10) in the three subjects,establishing the reproducibility of theHISTO technique.

Comparison of HISTO and STEAM toValidate TRConventional single-echo STEAM acqui-sition and HISTO acquisition showedhigh correlation in corrected lipid levelcalculations in humans (r � 0.97). TheT2-corrected percentage of lipids, R2 ofwater, and R2 of lipid are given in Table 4;the values obtained with both tech-niques (STEAM and HISTO) were sig-nificantly similar (P � .01), thus validat-ing an adequate TR was selected for theHISTO sequence and that the sequencemodification used to formulate HISTOdid not introduce differences. In addi-tion, the measured difference of thethree parameters showed no lineartrend.

Discussion

This study demonstrated the feasibility,accuracy, and reproducibility of a fastMR spectroscopic technique for quanti-fying hepatic lipid. We reconfigured anMR spectroscopic STEAM sequence toacquire all of the data necessary for T2-corrected lipid measurement in a singlebreath-hold acquisition. To our knowl-edge, previous studies have not em-

Table 2

Variation in Liver Lipid R2 and Water R2 and Effect on Calculated Lipid Fraction inRelation to TE in 25 Subjects

Parameter Range Mean � Standard Deviation

Lipid R2 14.5–21.8 17.6 � 1.8Water R2 24.3–35.7 28.8 � 2.8T2-corrected percentage of hepatic lipid 6.6–36.5 21.4 � 8.0Percentage of hepatic lipid with 24-msec TE 7.3–42.1 24.0 � 9.3Bias* 0.7–9.33† 5.1 � 2.6

* Bias is the difference between T2-corrected and 24-msec TE measures.† P � .001 (Wilcoxon signed rank test).

Table 3

Reproducibility Determined from Three Separate Imaging Sessions

ParameterWater R2 (sec�1) Lipid R2 (sec�1) Percentage of Hepatic Lipid

Session 1 Session 2 Session 3 Session 1 Session 2 Session 3 Session 1 Session 2 Session 3

Subject 1 33.6 � 0.6 33.3 � 0.6 33.4 � 0.9 16.4 � 0.2 17.2 � 0.6 16.7 � 1.1 14.5 � 0.3 15.2 � 0.5 13.2 � 0.8Subject 2 31.0 � 1.5 33.0 � 0.8 32.9 � 0.2 26.8 � 1.5 27.2 � 3.5 22.6 � 0.3 6.1 � 0.5 5.8 � 0.8 4.8 � 0.3Subject 3 34.0 � 1.3 35.2 � 0.5 32.6 � 0.8 17.5 � 0.4 18.3 � 0.7 17.2 � 1.2 11.8 � 0.2 11.7 � 0.5 11.7 � 0.2Pooled standard deviation 1.02 1.53 0.69P value* .49 .25 .13

Note.—Unless otherwise indicated, data are means � standard deviations.

* Significance level for three repeated examinations (within-subject analysis of variance).



ployed T2 corrections independently forwater and lipid with a single breath-holdMR spectroscopic acquisition methodthat could be used clinically.

The advantages of HISTO in compar-ison to other methods include the follow-ing: (a) it does not require complex mul-tisequence acquisitions to achieve correc-tions for T2 effects; (b) it allows theassessment of hepatic lipid and simulta-neous measurement of R2 of water andR2 of lipid for corrected hepatic lipid cal-culation; (c) it has been optimized for fastclinical application (15 seconds), makingit feasible in both adult and pediatric pa-tients; and (d) it provides technical ro-bustness with high reproducibility. In clin-ical applications, the postprocessing stepscan be performed offline and can be au-tomated for speed and simplification.

We demonstrated the effect of T2correction on the accuracy of the HISTOtechnique in phantoms. In phantoms withknown lipid content, the error on lipidestimation without T2 correction in-creased with increasing susceptibility ef-fects, which was controlled by iron con-tent. Although the need for correctingsusceptibility effects diminishes as thephantom iron level decreases, the use ofuncorrected measurements with a TE ofmore than 20 msec, as have been previ-ously reported (10,11), will lead tomarked miscalculation of the lipid frac-tion even at low iron levels. The estima-tion of lipid content, therefore, will de-pend highly on the TE of the MR spectro-scopic sequence, as exemplified in thisstudy with both phantom and human ev-idence. The reason for this effect is partlydue to the observation that R2 of waterand R2 of lipid varies differently, asshown in Table 3 and in a previous study(10). Human liver normally has variablelevels of hepatic iron that would accountfor some of this T2 variability; pathologi-cally increased iron deposition occurs in anumber of chronic liver diseases (26), al-though it is not generally seen to bepathologically elevated as part of nonalco-holic fatty liver disease. Other factors ex-pected to alter liver T2 values include in-terstitial edema related to inflammatoryconditions, which may occur because ofunderlying acute and chronic changes ofhepatitis. It may be expected that factors

affecting susceptibility in the liver maychange over time within an individual.Application of HISTO would account forthese changes and would be expected toimprove accuracy for detection ofchanges in hepatic lipid within an individ-ual. This is critical for monitoring diseaseprogression or improvement and for ther-apy monitoring.

Another requirement of a practicalMR spectroscopic method is high repro-ducibility. With HISTO, we have shownreproducibility in phantoms (pooled stan-dard deviation � 1.0% in repeated phan-tom measurements) and in humans(pooled standard deviation, 0.69%). Infurther evaluation by using between-session comparisons of HISTO, we dem-onstrated highly consistent results in hu-mans, which involved repeat examinationfollowing subject and voxel repositioning.As expected with decreased signal-to-noise ratio, reproducibility diminishes ashepatic lipid decreases; however, the ef-fect of signal-to-noise ratio does not sig-nificantly deteriorate the ability to mea-sure a broad range of hepatic lipid levels,as shown with this study. As further stud-ies are performed, the critical lower he-patic lipid threshold should be investi-gated as part of a comprehensive under-standing of both the clinical importance oflow hepatic lipid and the overall sensitiv-ity of HISTO.

Selection of TEs were guided by our

data showing that R2 decay for waterspectra has the greatest change duringthe first 20–30 msec. The HISTO se-quence was designed to allow for a shortfirst echo (TE � 12 msec) to account forthis requirement. We additionallyshowed that a TR of 3000 msec providedan optimal balance between allowing sig-nal recovery (at least 97%) and overallspeed of acquisition. We showed this bydirectly measuring the T1 of liver and alsoby comparing HISTO to separately ac-quired MR spectroscopic STEAM echoeswhere the TR is essentially infinite. In thislatter comparison, the difference be-tween HISTO and STEAM results werenot significantly different.

The concepts of MR spectroscopicT2 correction have been raised previ-ously, as has the use of MR spectros-copy for hepatic lipid measurement.However, a rapid method suitable forroutine clinical use has not been fullydeveloped. Most of the reported MRspectroscopic methods require severalminutes of acquisition, which may notbe amenable to broad application. Someprevious methods have required coordi-nated breath holding within the TR in-terval or synchronization to the breath-ing pattern of the patient to minimizeline broadening (25,27). The HISTOtechnique overcomes these limitationsof long imaging times, and simulta-neously provides the data required for

Table 4

Comparison of T2-corrected Single-Echo STEAM and HISTO Sequences

ParameterWater R2 (sec�1) Lipid R2 (sec�1)

Percentage of HepaticLipid

STEAM HISTO STEAM HISTO STEAM HISTO

Subject 1 26.9 26.2 � 0.4 17.2 17.2 � 0.5 22.7 23.5 � 0.7Subject 2 24.4 25.4 � 0.4 19.1 20.9 � 0.9 16.6 18.2 � 0.4Subject 3 30.1 29.8 � 0.4 18.0 16.6 � 0.7 35.5 34.0 � 1.4Subject 4 31.2 31.2 � 0.6 17.3 15.4 � 1.3 33.5 36.6 � 1.6Subject 5 33.9 33.9 � 1.3 18.2 17.7 � 0.6 12.1 11.8 � 0.1Subject 6 31.8 33.4 � 0.1 17.2 16.8 � 0.4 14.9 14.3 � 1.0Subject 7 32.5 32.3 � 1.1 23.2 25.5 � 2.6 5.5 5.6 � 0.7Pearson correlation coefficient* 0.97 (0.81)† 0.96 (0.75)† 0.99 (0.93)†

Bias‡ �0.20 (1.39, �1.80) 0.04 (3.18, �3.11) �0.46 (2.58, �3.50)

Note.—Unless otherwise indicated, data are means or means � standard deviations.

* Data in parentheses are the lower bound of the 95% confidence interval.† Significant correlation (P � .01).‡ Bias is the average difference between STEAM and HISTO, with 95% confidence intervals in parentheses.



T2 correction. Prior methods have de-pended on separately acquired se-quences for this purpose. Wang et al (6)applied 12 TEs (from 1.5 to 100 msecwith TR of 2000 msec) for R2 estima-tion. The total time for the MR imagingand MR spectroscopic acquisitions wasabout 25 minutes. By using HISTO, wedistributed the five TEs from 12 to 72msec, enabling monoexponential fitting,while balancing the requirements to de-velop a clinical method obtained in asingle breath hold.

Another potential application ofHISTO is simultaneous quantification ofhepatic iron. We found a highly signifi-cant linear relationship between the R2of water and the amount of iron in thephantoms. In addition, the predicted re-lationship between R2 of water and ironis independent of the lipid concentra-tion. For in vivo iron quantification byusing HISTO, it would be necessary torecalculate the iron relaxivity for thespecific HISTO parameters used to en-sure accuracy. Although not part of ourprimary aim, this does represent an at-tractive additional capability for the MRspectroscopic technique presented here.

MR imaging techniques performedby using Dixon methods or opposed-phase imaging have been evaluated inseveral studies for assessing lipid con-tent in liver and have even been com-bined with MR spectroscopy. For exam-ple, Chang et al (25) reported an accu-rate estimation of lipid fraction by usingopposed-phase imaging with qualitative1H MR spectroscopy to resolve the am-biguity of lipid or water dominance.However, signal loss on in-phase im-ages, caused by the presence of liveriron, is a potential pitfall in the determi-nation of liver lipid percentage at op-posed-phase MR imaging. Machann et al(21) proposed a fat-selective MR imagingtechnique for the assessment of the amountand spatial distribution of hepatic lipid. Inthat study, fat-selective MR imaging and 1HMR spectroscopy were compared and ledto very similar results showing correlationcoefficients of more than 0.95. Unfortu-nately, for both of these techniques, the T2effects were uncorrected, and T2 measuresof the subjects’ livers could not be deter-mined from the results.

The single voxel analysis was a limi-tation of the HISTO technique. In somepatients, lipid accumulation may be het-erogeneous throughout the liver, whichmay lead to sampling errors. One pro-posal is to use a dual-echo acquisitionfor determining heterogeneous foci offatty accumulation; another proposal isto use multiple sample sites within ana-tomically defined liver locations and toaverage the results. Even with MR imag-ing techniques for lipid detection, quanti-fication ultimately involves anatomicplacement of regions of interest. This pro-cess becomes comparable to placementof a spectroscopic voxel of interest.

Another limitation of the study wasthat we did not compare MR spectro-scopic results against liver biopsy sam-ples, so we were not able to establishthe accuracy of the technique in vivo. Inaddition, in vivo data were obtained in alimited number of human subjects. Theaim for this technical study was to char-acterize a practical MR spectroscopicapproach. We did not attempt a fullclinical investigation of the diagnosticperformance and utility in our study.

In summary, we have developed arapid breath-hold MR spectroscopictechnique, HISTO, which acquires dataat multiple TEs for more accurate quan-tification of T2-corrected hepatic lipidlevels.

References1. Patton HM, Sirlin C, Behling C, Middleton

M, Schwimmer JB, Lavine JE. Pediatric non-alcoholic fatty liver disease: a critical ap-praisal of current data and implications forfuture research. J Pediatr GastroenterolNutr 2006;43(4):413–427.

2. Schwimmer JB, Deutsch R, Kahen T, LavineJE, Stanley C, Behling C. Prevalence of fattyliver in children and adolescents. Pediatrics2006;118(4):1388–1393.

3. Fishbein MH, Miner M, Mogren C, Chalekson J.The spectrum of fatty liver in obese children andthe relationship of serum aminotransferases toseverity of steatosis. J Pediatr GastroenterolNutr 2003;36(1):54–61.

4. Radetti G, Kleon W, Stuefer J, Pittschieler K.Non-alcoholic fatty liver disease in obese chil-dren evaluated by magnetic resonance imaging.Acta Paediatr 2006;95(7):833–837.

5. Reddy JK, Rao MS. Lipid metabolism and

liver inflammation. II. Fatty liver disease andfatty acid oxidation. Am J Physiol Gastroin-test Liver Physiol 2006;290(5):G852–G858.

6. Wang ZJ, Haselgrove JC, Martin MB. Evalu-ation of iron overload by single voxel MRSmeasurement of liver T2. J Magn Reson Im-aging 2002;15(5):395–400.

7. Thampanitchawong P, Piratvisuth T. Liverbiopsy: complications and risk factors.World J Gastroenterol 1999;5(4):301–304.

8. Fishbein MH, Gardner KG, Potter CJ,Schmalbrock P, Smith MA. Introduction offast MR imaging in the assessment of hepaticsteatosis. Magn Reson Imaging 1997;15(3):287–293.

9. Hollingsworth KG, Abubacker MZ, JoubertI, Allison ME, Lomas DJ. Low-carbohydratediet induced reduction of hepatic lipid con-tent observed with a rapid non-invasive MRItechnique. Br J Radiol 2006;79(945):712–715.

10. Schuchmann S, Weigel C, Albrecht L, et al.Non-invasive quantification of hepatic fatfraction by fast 1.0, 1.5 and 3.0 T MR imag-ing. Eur J Radiol 2007;62(3):416–422.

11. Chan DC, Watts GF, Ng TW, Hua J, Song S,Barrett PH. Measurement of liver fat bymagnetic resonance imaging: relationshipwith body fat distribution, insulin sensitivityand plasma lipids in healthy men. DiabetesObes Metab 2006;8(6):698–702.

12. Szczepaniak LS, Babcock EE, Schick F, et al.Measurement of intracellular triglyceridestores by 1H spectroscopy: validation in vivo.Am J Physiol Endocrinol Metab 1999;276:E977–E989.

13. Kim H, Taksali SE, Dufour S, et al. Compar-ative MR study of hepatic fat quantificationusing single-voxel proton spectroscopy, two-point Dixon and three-point IDEAL. MagnReson Med 2008;59(3):521–527.

14. Dixon WT. Simple proton spectroscopic im-aging. Radiology 1984;153(1):189–194.

15. Fishbein MH, Stevens WR. Rapid MRI usinga modified Dixon technique: a non-invasiveand effective method for the detection andmonitoring of fatty metamorphosis of theliver. Pediatr Radiol 2001;31(11):806–809.

16. Glover GH, Schneider E. Three-point Dixontechnique for true water/fat decompositionwith B0 inhomogeneity correction. Magn Re-son Med 1991;18(2):371–383.

17. Glover GH. Multipoint Dixon technique forwater and fat proton and susceptibility imag-ing. J Magn Reson Imaging 1991;1(5):521–530.

18. Reeder SB, Wen Z, Yu H, et al. Multicoil



Dixon chemical species separation with aniterative least-squares estimation method.Magn Reson Med 2004;51(1):35–45.

19. Thomas EL, Hamilton G, Patel N, et al. He-patic triglyceride content and its relation tobody adiposity: a magnetic resonance imag-ing and proton magnetic resonance spectros-copy study. Gut 2005;54(1):122–127.

20. Szczepaniak LS, Nurenberg P, Leonard D,et al. Magnetic resonance spectroscopy tomeasure hepatic triglyceride content: preva-lence of hepatic steatosis in the general pop-ulation. Am J Physiol Endocrinol Metab2005;288(2):E462–E468.

21. Machann J, Thamer C, Schnoedt B, et al.

Hepatic lipid accumulation in healthy sub-jects: a comparative study using spectralfat-selective MRI and volume-localized1H-MR spectroscopy. Magn Reson Med2006;55(4):913–917.

22. O’Regan DP, Callaghan MF, Wylezinska-Arridge M, et al. Liver fat content and T2*:simultaneous measurement by usingbreath-hold multiecho MR imaging at 3.0T—feasibility. Radiology 2008;247(2):550–557.

23. Kreis R. Quantitative localized (1)H MRspectroscopy for clinical use. J Prog NMR1997;31:155–195.

24. Provencher SW. Automatic quantitation of

localized in vivo 1H spectra with LCModel.NMR Biomed 2001;14(4):260–264.

25. Chang JS, Taouli B, Salibi N, Hecht EM,Chin DG, Lee VS. Opposed-phase MRI forfat quantification in fat-water phantomswith 1H MR spectroscopy to resolve ambi-guity of fat or water dominance. AJR Am JRoentgenol 2006;187(1):W103–W106.

26. Westphalen AC, Qayyum A, Yeh BM, et al.Liver fat: effect of hepatic iron depositionon evaluation with opposed-phase MR im-aging. Radiology 2007;242(2):450–455.

27. St Pierre TG, Clark PR, Chua-anusorn W.Noninvasive measurement and imaging of liveriron concentrations using proton magnetic res-onance. Blood 2005;105(2):855–861.



Documents

Measurement of Hepatic Lipid: High-Speed T2-Corrected Multiecho Acquisition at 1 H MR Spectroscopy—A Rapid and Accurate Technique 1