rights / license: research collection in copyright - non …29643/et… · diss. eth no. 17140...

Research Collection

Doctoral Thesis

Detection of J-coupled metabolites in magnetic resonancespectroscopy

Author(s): Lange, Thomas

Publication Date: 2007

Permanent Link: https://doi.org/10.3929/ethz-a-005398016

Rights / License: In Copyright - Non-Commercial Use Permitted

This page was generated automatically upon download from the ETH Zurich Research Collection. For moreinformation please consult the Terms of use.

ETH Library

https://doi.org/10.3929/ethz-a-005398016

http://rightsstatements.org/page/InC-NC/1.0/

https://www.research-collection.ethz.ch

https://www.research-collection.ethz.ch/terms-of-use

Diss. ETH No. 17140

Detection of J -Coupled Metabolites

in Magnetic Resonance Spectroscopy

A dissertation submitted to the

Swiss Federal Institute of Technology Zurich

for the degree of

Doctor of Sciences

presented by

Thomas LangeDipl. phys.

born August 22nd, 1974citizen of Germany

accepted on the recommendation of

Prof. Dr. Peter BoesigerProf. Dr. Roland Kreis

Zurich, 2007

Contents

Summary 3

Zusammenfassung 7

Introduction and Objectives 11

1 Basics of Magnetic Resonance Spectroscopy 151.1 The Chemical Shift . . . . . . . . . . . . . . . . . . . . . . . 151.2 Scalar Coupling . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.2.1 Classical Picture . . . . . . . . . . . . . . . . . . . . 161.2.2 Quantum-Mechanical Picture . . . . . . . . . . . . . 20

1.3 Spatial Localisation . . . . . . . . . . . . . . . . . . . . . . 211.4 Two-Dimensional Spectroscopy . . . . . . . . . . . . . . . . 25

2 Pitfalls in Lactate Measurements at 3 T 272.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.2 Description of the Technique and Results . . . . . . . . . . 28

2.2.1 Patient Measurements . . . . . . . . . . . . . . . . . 282.2.2 Phantom Measurements . . . . . . . . . . . . . . . . 29

2.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342.4 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3 Improved Two-Dimensional J-Resolved Spectroscopy 413.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 413.2 Theory and Methods . . . . . . . . . . . . . . . . . . . . . . 433.3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . 503.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 553.5 Appendix: Sensitivity Comparison . . . . . . . . . . . . . . 55

1

2 CONTENTS

4 Prostate Spectroscopy at 3 Tesla Using Two-DimensionalS-PRESS 574.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 574.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 594.3 Materials and Methods . . . . . . . . . . . . . . . . . . . . . 624.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 664.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 724.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 774.7 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

5 Quantitative J-Resolved Prostate Spectroscopy Using 2DFitting 795.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 795.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . 825.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 865.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

Conclusions and Outlook 95

References 99

Acknowledgements 111

Curriculum Vitae 113

Summary

During the past twenty years, magnetic resonance spectroscopy (MRS) hasemerged as a valuable method for the investigation of physiological pro-cesses in the human body. It provides useful metabolic information com-plementing the insight delivered by magnetic resonance imaging (MRI) innumerous applications. While the detection and quantification of uncou-pled metabolites is relatively straightforward, J-coupled metabolites stillpresent a major challenge for spectroscopists working in the medical MRfield, because their resonances are usually split into multiplets and canshow a phase and amplitude modulation depending on the sequence tim-ing. In this thesis, problems arising from scalar coupling are discussed andseveral solutions proposed.

Localization techniques such as the double spin-echo method calledpoint-resolved spectroscopy (PRESS), which apply radio-frequency (RF)pulses in combination with field gradients for volume selection, suffer fromchemical shift displacement artifacts, i.e., different volumes are selected forspins with different Larmor frequencies. For weakly coupled spin systems,this gives rise to a considerable signal intensity loss in echo sequences forcertain echo times, an effect called anomalous J-modulation. In the firstcontribution to this thesis, this effect was investigated analytically and ex-perimentally for lactate, a weakly coupled metabolite which plays a pivotalrole in many brain disorders. In vivo and in vitro spectra were acquiredwith human imaging systems from different vendors at 1.5 T and 3 T todemonstrate the signal loss. Furthermore, the signal loss was quantifiedusing a modified spectroscopic imaging (SI) sequence and compared withtheoretical predictions. Finally, several strategies for avoiding or at leastmitigating the signal cancellation are discussed.

The detection and quantification of coupled metabolites is usually im-

3

4 Summary

paired through a strong spectral overlap in one-dimensional (1D) spectra.A viable remedy are two-dimensional (2D) spectroscopy methods, whichenable a better separation and resolution of multiplet resonances, hencefacilitating the extraction of coupling information. A widely used 2D tech-nique, which spreads the coupling information into a second dimension, isJ-resolved spectroscopy. It can easily be implemented by acquiring PRESSspectra with an incremental echo time (TE) variation encoding the indi-rect spectral dimension (JPRESS). In the second part of this thesis, animproved JPRESS method, which tackles problems arising at human MRIscanners, is presented and validated for brain spectroscopy. Through theimplementation of a maximum-echo sampling scheme, the sensitivity is in-creased and the spectral degradation due to imperfect water suppressionis alleviated. Besides, adverse eddy current effects are investigated andpossible correction schemes proposed.

While JPRESS enables the separation of chemical shift and scalar cou-pling information, 2D S-PRESS is introduced as a novel approach for thedetection of strongly coupled spin systems. Instead of varying TE for en-coding all scalar coupling information in the indirect spectral dimension,the echo time is kept constant, but the two partial echo times TE1 andTE2 are varied simultaneously and TE1 is encoded in the indirect dimen-sion. In the resulting 2D spectra, only the resonances of strongly coupledspin systems are spread into the second dimension, leading to more clearlyarranged spectra. The method is particularly useful for the detection ofcitrate representing a strongly coupled AB spin system and was thereforeanalytically optimized for prostate spectroscopy at 3 T. Two-dimensionalS-PRESS spectra were compared with JPRESS spectra in vitro as well asin vivo and the spectral parameters of citrate (coupling constant J andchemical shift difference δ) could be determined in vivo with reasonableaccuracy from 2D S-PRESS spectra acquired in healthy subjects.

In the final part of this thesis, two-dimensional prior-knowledge fitting(ProFit) is used for the quantification of JPRESS spectra acquired at afield strength of 3 T from the human prostate in vivo. Since ProFit is basedon the linear combination of two-dimensional basis spectra, it yields me-tabolite concentration ratios which are independent of sequence and echotime. Furthermore, ProFit turns out to be superior to one-dimensionalfitting methods (e.g. LCModel) for the quantification of coupled metabo-lites like citrate, spermine and myo-inositol. A study carried out on tenhealthy subjects shows that the prostate metabolites creatine, choline, cit-

Summary 5

rate, spermine, myo-inositol and scyllo-inositol can be reliably detected invivo. Among these especially choline and citrate have proven to be usefulmarkers for the detection of prostate cancer.

6 Summary

Zusammenfassung

In den letzten zwanzig Jahren hat sich die Magnetresonanzspektroskopieals wertvolle Methode fur die Untersuchung physiologischer Prozesse immenschlichen Korper etabliert. Sie eignet sich fur die Untersuchung desStoffwechsels und erganzt damit ideal die Magnetresonanzbildgebung, diemit einer Vielzahl von Anwendungen Einblick in die menschliche Ana-tomie und Physiologie gewahrt. Wahrend die Detektion und Quantifizie-rung von ungekoppelten Metaboliten relativ unproblematisch ist, stellenJ-gekoppelte Metaboliten nach wie vor eine grosse Herausforderung furMR-Spektroskopiker dar, weil ihre Resonanzen in sogenannte Multiplettsaufgespalten werden und eine sequenzabhangige Phasen- und Amplituden-modulation durchlaufen. In dieser Dissertation werden Probleme erortert,die sich aus der skalaren Kopplung ergeben, und Losungswege aufgezeigt.

Lokalisierungsverfahren wie die Spin-Echo-Methode PRESS, die Radio-frequenzpulse in Verbindung mit Feldgradienten fur die Volumenselektionverwenden, werden durch sogenannte “Chemical Shift”-Artefakte beein-trachtigt, d.h. verschiedene Volumina werden fur Spins mit verschiedenenLarmor-Frequenzen selektiert. Fur schwach gekoppelte Metaboliten fuhrtdies zu einem erheblichen Signalverlust bei bestimmten Echozeiten, ein Ef-fekt, der in der Literatur als “anomale J-Modulation” bezeichnet wird. Imersten Beitrag der vorliegenden Arbeit wird dieser Effekt analytisch undexperimentell fur Laktat untersucht, einen schwach gekoppelten Metabo-liten, der eine zentrale Rolle bei vielen zerebralen Erkrankungen spielt.MR-Spekren wurden mit medizinischen Scannern verschiedener Herstellerbei Feldstarken von 1.5 T und 3 T in vitro und in vivo akquiriert, um denSignalverlust zu demonstrieren. Daruber hinaus wurde der Signalverlustmittels einer modifizierten Sequenz fur die spektroskopische Bildgebungquantifiziert und mit theoretischen Vorhersagen verglichen. Schliesslich

7

8 Zusammenfassung

werden mehrere Strategien zur Vermeidung oder Reduzierung des Signal-verlusts diskutiert.

Die Detektion und Quantifizierung von gekoppelten Metaboliten wirdnormalerweise durch die starke spektrale Uberlappung in eindimensiona-len Spektren beeintrachtigt. Diesem Problem kann mit zweidimensionalen(2D) Spektroskopieverfahren begegnet werden, die eine bessere Auflosungvon Multiplett-Resonanzen ermoglichen und damit die Extraktion der Kopp-lungsinformation erleichtern. Eine weit verbreitete 2D-Technik, die dieKopplungsinformation in der zweiten spektralen Dimension auflost, ist dieJ-aufgeloste Spektroskopie. Die Methode kann einfach implementiert wer-den, indem PRESS-Spektren mit inkrementeller Echozeit-Variation akqui-riert werden und damit die indirekte spektrale Dimension kodiert wird(JPRESS). Im zweiten Teil dieser Arbeit wird eine verbesserte JPRESS-Methode fur die Spektroskopie an medizinischen MR-Geraten vorgestelltund fur die Hirnspektroskopie validiert. Durch sogenanntes “Maximum-Echo-Sampling” wird die Sensitivitat erhoht und die Beeintrachtigungdes Spektrums durch schlechte Wasserunterdruckung reduziert. Ausser-dem werden Wirbelstrom-Effekte untersucht und Korrekturverfahren vor-geschlagen.

Wahrend JPRESS die Trennung von Chemical-Shift- und Kopplungs-information ermoglicht, wird 2D S-PRESS als neuer Ansatz fur die De-tektion von stark gekoppelten Spinsystemen presentiert. Anstatt mit derVariation der Echozeit (TE) die gesamte Kopplungsinformation in derindirekten Dimension zu kodieren, wird TE konstant gehalten, aber diepartiellen Echozeiten TE1 und TE2 werden gleichzeitig variiert und mitTE1 die indirekte Dimension kodiert. In den resultierenden 2D-Spektrenerfahren nur stark gekoppelte Spinsysteme eine Aufspaltung in der indirek-ten Dimension, was zu einer ubersichtlicheren Anordnung der Resonanzenfuhrt. Die Methode ist vor allem fur die Detektion von Citrat geeignet, dasein stark gekoppeltes AB-Spinsystem besitzt, und wurde deshalb analy-tisch fur Prostata-Spektroskopie bei 3 T optimiert. 2D-S-PRESS-Spektrenwurden mit JPRESS-Spektren in vitro und in vivo verglichen, und diespektralen Parameter von Citrat (Kopplungskonstante J und Chemical-Shift-Abstand δ) konnten mit relativ hoher Genauigkeit aus 2D S-PRESS-Spektren bestimmt werden, die in der Prostata von gesunden Probandenaufgenommen wurden.

Im letzten Teil dieser Doktorarbeit wird 2D Prior-Knowledge-Fitting(ProFit) fur die Quantifizierung von JPRESS-Spektren verwendet, die bei

Zusammenfassung 9

einer Feldstarke von 3 T in der menschlichen Prostata aufgenommen wur-den. Da ProFit auf einer Linearkombination von 2D-Basisspektren ba-siert, liefert der Algorithmus Metabolitenkonzentrationsverhaltnisse, dieunabhangig von der Sequenz und der Echozeit sind. Daruber hinaus zeigtsich ProFit der 1D-Fitting-Methode LCModel fur die Quantifizierung vongekoppelten Metaboliten wie Citrat, Spermin und Myo-Inositol uberlegen.Eine Studie an zehn gesunden Probanden zeigt, dass die Prostatameta-boliten Kreatin, Cholin, Citrat, Spermin, Myo-Inositol und Scyllo-Inositolzuverlassig in vivo quantifiziert werden konnen. Von diesen sind vor allemCholin und Citrat als nutzliche Marker fur Prostatakrebs bekannt.

10 Zusammenfassung

Introduction andObjectives

Since its invention in the early 1970s, magnetic resonance imaging (MRI)has been developed into an indispensable diagnostic tool for medical ex-aminations. MRI exploits the fact that the body is transparent for radio-frequency (RF) waves and gives rise to an excellent soft tissue contrast.Its advantages over other imaging methods such as X-ray based computertomography (CT) and positron emission tomography (PET) are its non-invasiveness and the lack of nocuous ionizing radiation. In addition to themorphological information provided by MRI, magnetic resonance spectros-copy (MRS) allows for the investigation of metabolic processes in the bodyand has therefore proved to be an invaluable method for the examinationof tumors and many neuro-pathological and mitochondrial disorders.

The most abundant isotope in the human body that gives rise to anMR signal is 1H, which can be found in nearly all organic substances. Invivo 1H spectroscopy focuses on the detection of carbon-bound hydrogennuclei, since protons in highly polar bindings to oxygen or nitrogen usuallydissociate very fast compared to the spectral timescale and hence losecoherence. Due to the relatively low sensitivity of human MR systems,only those metabolites that occur in concentrations on the order of 1 mMcan be detected in the human body. In vivo 1H spectra exhibit a relativelysmall chemical shift dispersion and rather broad linewidths, which leadsto a considerable overlap of detectable resonances.

Furthermore, the detection and quantification of most metabolites arehampered by scalar coupling (or J coupling), which gives rise to a linesplitting and a more or less complicated phase and amplitude evolution ofthe observable resonances. In MRS one distinguishes two coupling regimes

11

12 Introduction and Objectives

depending on the coupling strength J and the chemical shift difference δof two coupled spins. When J is small compared to δ, the scalar couplingcan be treated as a perturbation (weak coupling limit) and the J evolutionis periodic. In contrast, when this approximation is not valid, the spinsystem is said to be strongly coupled and its resonance undergoes a morecomplicated phase and amplitude modulation.1

A popular approach for disentangling overcrowded spectra are so-calledspectral editing techniques, which filter out certain J-coupled metabo-lites on the basis of their specific coherence evolution. One distinguishesdifference editing techniques, where two spectra with oppositely phasedresonances of interest are subtracted from each other, and coherence fil-ters, which discriminate coupled and uncoupled resonances via the genera-tion and filtering of certain coherence orders.2,3 However, spectral editingmethods have to be tuned for specific metabolites or coupling networksand are not appropriate for the simultaneous detection and quantificationof different J-coupled metabolites in the same spectral region.

A more versatile method for the detection of J-coupled metabolites istwo-dimensional (2D) spectroscopy.4 Originally used for molecular struc-ture investigations in the NMR field, 2D techniques have gained popularityfor in vivo applications over the past decade. Spreading the spectral in-formation into a second dimension helps to decrease the peak overlap andincrease the distinguishability of multiplet resonances.

Since the signal-to-noise ratio (SNR) and the chemical shift scale withthe external magnetic field (B0), most spectroscopy techniques have gainedsensitivity as well as specificity with the advent of higher field strengths(B0 ≥ 3 T). On the downside, problems such as localisation artifacts andfield inhomogeneities are exacerbated through a higher B0 and often needto be tackled with new concepts.

The overall objective of this thesis is the optimisation of localized spec-troscopy for coupled metabolites in terms of SNR, resolution, artifact re-duction and hence the improvement of peak assignment and quantification.Various problems encountered in the detection of weakly and stronglycoupled spin systems are addressed in this work and specific solutionsproposed. This includes the optimisation of existing and the design ofnew sequences as well as their validation by experimental and analyticalmeans. Furthermore, new postprocessing and quantification strategies arepresented and investigated.

The basic principles of J-coupling are introduced in Chapter 1, along

Introduction and Objectives 13

with a brief description of techniques applied in this thesis, such as PRESSlocalisation, spectroscopic imaging and two dimensional spectroscopy meth-ods.

Chapter 2 deals with an effect called anomalous J-modulation, whichleads to a signal loss for coupled resonances at certain echo times. Thiseffect is quantitatively investigated for the weakly coupled spin system oflactate with single voxel as well as spectroscopic imaging (SI) experiments,carried out at human MR systems from different vendors and at differentfield strengths.

An improved JPRESS method is proposed in Chapter 3 and validatedfor brain spectroscopy at 3T. It focuses on the optimization of SNR andthe mitigation of artifacts in 2D spectroscopy.

In Chapter 4 a novel 2D sequence for the detection of strongly coupledspin systems dubbed 2D S-PRESS is presented. The method is optimisedfor citrate detection in the human prostate and validated in vitro as wellas in vivo.

Finally, Chapter 5 covers the quantification of 2D spectra. The 2Dfitting method ProFit, originally proposed for the quantification of brainspectra, is adapted and optimised for metabolite quantification in the hu-man prostate.

14 Introduction and Objectives

Chapter 1

Basics of MagneticResonance Spectroscopy

1.1 The Chemical Shift

The spectral dispersion in MRS is based on a phenomenon commonlyreferred to as chemical shift, a small change in the Larmor frequenciesof individual nuclei, depending on their chemical environment. When amolecule is placed in a magnetic field, its electronic eigenstates are slightlymodified. This gives rise to small additional magnetic fields at the nuclei,which oppose the externally applied field. The electron density aroundeach nucleus in a molecule varies according to the types of nuclei andbonds in the molecule. The opposing field and therefore the effective fieldat each nucleus will vary with the chemical environment.

The chemical shift is proportional to the exterior magnetic field strength.It is usually defined as the difference between the measured resonance fre-quency ν and the Larmor frequency νref of nuclei of a reference substance,normalized with the resonance frequency of unbound atoms ν0. It is giventhe symbol δ and measured in ppm:

δ =ν − νref

ν0× 106 (1.1)

This definition makes the chemical shift field-independent and thereforefacilitates the comparison of spectra acquired at different field strengths.

15

16 Basics of Magnetic Resonance Spectroscopy

Figure 1.1: Schematic visualization of two coupled protons A andB. Scalar coupling is mediated via the binding electrons between theatoms.

The most widely used spin species observed in MRS experiments is 1H,which has a favorable signal yield due to its high natural abundance andgyro-magnetic ratio. The chemical shift range of in vivo 1H spectra isabout 5 ppm and tetramethylsilane (TMS) is used as a reference substance.Up to twenty metabolites can be detected in vivo with currently usedclinical MR systems, operating at field strengths of 1.5 T or 3 T. Most ofthem give rise to multiple resonances stemming from different 1H nucleiin the molecule.

1.2 Scalar Coupling

Every metabolite gives rise to a characteristic pattern of resonances. How-ever, their detection and quantification is complicated by scalar coupling,a magnetic spin-spin interaction which is mediated via Fermi contacts, i.e.the binding electrons within the molecules (Fig. 1.1).

1.2.1 Classical Picture

Consider a molecule with two coupled spins A and B: Depending on theorientation of spin B with respect to the external magnetic field, the mag-netic field experienced by spin A and therewith also its Larmor frequency

1.2. SCALAR COUPLING 17

Figure 1.2: Energy level diagram for two spins A and B in theuncoupled (left) and weakly coupled (right) case, with four allowedtransitions (transition frequencies: ν′A, ν′′A, ν′B, ν′′B. Under theinfluence of weak scalar coupling, the energy levels are shifted by/Delta = /pmJAB/4 (with JAB being the coupling constant).

are slightly different. This gives rise to a small shift of the energy levelsand hence also changes the frequencies of the transitions, which are ob-served in an MRS experiment (Fig. 1.2). While in the uncoupled case thefrequencies of the two allowed transitions for spin A as well as spin B areequal (ν′A = ν′′A, ν′B = ν′′B), giving rise to two singlets, the two resonancesare split into doublets in the coupled case (Fig. 1.3). Coupling networkswith more than two coupled spins cause a splitting into triplets, quartetsand so on. In addition to the line splitting, scalar coupling gives rise toa phase evolution of the coupled resonances as a function of echo time(TE). This is illustrated in Fig. 1.4: For echo times of TE = 1/(2J) andTE = 3/(2J) the two peaks of a doublet are in anti-phase with respectto each other, while at TE = 1/J the resonance appears as an invertedin-phase doublet. After an echo time of TE = 2/J the coupled resonancehas experienced a phase evolution of 2π and appears as a positive in-phasedoublet again.

Basic rules for spectra of J-coupled nuclei:


Figure 1.3: Resulting spectra in the uncoupled (left) and coupled(right) case. Instead of a singlet a the resonance frequency νA ofthe uncoupled proton, one observes a doublet at resonances ν′A =νA + JAB/2 and ν′′A = νA − JAB/2

• J does not depend on the main field strength.

• The coupling strength decreases with increasing number of bondsrelating coupled spins.

• Coupling and therefore splitting is mutual: JAB = JBA.

• Spin-spin coupling between magnetically equivalent nuclei producesno observable splitting.

• A nucleus coupled to nI equivalent nuclei of spin I gives rise to2nI + 1 lines.

• Coupling to different spin groups is additive.

• The line intensities in a weakly coupled resonance are distributedaccording to a binomial pattern.

• Strong coupling gives rise to more complicated spectra.

Spin echoes in MRS sequences are usually created with refocusingpulses. Those invert the spin dispersion from a previous evolution pe-riod and hence cancel the Zeeman interaction, when placed in the middlebetween excitation and acquisition. In contrast to that, a non-selective

1.2. SCALAR COUPLING 19

Figure 1.4: Signal modulation in dependence of echo time TE.During the evolution period the transverse magnetisation is con-verted from in-phase to anti-phase magnetisation and vice versa.


180◦ pulse does not refocus the effects of J-coupling, which results in aphase modulation in a spin echo experiment. However, when instead aspin-selective refocusing pulse is applied, i.e., only one of the two coupledspins is inverted, the J-evolution is refocused and the resonance appears asa positive in-phase doublet regardless of echo time. This phase modulationis exploited in spectral editing techniques.2

1.2.2 Quantum-Mechanical Picture

In liquid phase NMR, the unperturbed nuclear spin Hamiltonian consistsof two parts, a Zeeman term HZ and a coupling term HJ :

H = HZ + HJ , (1.2)

with

HZ =N∑

k=1

ω0kIkz, HJ = 2π

N∑

k=1,l<k

−→IkJkl

−→Il . (1.3)

−→Ik denotes the angular momentum spin operator and ω0k the Larmor fre-quency of spin k, Jkl denotes the coupling strength between spin k andspin l.

One distinguishes two different coupling regimes depending on the cou-pling strength J and the difference in Larmor frequencies ∆ω of two cou-pled spins: In the weak coupling limit (2πJ ¿ ∆ω) the non-secular partof HJ can be neglected:

HwcJ = 2π

N∑

k=1,l<k

JklIkzIlz. (1.4)

In this case, coupling and Zeeman Hamiltonian commute: [HwcJ ,HZ ] = 0.

As a consequence, chemical shift and coupling evolution are independentof each other and can be calculated separately, using the product operatorformalism.5 If the secular approximation (2πJ ¿ ∆ω) is not valid, as formost homonuclear spin systems at typical field strengths of 1.5 or 3 T,the full coupling Hamiltonian has to be considered and the spin systemis said to be strongly coupled. Since Hsc

J and HZ do not commute, theproduct operator formalism cannot be applied any more and the analyticalcalculation of the spin evolution becomes more complicated.

1.3. SPATIAL LOCALISATION 21

1.3 Spatial Localisation

For spectroscopy application in the human body, where specific organs andtheir sub-structures are to be investigated, it is desirable to obtain spatiallyresolved spectral information. Especially for 1H spectroscopy, localisationis usually based on RF pulses in combination with linear field gradients. Ingeneral, RF pulses have a fixed finite bandwidth of typically a few hundredto a few thousand Hz. If a linear field gradient is applied to the sample, theLarmor frequencies of the spins show a linear dependence on the spatialposition. When an RF pulse is irradiated at the same time, it affectsonly those spins, whose Larmor frequency lies inside the frequency bandof the pulse. Applying this method consecutively with gradients alongthree orthogonal directions, a volume of interest (VOI) can be defined,from which the MR signal is acquired.

The most widely used technique based on this localisation approach ispoint-resolved spectroscopy (PRESS), which uses one 90◦ excitation pulseand two 180◦ refocusing pulses for localisation and echo formation at thesame time (Fig. 1.5). The acquisition of the echo can be started eitherimmediately after the second refocusing pulse (maximum echo) or only atthe echo top (half echo). The excitation pulse and the two refocusing pulsesbecome slice-selective by field gradients in the three spatial directions. Thefinal echo signal therefore originates only from the volume element definedby the intersection of the three slice planes.

It is often desirable to obtain spectroscopic information with a spatialresolution. This can be achieved with spectroscopic imaging (SI), whichis typically applied along two spatial dimensions. Other than in Fourierimaging where the signal is phase-encoded in one k-space direction andfrequency-encoded in the other during acquisition, both spatial dimensionswithin an SI slice, which is typically selected with the excitation pulse, arephase-encoded (Fig. 1.6). This typically amounts to a scan time of severalminutes, because in conventional SI experiments one excitation is requiredfor each point in k-space that is to be read out. However, SI sequencescan be accelerated by acquiring several echo signals per excitation. Thismethod is called turbo spectroscopic imaging (TSI).

The localised spectra are reconstructed through a Fourier transforma-tion along the temporal and the two spatial dimensions. The spatial dis-tribution of a certain metabolite can then be determined by integratingall spectra within a specified frequency interval around the resonance fre-


Figure 1.5: Schematic representation of a PRESS sequence. Vol-ume localisation is achieved with one excitation and two refocusingpulses in combination with selection gradients in three different di-rections.

1.3. SPATIAL LOCALISATION 23

Figure 1.6: K-space sampling in 2D SI (b) versus Fourier imag-ing (a). In conventional Fourier Imaging one spatial dimension isfrequency-encoded during the acquisition (x) and the other phase-encoded (y). In 2D SI both spatial dimensions have to be phase-encoded, since the spectral information is read during the acquisi-tion.


Figure 1.7: Creation of metabolite maps from SI data through peakintegration.

1.4. TWO-DIMENSIONAL SPECTROSCOPY 25

quency (Fig. 1.7). With a Fourier interpolation of the calculated matrixof integration values, smoothed metabolite maps can be obtained and su-perimposed with an anatomical image for the clinical interpretation of theresults.

For signal-to-noise considerations, the voxel size has to be significantlylarger than in imaging sequences, typical minimal values are 1 cm3 for 1Hspectroscopy. It is important to suppress signal from fatty tissue, whichoften surrounds the region of interest (e.g. in the brain). Therefore, inbasic applications, the SI method is combined with PRESS localisationand outer volume suppression (OVS).

1.4 Two-Dimensional Spectroscopy

In one-dimensional (1D) MRS the signal is recorded as a function of onetime variable and then Fourier transformed to give a spectrum which isa function of one frequency variable. In two-dimensional (2D) MRS thesignal is recorded as a function of two time variables, t1 and t2. Theresulting data is Fourier-transformed along both dimensions to yield aspectrum which is a function of two frequency variables f1 and f2. Thegeneral scheme for 2D spectroscopy is illustrated in Fig. 1.8.

In the first period, called the preparation interval, the spins are excitedby one or more pulses. The resulting transverse magnetization, which canconsist of single and multiple quantum coherences, is allowed to evolvefor the first time period t1. Then another period follows, called the mix-ing interval, which can contain further pulse/gradient elements. After themixing period the signal is recorded as a function of the second time vari-able t2. The sequence components applied in the preparation and mixingperiods (RF pulses, gradients, etc.) determine the information found inthe spectrum and its spectral representation. A 2D experiment consistsof a row of measurements with incrementally changed t1, which are storedseparately. Thus recording a 2D data set involves repeating a pulse se-

Figure 1.8: General scheme of a 2D pulse sequence.


quence for increasing values of t1 and acquiring a signal as a function oft2 for each value of t1.

In the NMR field, two-dimensional spectroscopy methods play a pivotalrole for molecular structure investigation, since they help to identify cou-pling networks and measure coupling constants as well as intra-moleculardistances. In medical applications, which usually suffer from low SNR andlimited spectral resolution, 2D sequences are used to enhance the distin-guishability of resonances by spreading them into a second dimension. Oneof the most popular and robust 2D method is JPRESS, a 2D J-resolvedtechnique, which is based on PRESS localisation. For a JPRESS exper-iment the echo time is varied and encoded in the indirect dimension t1.The indirect frequency domain f1 contains only coupling, but no chemicalshift information as the direct dimension f2.

Chapter 2

Pitfalls in LactateMeasurements at 3 T

(published in: American Journal of Neuroradiology6)

2.1 Introduction

Lactate is a metabolite that plays a pivotal role in many brain pathologiessuch as tumors, stroke, cerebral ischemia, hypoxia and several mitochon-drial disorders.7 The lactate concentration in healthy brain under normalconditions is about 1 mM. However, when oxygen availability is low dueto metabolic stress, glucose molecules are no longer oxidized completelyand pyruvate is produced. The pyruvate is converted to lactate, whichcan rise to concentrations above 10 mM. Lactate is therefore an importantmarker of anaerobic glycolysis taking place in the previously mentionedbrain disorders.

The lactate molecule has two weakly coupled resonances in 1H mag-netic resonance spectroscopy (MRS): a doublet (split by coupling to themethine (CH) proton) at 1.33 ppm arising from three magnetically equiva-lent methyl (CH3) protons and a quartet (split by coupling to the protonsof the methyl group) at 4.11 ppm, arising from the methine proton, whichis usually not visible in vivo. The scalar coupling gives rise to a phase evo-lution of the methyl doublet, which depends on the echo time (TE). ForTE = 144 ms, the resonance shows a phase of 180◦ leading to a negative

27

28 Pitfalls in Lactate Measurements at 3 T

in-phase doublet, whereas an echo time of 288 ms gives rise to a positivein-phase doublet. Since only in-phase resonances can be quantified, echotimes of 144 ms and 288 ms are preferable for lactate detection and assign-ment. The coupling evolution can also be exploited for difference editingtechniques increasing the sensitivity of lactate detection.2

Localization techniques such as the double spin echo method PRESS,which is the standard localization technique used on clinical MRI systems,suffer from chemical shift displacement artifacts. Importantly, this can giverise to signal misregistration for almost all metabolites, as only the signalfrom one specific frequency, usually the NAA frequency, originates fromthe selected volume of interest. Signal from protons with different chemicalshifts, e.g. from other metabolites, stem from spatially shifted volumes.Furthermore, for weakly coupled resonances, there is an additional signalcancelation due to anomalous J-modulation.8,9 This additional artifactand its effect on the interpretation of lactate levels in clinical spectra arediscussed in this work. As the chemical shift displacement roughly scaleswith the square of the field strength (due in part to reduced radio frequency(RF) pulse bandwidths, as well as increased chemical shift frequency sep-aration), a severe underestimation of lactate occurs at 3 T, when PRESSlocalization is used with an echo time of 144 ms. A detailed explanationof the origin of anomalous J-modulation can be found in the appendix.

This work shows in vitro and in vivo examples of signal cancellationfor several MRI systems. A strategy is presented to quantify the lactatesignal loss and thus validate the underlying theory. Finally, suggestionsfor parameter choices on clinical systems are given to avoid or at leastdiminish the problem of lactate underestimation at 3 T.

2.2 Description of the Technique and Results

2.2.1 Patient Measurements

Two patients with high-grade gliomas underwent MRS both with echotimes of 144 and 288 ms, performed on Philips Intera whole-body systems(Philips Medical Systems, Best, The Netherlands) by using a send-receivehead coil. One patient was measured at a field strength of 1.5 T, while theother was measured at 3 T.

Another patient with mitochondrial myopathy, encephalopathy, lacticacidosis, and stroke (MELAS) was examined on 1.5 T and 3 T systems from

2.2. DESCRIPTION OF THE TECHNIQUE AND RESULTS 29

a second vendor (GE Healthcare, Milwaukee, WI). Two-dimensional multi-voxel spectra were acquired at both field strengths consecutively the sameday, repeated with both TE = 144 ms and TE = 288 ms. Standard PRESSlocalization was used with the region of interest centered over an acuteoccipital lesion. On this system very selective saturation (VSS) bands10

are placed around the volume of interest by default. Some of the datapresented here were acquired on a clinical basis, but all patients gave theirwritten informed consent prior to participating in research experiments.

In the high-grade tumor in Fig. 2.1 the spectra acquired at 1.5 T bothshow a prominent lactate doublet, inverted for TE = 144 ms and in phasewith NAA for TE = 288 ms. The lactate peak, clearly visible as inverteddoublet at TE = 144 ms, is of greater area and amplitude at TE = 144 msthan at TE = 288 ms. This signal change reflects the known T2 relatedsignal attenuation with increasing echo times.

Comparable experiments were carried out on a 3 T system, scanninganother patient with a glioma (grade III). While for TE = 288 ms alarge lactate peak is observable at 1.33 ppm, the lactate resonance hascompletely vanished for TE = 144 ms due to anomalous J-modulation(Fig. 2.2).

The third patient example (Fig. 2.3), carried out on different systems,shows lactate peaks of similar relative amplitude with TE = 144 ms andTE = 288 ms at 1.5 T. The lactate signal loss, which is expected with TE= 288 ms compared to TE = 144 ms due to T2 relaxation, is approximatelycompensated by the signal loss due to anomalous J-modulation with TE =144 ms. At 3 T, however, anomalous J-modulation gives rise to a completedisappearance of the lactate peak.

2.2.2 Phantom Measurements

A standard brain metabolite phantom containing 5 mM lactate was mea-sured on three 3T MRI scanners from three different vendors (GE Health-care, Philips Medical Systems and Siemens Medical Solutions). Single-voxel MRS (volume of interest (VOI) size = 2 × 2 × 2 cm3, PRESSlocalization) was acquired from the same volume once with TE = 144 msand once with TE = 288 ms. The lactate peak seen at 1.33 ppm for TE= 288 ms would be expected to be smaller than the inverted doublet forTE = 144 ms due to T2 relaxation. However, at 3 T, spectra from allthree MR systems show significantly larger lactate resonances at TE =288 ms than at TE = 144 ms. For one system the lactate peak has even


Cho

NAA

Lac

Cho

NAA

Lac

a)

b) c)

TE = 144 ms TE =288 ms

Figure 2.1: Single-voxel spectra acquired at 1.5 T from the brain ofa patient (a) with a high-grade glioma by using PRESS localization:b) TE = 144 ms, c) TE = 288 ms. Cho indicates choline; Lac,lactate; NAA, N-acetylaspartate.


Cho

Cr

NAA

Lac

Cho

Cr

NAA

Lac

a)

b) c)

TE = 144 ms TE = 288 ms

Figure 2.2: Single-voxel spectra acquired at 3 T from the brain ofa patient (a) with a grade III glioma by using PRESS localization:b) TE = 144 ms, c) TE = 288 ms. Cr indicates creatine.


Figure 2.3: Multi-voxel spectra acquired one hour apart at 1.5 T and3 T from the same region in the brain of a patient with MELAS,by using standard PRESS localization with TE = 144 ms and TE= 288 ms. An inverted lactate doublet is clearly visible at 1.5 T,but not at 3 T (arrows). Upright lactate peaks at TE = 288 ms areseen equally well at both field strengths (arrows).


Figure 2.4: Single-voxel proton spectra acquired from a standardbrain metabolite phantom containing 5 mM lactate. The measure-ments were performed on three 3 T MR scanners from three dif-ferent vendors with TE = 288 ms (upper row) and TE = 144 ms(lower row). RF pulse bandwidths for the selective refocusing pulsesvary between vendors in the range 874-2300 Hz.

disappeared completely. This demonstrates that a significant amount ofthe lactate signal is lost (not visible) at TE = 144 ms (Fig. 2.4).

To quantify the signal loss due to anomalous J-modulation on the basisof the phantom spectra acquired for TE = 144 ms and TE = 288 ms, onewould have to determine the T2 relaxation constant of lactate first. Sincethe relaxation constant of lactate for the used brain metabolite phantomwas unknown and difficult to determine with high precision, a differentapproach was chosen: Spectroscopic Imaging (SI) offers the possibility todispense with spatially selective refocusing pulses, since the SI slice canbe selected only by the excitation pulse and the spatial encoding within


the slice is achieved with phase encoding only. However, in a standardSI measurement protocol without PRESS localization, both the excitationpulse and the refocusing pulse, which is needed for the echo formation,are slice-selective on some scanners, leading again to some signal loss. Toestimate the amount of signal loss due to anomalous J-modulation, SI datasets were acquired once with (“standard” SI sequence) and once withoutthe slice selection gradient during the refocusing pulse. Theory predictsthat the overall signal loss at TE = 144 ms is the same for a standard SIsequence as for a single-voxel protocol using PRESS (Appendix). It canalso be calculated that no signal loss should occur for lactate in a standardSI measurement with TE = 288 ms. Therefore, to validate the theoreticalpredictions experimentally, SI data sets were acquired for these two echotimes. Postprocessing of the spectra included exponential filtering of thetime domain signal, cosine filtering in k-space and B0 correction.

Fig. 2.5 shows the results of the SI measurements without PRESS. Forecho times of 144 ms and 288 ms one representative SI voxel is shown(acquired once with (Figs 2.5a, 2.5c) and once without a refocusing pulsegradient (Figs 2.5b, 2.5d), respectively). Integrating the modulus spec-tra and comparing the results with and without anomalous J-modulationyielded a signal loss of 72.2 % for TE = 144 ms, whereas the signal intensitywas approximately the same for TE = 288 ms. Theoretical calculationstaking into account the chemical shift difference of the coupled nuclei andthe bandwidth of the used refocusing pulses yield a relative signal loss of81.2 % for TE = 144 ms, whereas for TE = 288 ms no signal loss dueto anomalous J-modulation is predicted (Appendix). Thus the measuredsignal loss is consistent with the theoretically expected value. It should benoted that, in contrast to SI without PRESS, theory also predicts a smallsignal loss of 6.8 % at TE = 288 ms when using PRESS localization forsingle-voxel experiments. But this loss only needs to be taken into accountwhen high precision quantification is desired.

2.3 Discussion

Detection of lactate using MRS plays an important clinical role in the as-sessment of a number of brain abnormalities, including tumor, stroke andmitochondrial disorders. The presence of lactate in the context of a tumorcan be considered diagnostic for glioblastoma multiforme. Lactate is alsoelevated as a consequence of mitochondrial abnormalities in neurodegen-

2.3. DISCUSSION 35

NAA

Lac

NAA

Lac

NAA

Lac

NAA

Lac

a) b)

c) d)

Figure 2.5: Spectra from an SI dataset acquired at 3 T from a phan-tom containing 10 mM NAA and 20 mM lactate without PRESSlocalization: a) TE = 144 ms, with refocusing pulse gradient, b)TE = 144 ms, without refocusing pulse gradient, c) TE = 288 ms,with refocusing pulse gradient, d) TE = 288 ms, without refocusingpulse gradient.


erative disorders, such as Huntington’s disease. Since for very short echotimes there are often residual fat peaks visible in the spectral region ofthe lactate doublet, one has to resort to echo times near multiples of 1/J(J being the coupling constant) for proper lactate detection and quantifi-cation. As an inverted doublet can be discriminated more easily againstother resonances like lipids, PRESS with TE = 144 ms is often consideredthe most appropriate method for unambiguous lactate detection at 1.5 T.Although the effect of “anomalous J-modulation” and the potential signalloss for lactate at TE = 144 ms has been discussed in the literature,8,9

little attention has been given to this phenomenon in clinical routine sofar. Neglecting this effect can potentially lead to a severe misinterpretationin clinical diagnosis as shown in our examples. The extent of the signalloss due to anomalous J-modulation can vary considerably depending onthe field strength, the used coil and the sequence parameters. A prac-tical recommendation for clinical MRS at 3 T is to perform a phantomstudy at TE = 144 ms and TE = 288 ms; if the lactate signal at TE =144 ms is less than that seen at TE = 288 ms, it is recommended not touse TE = 144 ms, but rather only TE = 288 ms in clinical examinations,although in general the sensitivity decreases with longer echo times due toT2 relaxation. However it should be noted that the quantitative influenceof anomalous J-modulation can be different in specific in vivo examples,where inhomogeneous metabolite distributions can either aggravate or at-tenuate the effect compared to in vitro experiments.

Strategies to prevent or alleviate the signal loss due to anomalous J-modulation are discussed elsewhere in detail,8,9, 11 but usually requirechanges in the scanner software. An approach implemented on severalclinical scanners is to saturate the region of spin-selective refocusing withouter volume suppression (OVS) pulses prior to excitation. For example,on some scanners, quadratic phase suppression pulses have been imple-mented for this purpose,10 but they suppress the effect only partially, sincethey are not specifically tuned for lactate detection as can be seen bothfrom in vivo and phantom examples in this work. However, at higher fieldstrengths than 3 T, which have now become available for human studiesas well, the chemical shift displacement will be so pronounced that thiswork-around will be less effective.

Another approach for high-field systems is using pulses with muchlarger bandwidths such as adiabatic pulses to alleviate the chemical shiftdisplacement. However, since ordinary adiabatic pulses are usually not

2.4. APPENDIX 37

spatially selective, they cannot be used in PRESS sequences. Localizationby adiabatic selective refocusing (LASER)12 or single-shot adiabatic lo-calized volume excitation (SADLOVE)13 applies pairs of large-bandwidthadiabatic full passage pulses for volume selection and echo formation atthe same time and will probably be the method of choice for localizationin high-field spectroscopy applications in the future. A further popularlocalization technique is STimulated Echo Acquisition Mode (STEAM),which uses three 90◦ pulses giving rise to a stimulated echo. Since 90◦

pulses have much larger bandwidths than 180◦ pulses, the chemical shiftdisplacement is far less severe. However, for STEAM the signal intensityof coupled resonances not only varies with TE, but also shows a strongmodulation governed by the mixing time between the second and thirdpulse.7 Furthermore, SI offers the possibility to dispense with spatiallyselective refocusing pulses and, therefore, prevent the signal cancellation.However, SI protocols without PRESS localization are usually not imple-mented on purely clinical scanners and very good OVS is required for thisapproach to prevent the spectra from being impaired by subcutaneous fatsignal.

2.4 Appendix

The signal cancellation by anomalous J-modulation for coupled resonancesarises from the chemical shift displacement artifact. The relative voxeldisplacement for two protons with chemical shifts δ1 and δ2 equals the ratioof the chemical shift difference (∆ωCS = δ1−δ2) and the bandwidth of theRF pulse (∆ωRF ) used for volume selection. For the signal of the methyl(CH3) resonance of lactate at 1.33 ppm, the signal loss can be understoodas follows: Due to this chemical shift displacement, the volume selectedby one single refocusing pulse decomposes into a region where both themethyl and the methine (CH) protons of the lactate molecule are affectedby the pulse (“non-selective” pulse) and a region where only the methylspin is inverted (“selective” pulse) and therefore the coupling evolutionis refocused. This leads to a superposition of signal with different phasesgiving rise to signal cancellation in the spectrum. The volume selected by aPRESS sequence, which uses two refocusing pulses, consists of four partialvolumes with different phase evolutions, depending on whether none, oneor both of the two refocusing pulses are spin-selective (Fig. 2.6). In aPRESS sequence the signal loss is also determined by the sequence timing,


Figure 2.6: Partial volumes and their coupling evolution for asingle-voxel PRESS experiment: The 90◦ excitation pulse is ap-plied with a gradient in the z direction, whereas the two refocusingpulses are applied with gradients in the x and y direction, respec-tively. The size of the partial volumes is determined by the chemicalshift displacement. The signal phase of the magnetization is deter-mined by TE, the scalar coupling constant (J) of lactate and thetime interval (t1) between the excitation pulse and the first refocus-ing pulse.

since the spins are inverted twice by the two spatially selective refocusingpulses.14 By adding the signal terms for the four partial volumes, theresulting signal loss can be calculated. This effect termed “anomalous J-modulation”9 is particularly pronounced for TE = 144 ms, when the twosuperimposed signals from the partial volumes V1 and V4 have a phasedifference of 180◦ and therefore the cancellation is most effective, whereasfor TE = 288 ms almost no cancellation occurs (Fig. 2.6).

According to theory, the relative signal loss for lactate due to anoma-lous J-modulation at TE = 144 ms amounts to

∆S

S= 2

∆ωCS

∆ωRF, (2.1)

2.4. APPENDIX 39

where ∆ωCS is the chemical shift difference of the coupled nuclei and∆ωRF is the bandwidth of the refocusing pulse. At 3 T the chemical shiftdifference between the methyl and the methine proton is 355 Hz, and therefocusing pulse used in this example had a bandwidth of 874 Hz. Thispredicts a relative signal loss of 81.2 % due to anomalous J-modulationfor TE = 144 ms. The theoretical signal loss for single-voxel PRESS mea-surements at TE = 288 ms depends on the position of the two refocusingpulses within the sequence. Usually the PRESS sequence is rendered asasymmetric as possible with the first refocusing pulse being irradiated assoon as possible (at time t1) after the excitation pulse. For the standardPRESS sequence implemented on a 3T Philips Intera Scanner theory pre-dicts the following signal loss due to anomalous J-modulation at TE =288 ms:

∆S

S= 2

∆ωCS

∆ωRF·(

1− ∆ωCS

∆ωRF

)· (1− cos (π · J · 2t1)) = 6.8%. (2.2)

For an SI sequence without PRESS localization, but with a slice-selectiverefocusing pulse for the echo formation, there are only two interferingpartial volumes (corresponding to V1 and V4 in Fig. 2.6). For TE = 144 ms,this leads to the same relative signal loss as for a single-voxel PRESSsequence, whereas no signal loss at all occurs for TE = 288 ms.

To understand the basic behavior of the lactate resonances in a PRESSsequence, it is sufficient to approximate the used pulses as ideal hard pulses.However, pulse imperfections give rise to a far more complicated evolutionbehavior, which was investigated analytically, simulated numerically anddiscussed in detail in several publications.15–19 Since the chemical shift isproportional to B0 and, due to B1 limitations, the pulse bandwidths are ap-proximately inversely proportional to B0, the chemical shift displacementand therewith the lactate signal loss due to anomalous J-modulation atTE = 144 ms roughly scale with the square of the field strength.

Chapter 3

ImprovedTwo-DimensionalJ-Resolved Spectroscopy

(published in: NMR in Biomedicine20)

3.1 Introduction

The main objective of in vivo proton MRS is the determination of in-dividual metabolite concentrations. This can be a difficult task owingto generally complex and overcrowded spectra. A powerful and versatileapproach to increase specificity is multi-dimensional spectroscopy, whereindirect dimensions are encoded by varying the lengths of evolution pe-riods. These sequences greatly dominate the in vitro NMR field, yield-ing a wealth of information such as connectivities [COSY21] or moleculardistances [NOESY22]. However, the application of multi-dimensional spec-troscopy in vivo is unpopular for several reasons. Time, hardware andother constraints only permit the application of the most basic multi-dimensional sequences such as J-resolved spectroscopy23,24 or COSY.25

Additionally, post-processing techniques are not as advanced as regular1D spectroscopy, that is, no appropriate fitting procedures exist for ex-tracting the maximum information from the 2D spectra.

41

42 Improved Two-Dimensional J-Resolved Spectroscopy

Two-dimensional J-resolved spectroscopy26 is one of the simplest 2Dsequences and well suited for in vivo applications. It consists of a series ofspin-echo experiments with different echo times encoding the indirect t1dimension. The equivalent in vivo MRS sequence is dubbed JPRESS,24

since it employs point-resolved spectroscopy (PRESS)27 for volume locali-sation. JPRESS provides an additional encoding of the J coupling, henceimproving specificity in detecting J-coupled metabolites and alleviating amajor limitation of 1D MRS, the spectral overlap of resonances.

However, JPRESS has been applied for in vivo MRS in only a few stud-ies. A recent review of the application to small animals is given by Mericet al.28 JPRESS has been applied to the human brain,24,29–36 muscle,23,37

prostate38,39 and breast.40 The method is highly robust to large water andlipid signals (e.g. in the breast), hence permitting spectroscopy withoutwater suppression.31,38 Interesting variants of JPRESS include a weightedaccumulation scheme for filtering the indirect dimension already during ac-quisition,41 the combination with chemical shift imaging and spiral sam-pling42 and the constant-time version, termed CT-PRESS43 or CSSF,44

where the chemical shift (CS) is encoded in the indirect dimension. Thesensitivity and robustness of traditional J-resolved spectroscopy can beimproved by a maximum-echo sampling scheme,45 which was applied inthe human brain at 1.5 T.41 In maximum-echo sampling, the acquisitionstarts immediately after the final crusher gradient of the last refocusingpulse. This leads to both an increase in sensitivity and a tilt of the peaktails in the spectra. The experiment with the maximum-echo samplingscheme was further modified to acquire multiple echoes, all sampled aslong as possible, in rat brain at 7 T.46

JPRESS spectra are typically quantified in two different ways. Thefirst one is the spectral integration of metabolite peaks directly in two di-mensions, which is the most common method in in vitro NMR. As the line-shapes of 2D peaks in the real spectrum are phase-twisted, the magnitudespectrum is usually preferred for integration. This results in a consider-able loss of information and in an increased overlap due to broad dispersiontails of magnitude peaks.4 The other possibility is the application of 1DMRS techniques to cross-sections through the 2D spectra. However, thisexcludes information from the spectral analysis. In the cross-section att1 = 0 Hz, also called TE-averaged PRESS,35 most of the signal fromJ-coupled metabolites is lost. When taking another cross-section at thecorresponding J frequency,32,33,36 one loses the signal from other peaks

3.2. THEORY AND METHODS 43

of the same metabolite. This can be seen, for instance, in γ-aminobutyricacid (GABA), which has 11 peaks (i.e. two triplets and one quintet). Bytaking the cross-section, only one peak remains for quantification, hencelosing a large amount of information. Furthermore, the overlap with thepredominant singlets can still be considerable, as in case of GABA withcreatine at 3 ppm, leading to poor experimental reproducibility.47

In this work, several reconstruction issues are discussed, along withstrategies for post-processing. The specific benefits of the maximum-echosampling scheme for JPRESS45 are elucidated, which are mainly an in-crease in sensitivity and a decrease in spectral overlap. The sensitivity ofJPRESS is analytically compared with that of PRESS. The qualitativebehaviour of eddy currents occurring in this type of experiment is assessedand an efficient 2D eddy current correction scheme based on the 1D phase-deconvolution method48 is proposed. Furthermore, it is possible to devisea realistic model for two-dimensional fitting49 with the knowledge of thequalitative behaviour of eddy currents.

3.2 Theory and Methods

Localised two-dimensional J-resolved spectroscopy (JPRESS) is a simplespin-echo experiment with different echo times encoding the J couplingin the indirect t1 dimension. The echo top is used as reference point forthe reconstruction along t1 so that no chemical shift (CS) evolution ispresent in that dimension. The J evolution is not influenced by the 180◦

pulses and is hence resolved along t1. The acquisition is encoded alongthe direct t2 dimension, which contains both CS and J evolution. Aftera Fourier transformation of the data in two dimensions, a spectrum isobtained with its resonances aligned on the horizontal (0 Hz) axis. J-coupled spin systems are split up into multiplets tilted by 45◦, as depictedfor GABA in Fig. 3.1. The preceding considerations hold strictly onlyin the weak coupling limit. Strongly coupled spin systems give rise toadditional resonances at the mean chemical shift frequency.4,24

The JPRESS implementation used for this work is derived from theregular, asymmetric PRESS sequence27 (Fig. 3.2). The first echo time(TE1) is always chosen as short as possible and only the second refocusingpulse is shifted for encoding t1. The sampling of the echo signals startsimmediately after the final crusher gradient of the last 180◦ pulse,45 hence-forth called maximum-echo sampling. This acquisition scheme has several


f 1 [Hz]

f2 [ppm]

1.61.822.22.42.62.833.2

−20

−10

0

10

20−5

0

5

f 1 [Hz]

f2 [ppm]

1.61.822.22.42.62.833.2

−20

−10

0

10

20−5

0

5

Figure 3.1: In vitro spectrum of GABA plotted in magnitude mode(top) and phased complex mode (bottom). Two triplets are visibleat 3.01 and 2.28 ppm and one quintet at 1.89 ppm. The CS evo-lution is only effective in the direct f2 dimension. One multipletis therefore located at one CS frequency along f2. J coupling iseffective in both f1 and f2, hence the resonances are split up intomultiplets tilted by 45◦. The minimum echo time was 31 ms, hencethe multiplets already exhibit some anti-phase evolution (bottom).Almost all structure is lost in the magnitude spectrum (top) due tothe broad dispersion tails. (100 mM GABA; 2 Hz Gaussian linebroadening in both t1 and t2)


Figure 3.2: Principle functioning of the applied JPRESS sequence.The acquisition starts at TAS directly after the final crusher gra-dient. The indirect dimension is given by t1 + TEmin = TE =TE1 + TE2, with a varying second echo time TE2 and a constantfirst echo time TE1. TEmin is also the minimum echo time of thecorresponding PRESS sequence.

advantages over the traditional half-echo sampling, where the acquisitionstarts at the echo top. Highly beneficial, as described later, is the re-sulting tilt of the peak tails. Furthermore, the sensitivity is increased byacquiring more signal. This gain in sensitivity is calculated analytically inthe Appendix by integrating over the damping curves in the time domain,which corresponds to the peak height in the frequency domain. The noiseis the same for experiments of the same duration and repetition time. Thesensitivity can be compared by calculating the ratio of peak heights. Thisis analytically derived in the Appendix for shortest echo-time PRESS andJPRESS with two different sampling schemes (i.e. maximum and halfecho).

The integral of a resonance line is proportional to the first point ofthe FID. This applies also to the two-dimensional case; the point in theecho signal responsible for the integral under the spectrum is the centre ofthe reference, as seen in the middle of Fig. 3.3(a) and (b). This point isnot affected by the sampling scheme and therefore the areas underneaththe resonances remain constant. The reconstruction of JPRESS spectrafrom the maximum-echo data is straightforward. Its reference points arethe minimum echo time TEmin and the echo top of the t1 and t2 dimen-sion, respectively. TEmin is identical with the minimum echo time of the


Figure 3.3: Traditional half-echo sampling (a and c) in comparisonwith maximum-echo sampling (b and d). The acquisition of the echostarts either at the echo top (a) or as soon as possible (b). The tiltof the truncation line for the acquisition begin (b) translates directlyinto a tilt of the peak tails away from f2-axis (d). In the traditionalsampling, both tails are rectangular to each other. The referencepoint of the 2D Fourier transformation is shifted to the middle inboth domains. The echo signals are additionally zero-padded totwice the number of rows in t1.


corresponding half echo PRESS sequence and governed mainly by the du-ration of the RF pulses and gradients. For every time increment ∆TE, themaximum-echo sampling starts the acquisition ∆TE/2 earlier with respectto the echo top. Therefore, the different rows have to be time-shifted tothe same reference point (i.e. the echo top) in the time domain. This canbe performed flexibly in the frequency domain by multiplication with thelinear phase

τ(t1, f2) =12t1f2 (3.1)

where t1 = TE − TEmin is the time in the indirect dimension (Fig. 3.3)and f2 is the frequency in the direct dimension. For this time shift, thedata is Fourier transformed only along the direct t2 dimension. A JPRESSspectrum is obtained after Fourier-transforming the time-shifted signal alsoalong the indirect t1 dimension.

In the shifted JPRESS time domain data, the echo truncation lineis therefore tilted by arctan(1/2) ≈ 26.6◦ against the vertical t1-axis(Fig. 3.3). This truncation line directly translates into the frequency do-main, where the peak tails along f2 are tilted by the same angle againstthe horizontal f2-axis (Fig. 3.3). The angle is independent of the samplingfrequencies in f1 and f2. However, the f1 dimension should be oversam-pled sufficiently in order to prevent folding of the residual water peak tailback into the spectral region of interest. These additional acquisitionsare usually free of charge, because oversampling amounts to averaging;and a main limitation for the detection of J-coupled metabolites is thesensitivity.

Nuisance gradients, caused mainly by eddy currents, lead to a lineshapedistortion. Knowledge of the qualitative behaviour of these gradients isimportant for devising correction strategies. Eddy currents are space andtime dependent and induced by switching the gradients of the MR system.7

They decay in a multi-exponential fashion, leading to additional multi-exponential gradients. A specific voxel at a single echo time always showsthe same decay, as long as the experimental settings are not altered byany intermediate preparations. To correct PRESS spectra, it thereforesuffices to measure the reference signal with an identical experiment, butomitting the water suppression. The phase of the echo in the time domainis ideally constant (for water being on-resonant) and any deviation reflectsthe additional nuisance gradient. Symmetric and undistorted lineshapescan be restored by subtracting the phase of the reference signal from the


data under analysis, which amounts to a phase deconvolution.48

The eddy currents in a 2D JPRESS experiment are induced mainly bythe last pair of crusher gradients. Acquisition in maximum-echo samplingalways starts at the same relative time after these gradients. Hence itsdecay behaviour is similar for all echo times, despite the shifted echo top.This important finding was verified by phantom measurements shown inFig. 3.4 and can be used to devise an eddy current correction scheme. Asingle 1D spectrum, typically from the shortest echo time, has to be ac-quired without water suppression. The phase of this 1D reference signalcan be subtracted from the phase of each echo of the actual experimentin the time domain. Alternatively, this qualitative behaviour may serveas a physically realistic model of the gradient distortion in fitting rou-tines. The most feasible model for the phase distortions in JPRESS wasa biexponential decay with fixed decay times.49

Phase-sensitive JPRESS spectra exhibit phase-twisted 2D lineshapesarising from the product of two complex signals. Each domain contains ab-sorption (A) and dispersion (D) components. The real part of the (phased)spectrum is given by

Re{(A1 + iD1)(A2 + iD2)} = A1A2 −D1D2. (3.2)

Phase-twisted lineshapes are inherent to this kind of experiment owing tophase modulation in t1.4,7 However, the cross-section through the centreof the peak still exhibits a pure (1D) absorption lineshape and can be pro-cessed in complex mode. The 2D lineshape is further complicated by thetilt of the peak tails due to the maximum-echo sampling scheme (Fig. 3.3),resulting in a reduced peak tail of the Voigt lineshape. Moreover, the t1dimension starts at the minimum echo time and hence some J evolutionhas already taken place. The 2D peaks of J-coupled metabolites thereforeexhibit a small anti-phase component (Fig. 3.1, bottom). Nevertheless,these effects are deterministic and do not hamper a quantification by fit-ting. The advantages of quantifying the real part of the spectrum (i.e. lessoverlap) greatly exceed its disadvantages (i.e. distorted lineshapes withnegative components).

All experiments were performed on a Philips Intera 3T whole-bodyscanner (Philips Medical Systems, Best, The Netherlands) equipped witha transmit/receive head coil. The echo times ranged from 31 to 229 ms insteps of ∆TE = 2 ms and the bandwidths in f1 and f2 were 0.5 and 2 kHzwith 100 and 2048 sampling points, respectively. Four-step phase cycling


0

0.5

1

mag

nitu

de

−1

−0.5

0

orig

inal

pha

se

0 50 100 150 200−1

−0.5

0

eddy

cur

rent

co

rrec

ted

phas

e

t2 [ms]

Figure 3.4: Effect of eddy currents on the echo signal, depicted hereby plotting all echo times in lines with different colours/shades.The upper, middle and bottom rows show the magnitude, the orig-inal phase and the corrected phase of the echoes, respectively. Theecho is acquired in a phantom (15 mM Cr) without water suppres-sion and the original phase should be ideally zero. Any phase dis-tortion is due to experimental imperfections, attributed to mainlyeddy currents. Different echo times always give rise to the samephase distortion (middle) despite the shifted echo top (upper plot).Therefore, it is sufficient to acquire a reference spectrum for a sin-gle echo time and subtract its phase from the phase of all signalsacquired with other echo times (bottom).


for each echo and a repetition time of TR = 2.5 s amount to a totalscan duration of 17 min. Water was suppressed by selective excitationin conjunction with gradient spoiling. Twenty-seven healthy volunteers(age 35.4 ± 7.5 years; 20 female, 7 male), who provided written informedconsent, were scanned in the parietal lobe. The voxel was 2.5 cm × 2.5 cm× 2.5 cm in size, consisting mainly of grey matter, but some contaminationwith white matter was unavoidable. Quantification was performed with thetwo-dimensional fitting procedure ProFit.49 Another exemplary JPRESSspectrum was acquired in a transplanted kidney (age 39 years; female)with the same parameters as used for the brain.

3.3 Results and Discussion

The maximum-echo sampling scheme has several advantages. Highly ben-eficial for the robustness of the experiment is the tilt of the peak tails,hence improving robustness of the standard PRESS sequence, while re-taining its ease of application. A typical JPRESS spectrum of a healthyvolunteer scanned in the parietal lobe is shown in Fig. 3.5. The contami-nation from poorly suppressed water is greatly reduced. In the traditionalhalf-echo sampling, the spectra are sometimes contaminated, hence impair-ing quantification. However, the maximum-echo sampled spectrum oftenallows accurate quantification. This effect can be also utilised for spectracontaining a large amount of lipids, as shown in a transplanted kidney inFig. 3.6. The glutamate and glutamine resonance is separated from lipids,hence greatly reducing contamination.

In maximum-echo sampling, most of the signal is acquired, hence in-creasing sensitivity. An analytical sensitivity comparison is given in theAppendix. Some actual examples for 1D PRESS and 2D JPRESS withtypical values are given in Table 3.1. The maximum-echo sampling gener-ally leads to a considerable gain in sensitivity as compared with half-echoJPRESS. For longer T2 relaxation times, the sensitivity between PRESSand maximum-echo sampled JPRESS is approximately the same. It isimportant, however, to restrict the sampling time along t1 for metabo-lites with short T2s. The J evolution has to be sufficiently resolved, hencethe maximum echo time of 229 ms is a reasonable choice. The sensitivitycalculations (Appendix) considered only non-coupled spins. Although Jcoupling can be included, this is not particularly helpful and intuitive asit must be done individually for each metabolite. Most brain metabolites

3.3. RESULTS AND DISCUSSION 51

f 1 [Hz]

f2 [ppm]

12345

−20

−10

0

10

20

−5

0

5NAAH

2O Cr Cr

Cho

GluGln

GSH

mI mINAAGluGln

Figure 3.5: Typical in vivo JPRESS spectrum plotted with con-tour lines on logarithmic colour scale. Only the real part of thephased spectrum is shown, hence also giving rise to negative com-ponents. The data were zero-filled to 512 samples in the indirectdimension and apodised with a Gaussian filter with 2 Hz width inboth t1 and t2. Directly visible are the predominant singlets from N-acetylaspartate (NAA), total creatine (Cr) and choline-containingcompounds (Cho), but also some of the J-coupled metabolites,namely glutathione (GSH). More information is still present inthis spectrum, which can be extracted with advanced post-processingmethods such as two-dimensional fitting.


f 1 [

Hz]

f2

[ppm]12345

−20

−10

0

10

20 −0.3

−0.2

−0.1

0

0.1

0.2

0.3

lipids

lipids

TMA

osmolytes

Figure 3.6: JPRESS spectrum of a transplanted kidney. The as-signment of resonances is based on Dixon and Frahm.50 The lipidsand the partially suppressed water are tilted away from the spec-tral region of interest, hence contamination is greatly reduced. Theshort T2 relaxation time is visible by broad linewidths in both f1

and f2.

T2 (ms) 100 150 200 250Short-TE PRESS 100 100 100 100

Maximum-echo JPRESS 62 82 95 104Half-echo JPRESS 43 55 63 69

Table 3.1: Comparison of sensitivity (%) between 1D PRESS and2D JPRESS with both maximum-echo and half-echo sampling forT ∗2 = 50 ms (corresponding to a line-width of 6.4 Hz) and the actualt1 sampling range of TS1 = 200 ms. The sensitivity of the PRESSexperiment is scaled to 100 % for each T2 time The maximum-echosampling scheme is highly beneficial for gaining sensitivity in 2Dexperiments.

3.3. RESULTS AND DISCUSSION 53

3.8 4 4.2 4.4 4.6 4.8 5 5.2 5.4 5.6

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

f2

[Hz]

sig

na

l [a

.u.]

Figure 3.7: Zero-Hertz cross-section through JPRESS spectrum ofwater with (dashed) and without (solid) eddy current correction.Symmetry is increased, hence increasing also the peak height.


are strongly coupled at typical field strengths of B ≤ 3 T, hence imped-ing analytical calculations. However, comparing the experiments providessome insight. The J evolution depends in both PRESS and JPRESS onthe minimal echo time. A PRESS spectrum can be reconstructed fromJPRESS by projection onto the f2 axis. This can lead to signal loss whenanti-phase terms are summed. Regular 1D PRESS always suffers fromthese destructive interferences and therefore performing a 2D experiment,where the J evolution is encoded, is advantageous. The minimum echotime of 31 ms already leads to anti-phase components. However, such along minimum TE is state of the art on whole-body 3 T scanners, owingto low transmitting RF field strengths and hence long pulses.

The scan duration is closely related to the sensitivity. The relativelylong scan time of 17 min was mainly chosen to increase sensitivity and canbe readily shortened by decreasing the phase-cycling steps or the amountof oversampling in the indirect dimension. Another favourable option forthe sensitivity would be the reception with surface coils. However, this wasnot considered here, because of the reduced bandwidths of the RF pulseswhen transmitting with the body coil instead of the transmit/receive headcoil. As a consequence, the chemical shift displacement is large, giving riseto localisation errors and anomalous J evolution.9

The effect of the eddy current correction is clearly visible in the in vitroexperiments without water suppression (Figs. 3.4 and 3.7). The strengthof the last pair of crusher gradients is increased from 12 to 20 mT/min order to provoke eddy currents. The reference signal should be mea-sured in the same scan without intermediate preparations. However, thiswas impossible in the current implementation and the separately measuredreference scan showed slightly different eddy current effects compared withthe actual measurement. In general, it is difficult to see the spectral dis-tortions from eddy currents in the in vivo spectra, because of linewidthsof typically 6-9 Hz. Moreover, these distortions are small owing to highlyoptimised gradient systems.

The acquired data can also be reconstructed differently, for instanceas CSSF.44 Other options are JPRESS with different J evolutions, whichcan be achieved by moving the reference of the 2D Fourier transformationalong t1. For example, the medium echo time often implies more in-phaseevolution for some metabolites at the cost of greatly increased truncationwiggles. Shifting the data in either domain does not alter its informationcontent. For a subsequent fit of the spectra, the reconstruction choice is

3.4. CONCLUSIONS 55

mainly governed by ease of implementation, thus favouring JPRESS overCSSF.

3.4 Conclusions

The maximum-echo sampling scheme increases sensitivity, while at thesame time further improving the specificity of the JPRESS experiment.In addition to the description of the qualitative behaviour of the eddycurrents in 2D JPRESS, this is the basis for a successful and accuratequantification of the data using the 2D fitting procedure ProFit.49

3.5 Appendix: Sensitivity Comparison

A common definition of the sensitivity is the signal-to-noise (SNR) ratioper unit time.4 For identical experimental durations and a constant ac-quisition time window, the noise will remain the same and it is sufficientto compare the signal. The signal is commonly defined as the peak heightin the frequency domain, which is equivalent to the integral of the timedomain signal. For singlets, it suffices to integrate over the exponentialdamping curve of the echo along the times t1 and t2. The digitisation doesnot fundamentally change the signal and therefore it suffices to take thecontinuous integral. The 1D PRESS experiment can be compared to 2Dexperiments simply by integrating over the indirect t1 axis with always thesame intensity. The signal from PRESS is therefore given by

∫ TS1

0

∫ ∞

0

exp(− t2

T ∗2

)dt2dt1 = T ∗2 TS1, (3.3)

where TS1 is the sampling time along t1. The PRESS sequence with theshortest possible echo time is used for an equitable comparison. The echotime of this PRESS sequence is also equivalent to the shortest echo timeof JPRESS.

The signal of JPRESS can be split up into two halves. The dampingcurves in the traditional half-echo sampling is equivalent to the right halfand given by∫ TS1

0

∫ ∞

0

exp(− t2

T ∗2

)· exp

(− t1

T2

)dt2 dt1 = T ∗2 T2

(1− exp

(−TS1

T2

)).

(3.4)


The left half of the maximum-echo signal can be calculated by

∫ TS1

0

∫ 0

−t1/2

exp(

t2T ∗2

)· exp

(− t1

T2

)dt2 dt1 =

T ∗2 T2

(1− exp

(−TS1

T2

))− T ∗2 tx

(1− exp

(−TS1

tx

)), (3.5)

where tx = 2T2T∗2T2+2T∗2

is introduced for notational convenience. Addingboth halves together, yields the total signal of the maximum-echo sam-pled JPRESS

2T ∗2 T2

(1− exp

(−TS1

T2

))− T ∗2 tx

(1− exp

(−TS1

tx

)). (3.6)

These formulae can now be used to compare the various experiments witheach other.

The ratio maximum-echo JPRESS (Eqn. 3.6) to PRESS (Eqn. 3.3) isgiven by

2T2

TS1

(1− exp

(−TS1

T2

))− tx

TS1

(1− exp

(−TS1

tx

)), (3.7)

and half-echo JPRESS (Eqn. 3.4) to PRESS (Eqn. 3.3) by

T2

TS1

(1− exp

(−TS1

T2

)). (3.8)

Examples with typical values are given in Table 3.1.

Chapter 4

Prostate Spectroscopy at3 Tesla UsingTwo-DimensionalS-PRESS

(published in: Magnetic Resonance in Medicine51)

4.1 Introduction

Because of its non-invasive nature, magnetic resonance imaging (MRI)has proven to be an indispensable method for diagnosing many patholo-gies. However, the morphological and relaxometric information providedby MRI is not sufficient for unequivocally diagnosing certain tumor types.In such cases MR spectroscopy (MRS) can provide valuable complemen-tary metabolic information.52 For the diagnosis of prostate cancer, themetabolites choline (Cho) and citrate (Cit) are of particular interest.52,53

Cit is secreted in relatively large amounts by the epithelial cells of theprostate. High levels of zinc inhibit the oxidation of Cit in the Krebscycle.54 However, in cancerous epithelial tissue the zinc concentrationis considerably reduced, resulting in lower net Cit levels. On the otherhand, the Cho concentration is increased due to the rapid cell turnover in

57

58 Prostate Spectroscopy at 3 Tesla Using Two-Dimensional S-PRESS

tumorous tissue. Therefore, the Cho-to-Cit concentration ratio can helpto discriminate between malignant adenocarcinoma and benign hyperpla-sia.53,55,56

The preferred method for localized prostate 1H MRS is the double spin-echo sequence called point-resolved spectroscopy (PRESS).27 Because ofits high signal yield (e.g., compared to stimulated-echo acquisition mode(STEAM)57), it is the method of choice, especially when surface coils areused for signal reception.

While the detection and quantification of the uncoupled Cho resonanceis straightforward, Cit represents a strongly coupled AB spin system attypical field strengths available for in vivo MRS (B0 ≤ 7 T ) and is there-fore a challenge for spectroscopists. The strong coupling gives rise to acomplicated J-modulation of the Cit resonance that critically depends onthe sequence timing. Several groups have investigated the signal behav-ior of Cit and optimized the PRESS sequence to yield maximum signalintensity and a line-shape appropriate for Cit quantification.1,58–63

Expanding PRESS to a second dimension has been shown to be usefulfor disentangling overcrowded spectra. A powerful method that enablesthe separation of chemical shift and coupling information for J-coupledmetabolites is 2D J-resolved PRESS (JPRESS).24 PRESS spectra are ac-quired with varying echo times, which encode the indirect t1 dimension.The application of JPRESS in prostate examinations enables the spectralseparation of creatine (Cr) and Cho from spermine (Spm), which has asimilar resonance frequency.38,39 Kim et al.64 combined JPRESS withspectroscopic imaging (SI) to identify the region of abnormal metabolismdue to prostate cancer more accurately. The line-widths in the indirectdimension of JPRESS spectra are governed only by the T2 decay and notby field inhomogeneities as in 1D spectra, where T ∗2 determines the line-widths. Thus a better spectral resolution of coupled resonances is achieved.The possibility of determining T2 values from 2D J-resolved spectra is anadditional benefit.

However, strong scalar coupling leads to additional peaks in 2D spec-tra, which can complicate peak assignment and quantification.65 Whileweakly coupled two-spin systems give rise to four peaks in a JPRESSexperiment, the Cit molecule shows a spectral pattern with eight peaks.The additional peaks stem from the coherence transfer between the cou-pled spins. Thrippleton et al.65 suggested several methods for suppressingthese strong coupling peaks in JPRESS experiments. One approach is a

4.2. THEORY 59

multi-scan technique that employs an incremental simultaneous displace-ment of the two refocusing pulses within the sequence, leaving the totalecho time (TE) unchanged. Gambarota et al.66,67 proposed a differenceediting method based on the strong-coupling point-resolved spectroscopy(S-PRESS) sequence. It uses PRESS localization (90◦x−[TE1/2]−180◦y−[TE1/2 + TE2/2]− 180◦y − [TE2/2]− Acq) and consists of two measure-ments with the same overall TE (TE = TE1 + TE2 = const), but withdifferent TE1 and TE2. When the two resulting spectra are subtractedfrom each other, uncoupled and weakly coupled resonances are suppressed,while the strong coupling peaks remain. At a field strength of 3 T, densitymatrix simulations predicted a maximum J-modulation of Cit for a TEof 280 ms and helped to determine the optimal TE1 values for differenceediting.

In this work we combine the strengths of 2D spin-echo spectroscopyand the S-PRESS approach by implementing S-PRESS as a 2D technique.The method was analytically optimized for prostate MRS and validated invitro as well as in vivo. 2D S-PRESS was used to determine the spectralparameters of Cit (J and δ) in vivo with relatively high precision.

4.2 Theory

The 1H MR signal of Cit arises from two magnetically equivalent methy-lene groups (CH2) that constitute strongly coupled AB spin systems atcurrently available field strengths for in vivo MRS (B0 ≤ 7 T). In 1DPRESS the Cit resonance consists of four lines (Fig. 4.1) centered around2.6 ppm. While the separation between the two leftmost and the two right-most lines equals the field-independent coupling constant (J = 16.1 Hz),the distance between the two inner lines depends on the magnetic fieldstrength, due to the linear field dependence of the chemical shift differencebetween the A and B spins (δ = 0.149 ppm).68 As a characteristic featureof strongly coupled spin systems, the two inner lines show higher signal in-tensity than the two outer lines. Furthermore, the strong coupling causesa complicated amplitude and phase modulation of the Cit resonance as afunction of TE. This behavior cannot be easily understood in the classicsense, and requires a quantum mechanical description. The Hamiltonianfor an AB spin system can be written as

H = (ωAAz + ωBBz) + 2πJ(AxBx + AyBy + AzBz) (4.1)


Figure 4.1: Plot of the citrate signal as obtained in a pulse-acquireexperiment at 3 T. J denotes the strength of the scalar coupling inHz and Λ =

√δ2 + J2 the strong coupling frequency, where δ is the

difference in Larmor frequencies between the two coupled protons.The total intensity is normalized to one.

where Ax,y,z and Bx,y,z are the angular momentum operators in Carte-sian coordinates, ωA and ωB are the chemical shifts, and J is the couplingconstant of the two coupled spins A and B. The nonsecular contributionAxBx +AyBy is characteristic of strongly coupled spin systems and is usu-ally neglected in the limit of weak scalar coupling. Writing this expressionin terms of raising and lowering operators yields 1/2 · (A+B− + A−B+),indicating that there is a coherence transfer between spins A and B dueto strong coupling. By analytically calculating the density matrix evolu-tion with this Hamiltonian or applying the Kay-Mc-Clung formalism,69

the signal resulting from a PRESS sequence can be expressed in termsof four groups of Cartesian single-quantum coherence product operators{(Ay + By), (2AxBz + 2AzBx), (Ax − Bx), (2AyBz − 2AzBy)}.1 The am-plitudes of the detectable in-phase components My and Mx are:

〈My〉 = 〈Ay〉 = 〈By〉= − cos(πJ · TE) · [(δ/Λ)2 + (J/Λ)2 cos(πΛ · TE)]

− sin(πJ · TE) · [(J/Λ)3 sin(πΛ · TE)

+(2Jδ2/Λ3) sin(π

2Λ · TE) cos(

π

2Λ · (TE1 − TE2))] (4.2)

4.2. THEORY 61

〈Mx〉 = 〈Ax〉 = −〈Bx〉= − cos(πJ · TE) · [(J2δ/Λ3)2(2 sin(

π

2Λ · TE)

· cos(π

2Λ · (TE1 − TE2))− sin(πΛ · TE))]

− sin(πJ · TE) · [(2Jδ/Λ2) sin(π

2Λ · TE)

· sin(π

2Λ · (TE1 − TE2))] (4.3)

Most amplitude terms only show a dependence on the total TE, but areindependent of the particular choice of TE1 and TE2. However, the ampli-tude expressions also contain summands depending on the difference TE1

- TE2. These terms are characteristic for strongly coupled spin systemsand vanish in the weak-coupling limit.

A 2D S-PRESS experiment consists of multiple PRESS measurementswith a constant TE but a simultaneous incremental variation of TE1 andTE2. When TE1 is encoded in the indirect dimension (t1) and a 2DFourier transformation is applied to the data, the resulting 2D spectrumdisplays the strong coupling information resolved in the indirect spectraldimension (f1). Strongly coupled spin systems reveal a line-splitting inf1, whereas uncoupled and weakly coupled metabolites resonate only atf1 = 0 Hz. For AB spin systems, these strong coupling peaks arise fromthe terms depending on TE1 − TE2 in Eqs. (4.2, 4.3) and appear atfrequencies f1 = ±Λ/2, where Λ =

√δ2 + J2. Since all acquired echo

signals have undergone the same T2 decay and are aligned along their echotops in t1, the spectral resolution in the indirect dimension of a 2D S-PRESS spectrum is limited only by the TE1 encoding range, and not byfield inhomogeneities or T2 relaxation. The signal intensity of the strongcoupling S-PRESS peaks at f1 = ±Λ/2 shows a modulation with TE. Itcan be calculated as a function of TE from the amplitude terms given inEqs. (4.2, 4.3), as shown in the Appendix. The strong coupling peaks evenvanish for certain TEs (TE = {n · 2/Λ, nεN}). This has to be taken intoaccount when setting up a 2D S-PRESS experiment.

While the trigonometric factors in the amplitude terms (Eqs. (4.2, 4.3))can be adjusted and optimized via TE, the polynomial factors depend onlyon the coupling strength κ = J/δ of the AB spin system, which can be


shown with basic substitutions:

P1,x = 2J2δ/Λ3 =2 · κ2

(1 + κ2)1.5 (4.4)

P2,x = 2Jδ/Λ2 =2 · κ

1 + κ2(4.5)

P1,y = 2Jδ2/Λ3 =2 · κ2

(1 + κ2)1.5 (4.6)

Plotting these polynomial factors (Fig. 4.2) shows that they reach maximafor κmax =

√2, κmax = 1 and κmax = 1/

√2, respectively. This means

that the highest signal yield for S-PRESS experiments is achieved for spinsystems with κmax = J/δ ≈ 1, which is the case for Cit at a field strengthof 3 T.

The spectral parameters of Cit (J and δ) sensitively depend on thechemical environment. Van der Graaf et al.68 showed in phantom exper-iments that J and δ increase with the concentration of divalent cations,such as Mg2+, Ca2+, and Zn2+. They also reported a strong dependenceon the pH value. This raises the question as to whether there is a de-tectable difference in J and δ between cancerous and healthy tissues. TheZn2+ concentration plays a particularly important role in the pathophys-iology of adenocarcinoma. It is largely reduced in cancerous tissue andmay therefore be useful as a marker for prostate cancer. Prostate spec-troscopy usually exploits the fact that zinc ions prevent Cit from beingoxidized. A reduced Zn2+ concentration is accompanied by a diminishedCit level, which is detected in MRS experiments. Therefore, we carriedout a phantom experiment to investigate whether this zinc decrease couldbe detected on the basis of changes in J and δ.

4.3 Materials and Methods

All experiments were performed on a Philips Achieva 3T system (PhilipsMedical Systems, Best, The Netherlands). Phantom experiments werecarried out with a transmit/receive head coil. All in vivo spectra were ac-quired with a two-element ellipsoidal surface coil (semi-major axis = 17 cm,semi-minor axis = 14 cm), and the body coil was used for transmission.

4.3. MATERIALS AND METHODS 63

Figure 4.2: Polynomial amplitude factors of the strong couplingterms in Eqs. (4.2, 4.3) as functions of the coupling strength κ =J/δ: a) P1,x = 2J2δ/Λ3, b) P2,x = 2Jδ/Λ2, c) P1,y = 2Jδ2/Λ3.


Both 2D S-PRESS and JPRESS measurements are based on the PRESSsequence:

90◦x− [TE1/2]−180◦y− [TE1/2+TE2/2]−180◦y− [TE2/2]−Acq (4.7)

The acquisition employed a maximum-echo sampling scheme20,45 with1024 sampling points covering a bandwidth of 2000 Hz. For both 2DS-PRESS and JPRESS experiments the measurement started with themanufacturer’s default PRESS implementation, in which TE1 is as shortas possible. In the 2D S-PRESS experiments TE1 and TE2 were variedsimultaneously to keep the total TE constant. TE1 was encoded in theindirect t1 dimension. For the JPRESS experiments, starting from theshortest possible TE, only TE2 and hence the total TE were varied andencoded in the indirect t1 dimension.

The water signal was suppressed with selective excitation in conjunc-tion with gradient spoiling prior to the PRESS sequence. Band-selectiveinversion with gradient dephasing (BASING) further improved the wa-ter suppression for the 2D S-PRESS measurements.70,71 The combina-tion of these two techniques suppressed the water resonance almost com-pletely, and thus prevented folding artifacts from large water peak tails.Fat suppression was omitted because careful voxel selection ensured thatno periprostatic fat was included. The TEs for the 2D S-PRESS mea-surements were longer than 200 ms and therefore were much longer thantypical T2 values of macromolecules.

In vitro measurements were carried out on a phantom solution con-taining 15 mM Cho, 30 mM lactate (Lac, weakly coupled), and 50 mMCit (strongly coupled) to demonstrate the different behavior of weakly andstrongly coupled spin systems. A 2D S-PRESS spectrum was acquired atTE = 438 ms using 20 TE1 encoding steps with a spacing of ∆TE1 =20.86 ms, yielding a nominal spectral resolution ∆f1 = 2.4 Hz in the in-direct dimension. For comparison a JPRESS spectrum was acquired withthe same nominal resolution in both f1 and f2.

In vivo spectra were acquired from the prostates of healthy subjects,who provided written informed consent prior to participating in the study.The voxel (14 mm × 27 mm × 14 mm = 5.3 ml) was positioned in thecenter of the prostate. A repetition time (TR) of 1.2 s was chosen becausethe T1 relaxation time of Cit is quite short (≈ 470 ms) and a high sensitiv-ity for Cit detection was desired.72 Forty equidistant TE1 encoding stepsin the range of 32.54 to (TE - 21.12) ms with 16 spectral averages each


and a 16-step phase-cycling scheme were used for the 2D S-PRESS exper-iments, amounting to an overall scan time of 13 min. Two-dimensionalS-PRESS spectra were acquired for TEs of 235 ms, 282 ms, 313 ms, and353 ms with nominal spectral resolutions of 5.4 Hz, 4.3 Hz, 3.8 Hz, and3.2 Hz, respectively, in the indirect dimension. These TEs were chosen onthe basis of a pilot measurement that suggested intensity maxima (TE =282 ms and TE = 353 ms) and complete cancellation (TE = 235 ms andTE = 313 ms) of the strong coupling peaks for certain TEs. For compari-son a JPRESS spectrum was acquired using 80 encoding steps (TE range:53.67 - 282 ms), eight spectral averages, and eight phase cycles. Thisamounts to the same overall scan time. The nominal spectral resolutionin the indirect dimension (4.3 Hz) was chosen to be the same as for theS-PRESS experiment with TE = 282 ms. For the JPRESS acquisition,BASING water suppression was omitted to avoid a restriction to long TEsand an associated loss of sensitivity. Considerable oversampling in the in-direct dimension was necessary to prevent impairment of the 2D spectraby back-folded peak tails of the residual water signal. Three in vivo 2DS-PRESS spectra from the same subject, acquired with different TEs andin different scan sessions, were used to optimize the measurement protocolwith respect to the TE. The coupling constant J and the chemical shiftdifference δ of the two coupled protons were determined from the Cit peakpositions in the four spectra and averaged. Using these values, we opti-mized the 2D S-PRESS sequence for Cit detection by calculating the peakintensity dependence on the TE under in vivo conditions (see Appendix).Furthermore, the inter-subject variability of J and δ was determined by2D S-PRESS experiments (TE = 280 ms) on 11 healthy subjects (21-45years old).

To investigate whether a zinc decrease in the prostate due to prostatecancer could be detected on the basis of changes in J and δ, we preparedtwo phantom solutions with approximate in vivo metabolite concentrationsas found in literature (Table 4.1), with and without zinc ions. In two 2DS-PRESS experiments, the spectral parameters J and δ were determinedfor both phantoms.

Postprocessing of all acquired data was performed with in-house-writtenMATLAB code (The MathWorks, Inc.) and included zero-filling to 512samples in t1 and 2048 samples in t2, a 2D Fourier transformation of theacquired data, and 1.5 Hz Gaussian filtering in both f1 and f2. All spectrain this paper are presented in magnitude mode as logarithmically scaled


Metabolite Concentration [mM] ReferenceCit 90 68

Cho 9 73

Cr 12 73

Lac 10.3 70

Spm 18 73

Na+ 145 68

K+ 61 68

Ca2+ 18 68

Mg2+ 14.7 68

Zn2+ 8.5 68

Cl− 47.4 68

Table 4.1: Literature values for metabolite concentrations in thehealthy prostate.68,70,73 A solution with these concentrations wasused to investigate the influence of the Zn2+ concentration on thespectral parameters of citrate under in vivo conditions.

color plots.

4.4 Results

The results of the in vitro experiments are presented in Fig. 4.3. TheJPRESS spectrum (a) shows a simple splitting of the Lac resonance (twopeaks separated by J ≈ 7 Hz in f1) and the typical multiplet of eight peaksfor Cit. Substituting TE2 by TE − TE1 in the analytical terms of Eqs.(4.2, 4.3), expanding the trigonometric expressions and collecting contri-butions with a TE dependence yields resonances at f1 = {±J/2,±Λ/2 ±J/2,±Λ/2 ∓ J/2} and thus confirms the results from numerical simula-tions.39 In the 2D S-PRESS spectrum (b), the Lac resonance does notshow a line-splitting along f1. The Cit signal consists of the typical quar-tet at f1 = 0 Hz, which is also observed in 1D PRESS, and two doubletsat f1 = ±Λ/2 (henceforth called “strong coupling peaks”). The reso-nance frequencies of these peaks in f2 correspond to those observed in 1DPRESS spectra. The spectral parameters of Cit were determined fromthe positions of the strong coupling peaks in f1 and f2: J = 15.13 Hz,

4.4. RESULTS 67

Figure 4.3: JPRESS (a) and 2D S-PRESS (b) magnitude spec-tra (acquired at 3 T from a phantom solution containing 15 mMcholine (Cho), 30 mM lactate (Lac) and 50 mM citrate (Cit)).The JPRESS experiment contained 80 TE encoding steps in therange [41.58 ms, 438 ms]. The 2D-SPRESS experiment (TE =438 ms) contained 20 TE1 encoding steps in the range [25.96 ms,422.40 ms]. The citrate multiplets with peak assignments are shownin zoomed sections (right graphs).


Figure 4.4: JPRESS (a) and 2D S-PRESS (b) magnitude spectraacquired at 3 T from the prostate of a healthy subject. The JPRESSexperiment contained 80 TE encoding steps in the range [53.67 ms,282 ms]. The 2D-SPRESS experiment (TE = 282 ms) contained40 TE1 encoding steps in the range [32.54 ms, 260.36 ms]. Detectedmetabolites are: choline (Cho), citrate (Cit), creatine (Cr), lipids(Lip), spermine (Spm).

δ = 14.34 Hz, Λ = 20.84 Hz.

Figure 4.4 shows an in vivo JPRESS spectrum (a) with TE rang-ing from 53.67 to 282 ms and a 2D S-PRESS spectrum (b) with TE =282 ms. Both spectra have a nominal spectral resolution of 4.3 Hz inthe f1 dimension. In both spectra, Cho, Cr, and Cit resonances are visi-ble. Additionally, the JPRESS spectrum shows a Spm multiplet around3.1 ppm and lipid (Lip) resonances around 1.3 ppm. Because of their shortT2 relaxation time, those metabolites are not visible in the 2D S-PRESSspectrum, which does not contain short-TE data. The multiplet patternof Cit is similar to the in vitro case (Fig. 4.3). Only the relative signalintensities of the JPRESS peaks are different. The following in vivo Citparameters were determined as averaged values from this and two other 2DS-PRESS spectra acquired from the same subject: J = (15.77± 0.20) Hz,δ = (19.94± 0.40) Hz, Λ = (25.42± 0.66) Hz. Figure 4.5 shows two crosssections through the 2D S-PRESS spectrum at the frequencies f1 = 0 Hzand f1 = −12.7 Hz. The cross section at f1 = 0 Hz shows four reso-

4.4. RESULTS 69

11.522.533.5

0

0.02

0.04

0.06

0.08

0.1

f2

[ppm]11.522.533.5

0

0.02

0.04

0.06

0.08

0.1

0.12

f2

[ppm]

a) b)

Figure 4.5: Cross section through the 2D S-PRESS magnitude spec-trum in Fig. 4.4b at f1 = 0 Hz (a) and f1 = −12.7 Hz (b).

nance lines and looks similar to a usual PRESS spectrum acquired withthe same TE. In the cross section at f1 = −Λ/2 = −12.7 Hz, only aCit doublet is visible at the two leftmost Cit resonance frequencies in f2.These two peaks are separated by ∆f2 = J . The intensity modulation ofthe strong coupling peaks was determined according to Eq. (4.14) withthe measured in vivo parameters J and δ. The intensity graph (Fig. 4.6)shows zero-crossings for TE = {n · 2π/Λ, nεN}. We verified the ac-curacy of the determined spectral parameters by acquiring additional 2DS-PRESS spectra with TEs of 235 ms, 313 ms, and 353 ms (Fig. 4.7). ForTE = 235 ms and TE = 313 ms, the strong coupling peaks of Cit areexpected to have zero intensity, and indeed they vanished at these specificTEs. The maxima of the intensity graph in Fig. 4.6 are preferable TEsfor 2D S-PRESS in the human prostate. The TE should be large enoughto yield the required spectral resolution in f1 and short enough to avoidtoo much signal loss due to T2 relaxation. Therefore, TE = 278 ms andTE = 353 ms are most appropriate for S-PRESS prostate spectroscopy.

The spectral parameters of Cit (J and δ) for 11 healthy subjects arelisted in Table 4.2. They yield mean values of J = (15.99 ± 0.49) Hzand δ = (19.99 ± 0.70) Hz. The inter-subject standard deviation (SD) isslightly larger than the SD due to different voxel positions and measure-ment imperfections only.

The 2D S-PRESS experiments with the two prostate phantom solutionsyielded the following spectral parameters:


Figure 4.6: In-vivo signal intensity of the strong coupling S-PRESSpeaks at f1 = ±Λ/2π as a function of TE, neglecting T2 relaxation.The echo times, for which 2D S-PRESS spectra were acquired, areindicated with dotted lines: TE = 235 ms, 282 ms, 313 ms, 353 ms(see Figs. 4.4b and 4.7a-c).

4.4. RESULTS 71

Figure 4.7: 2D S-PRESS magnitude spectra acquired at 3 T fromthe prostate of a healthy subject for echo times a) 235 ms, b)313 ms, c) 353 ms. The strong coupling peaks of citrate vanishfor TE = 235 ms and TE = 313 ms.


Subject Age J [Hz] δ [Hz]1 22 15.77 19.942 22 15.78 20.613 27 16.35 18.914 26 16.35 20.945 24 15.87 20.086 27 15.40 19.937 22 16.35 19.918 21 15.58 20.739 25 17.05 19.2210 45 15.46 19.0311 44 15.90 20.58

mean 27.7 ± 8.6 15.99 ± 0.49 19.99 ± 0.70

Table 4.2: Spectral parameters of citrate (J and δ) for 11 healthysubjects aged between 21 and 45, determined with 2D S-PRESSspectroscopy.

1. Phantom solution with zinc ions (8.5 mM, corresponding to healthytissue): J = 15.97 Hz, δ = 18.62 Hz.

2. Phantom solution without zinc ions (corresponding to cancerous tis-sue): J = 15.84 Hz, δ = 18.63 Hz.

4.5 Discussion

Two-dimensional S-PRESS facilitates the unequivocal detection and quan-tification of strongly coupled metabolites. The combination of 2D spec-troscopy techniques with PRESS localization is an obvious and promisingapproach since no additional pulses or gradients are required and the tech-nique is fairly robust. Spreading the spectral information into a seconddimension helps to disentangle overcrowded spectra and improve the res-olution of coupled resonances. JPRESS has been established as a usefulmethod for detecting J-coupled metabolites because it helps to distin-guish coupled from uncoupled resonances. Furthermore, it provides betterspectral resolution in the indirect dimension, where the line-widths are

4.5. DISCUSSION 73

governed only by T2 decay and not by field inhomogeneities. The appli-cation of a maximum-echo sampling scheme20 tilts the peak tail of theresidual water resonance away from the f2 axis and prevents them fromcontaminating the spectral region of interest (ROI).

However, JPRESS spectra can be relatively complex, especially whenstrongly coupled spins are involved, giving rise to additional peaks in thespectrum. This also hampers the determination of exact peak positions inf1 (Fig. 4.4a). For this reason Thrippleton et al.65 devised several strate-gies to suppress these strong coupling peaks in JPRESS and correlationspectroscopy (COSY) experiments. The 2D S-PRESS method represents acomplementary approach that enables strongly coupled metabolites to beidentified on the basis of their strong coupling peaks. In the case of an ABspin system, the 2D S-PRESS sequence gives rise to four strong couplingpeaks of equal intensity at one characteristic frequency: f1 = ±Λ/2. Thus,strongly coupled metabolites can easily be distinguished from uncoupledand weakly coupled ones, which resonate only at f1 = 0 Hz. A cross sec-tion through the 2D S-PRESS spectrum at f1 = ±Λ/2 can be used forvisual inspection and quantification of the Cit resonance (Fig. 4.5b). Thisparticular choice of a cross section is equivalent to the S-PRESS differenceediting approach presented by Gambarota et al.66,67 A cross section at f1

= 0 Hz, as presented in Fig. 4.5a, corresponds to the integral over the 2Dtime domain data along t1 with a subsequent 1D Fourier transformation.This is analogous to a TE-averaged spectrum,35 which can be retrievedfrom a JPRESS spectrum as a cross section at f1 = 0 Hz. Cross sectionsat f1 = ±Λ/2 can be reconstructed by multiplying the 2D time domaindata with the linear phase factor exp(iϕ) (with ϕ = πΛt1) prior to integra-tion along t1 and Fourier transformation. For S-PRESS difference editing,two spectra are acquired with TE1 = τ1 and TE1 = τ2, which are chosensuch that the Cit resonances for these two particular choices of TE1 havea phase difference 2π · Λ/2 · (τ1 − τ2) = π with respect to each other (Λ/2being the modulation frequency).66,67 Applying the above-mentioned lin-ear phase factor to the 2D S-PRESS data amounts to a phase difference∆ϕ = ϕ1 −ϕ2 = πΛ(τ1 − τ2) = π for the two acquired spectra, which cor-responds to a difference editing scheme when integrating over the indirectdimension (two-point Fourier transformation). Therefore, the S-PRESSspectrum shown by Gambarota et al.66,67 corresponds to a superpositionof the two cross sections in the 2D S-PRESS spectrum at f1 = ±Λ/2.

As an additional benefit, the spectral resolution in the indirect dimen-


sion of a 2D S-PRESS spectrum is limited only by the TE1 encoding range,and not by field inhomogeneities or T2 relaxation as in JPRESS. However,the truncation in the t1 domain requires apodization (∼ 1− 2 Hz) to pre-vent wiggles in the spectrum. For studies in which sensitivity is not amajor concern, a sufficiently long TE can be used to obtain the desiredspectral resolution in f1. To obtain an optimal SNR for the strong couplingpeaks, the TE has to be adjusted individually for every strongly coupledspin system. We found that TE = 278 ms and TE = 353 ms are prefer-able TEs for in vivo Cit detection. This result is in good agreement withGambarota et al.,66,67 where an optimal TE of 280 ms was suggested inthe context of the original S-PRESS implementation as a difference editingapproach.

As shown in this work, the spectral parameters J and δ of stronglycoupled metabolites can be determined with reasonable precision from a2D S-PRESS spectrum. Prostate spectroscopy using a 1D PRESS se-quence does not properly resolve the Cit multiplet. Even at 3 T, the twoinner lines usually overlap so heavily that they appear as one peak. 2D S-PRESS enables complete and unequivocal resolution of the Cit resonancesand thereby facilitates the determination of J and δ. The advantage of 2DS-PRESS over COSY,21 which is often used to determine in vivo couplingconstants, is that spatial localization can be achieved at no extra cost. Thedetermined spectral parameters (J = 15.99 Hz, δ = 0.157 ppm) slightlydeviate from the results obtained by van der Graaf et al.68 (J = 16.1 Hz,δ = 0.149 ppm), who measured a phantom solution with approximate invivo metabolite concentrations. It is particularly important to know thecoupling parameters when they are susceptible to changes in the chemi-cal environment of the molecule. A comparison of the measured in vitroand in vivo parameters of Cit revealed the following findings: In the Citphantom, J and δ were decreased by 5 % and 28 % compared to theirrespective in vivo values in healthy prostate tissue. Even in the prostatephantom that contained all major in vivo metabolites and ions, δ was sig-nificantly reduced compared to its value in healthy subjects. However,when the 2D S-PRESS spectra from the two prostate phantoms (with andwithout zinc ions) were compared, there was no significant difference inδ and only a marginal difference (0.13 Hz = 0.8 %) in J , which wouldnot be detectable under in vivo conditions. The experiments carried outby van der Graaf et al.68 on a phantom solution containing 90 mM Citand 8.5 mM zinc chloride yielded much larger differences (2.4 % in J and

4.5. DISCUSSION 75

5.5 % in δ), which may be detectable even under in vivo conditions con-sidering the estimates for the intra-individual error margins found in thiswork (1.3 % for J and 2.0 % for δ). This discrepancy reflects the factthat large concentrations of magnesium (14.7 mM) and calcium (18 mM)ions reduce the influence of zinc ions. The spectral parameters appearto depend on the total number of divalent ions, which undergoes only asmall relative change, even when the concentration of zinc ions drops dra-matically. Therefore the coupling evolution of Cit does not change signif-icantly under pathological conditions. This is an important point to notebecause a different coupling evolution in pathological and healthy tissuemight impair quantitative studies in which the Cho-to-Cit ratio in differ-ent parts of the prostate is compared (e.g., in an SI experiment) and usedfor clinical diagnosis. J and δ may also be influenced by the concentrationof Spm molecules, which have opposite polarity with respect to the Citmolecules.74 It correlates with the Cit concentration and is therefore alsodecreased in cancerous tissue. Apart from that, J and δ depend on theCit concentration itself because a reduced Cit concentration enhances theinfluence of the divalent ions. However, the reduced Cit and Spm concen-trations under pathological conditions were not taken into account for theprostate phantom. The pH dependence of the spectral parameters75 wasnot considered either, because it is primarily adjusted through the con-centration of Cit and divalent ions. Knowledge of the spectral parametersand the exact quantum-mechanical evolution is of great importance for thequantification of coupled metabolites. LCModel has been established asthe most objective and robust method for metabolite quantification in re-cent years.76 It models the measured spectrum as a linear combination ofsingle metabolite basis spectra, which can either be acquired in phantomexperiments or simulated. This makes LCModel particularly appropriatefor fitting metabolites with complicated coupling networks, for which nocompact analytical description of the coupling evolution under a PRESSsequence is available. It is difficult to exactly match in vivo conditions in aphantom experiment, and therefore the coupling evolution of metabolites,such as Cit, in a phantom solution can differ from the evolution in vivo.Since this results in quantification errors, simulating the basis spectra maybe a better approach for coupled resonances, such as those of Cit. For acorrect simulation, however, the exact spectral parameters J and δ mustbe known. A 2D S-PRESS spectrum delivers these parameters at no ex-tra cost. Instead of fitting and quantifying cross sections through the 2D


S-PRESS spectrum with LCModel, a 2D fitting method, such as ProFit,49

can be directly applied to the 2D S-PRESS spectrum to include all priorknowledge.

An alternative method for detecting strongly coupled spin systems waspresented by Trabesinger et al.,77 who suggested spectral editing witha single quantum coherence filter for taurine detection. This techniqueexploits the fact that detectable single quantum coherences are createdthrough the coherence transfer between the strongly coupled spins, whilefor uncoupled and weakly coupled spin systems only even coherence or-ders are produced, which are not directly observable in the acquisition.However, spectral editing techniques based on coherence pathway selec-tion usually have an inherent signal loss of 50 %. On the other hand, 2DS-PRESS requires longer TEs than common coherence filtering techniquesand is therefore not the method of choice for large metabolites with veryshort T2 relaxation times.

Since prostate cancer usually occurs in the peripheral zone of theprostate, single-voxel measurements, as performed in this work, are notappropriate for clinical diagnosis. However, one could increase the SNRconsiderably by using an endorectal coil for signal reception instead ofsurface coils. This might enable a combination of 2D S-PRESS with SI inorder to obtain spatial distributions of metabolites over the whole prostate.Fat suppression, which is not an issue for the single-voxel approach, couldbe achieved with either broadband outer volume suppression (OVS) pulsesor a dual-band BASING pulse, which inverts water and fat at the sametime.

Analytical considerations presented in the Theory section show that Citis an ideal metabolite for S-PRESS. Its coupling strength ensures a near-optimal signal yield at 3 T, and the comparably simple signal structure ofan AB spin system is appropriate for analytical investigations, which canbe used to optimize the sequence. However, 2D S-PRESS may also turnout to be useful for spectroscopy in the brain, where several metabolites ofinterest (e.g., glutathione, GABA, myo-inositol) are strongly coupled. Inparticular, the cysteinyl group of glutathione, which has a relatively simplecoupling network (ABX), may be an interesting target metabolite. Gam-barota et al.78 recently investigated the signal modulation of glutathionewith TE1 and TE applying numerical and experimental methods. How-ever, a compact analytical description for the signal response of an ABXspin system to a PRESS sequence would be helpful for optimizing 2D

4.6. CONCLUSIONS 77

S-PRESS for glutathione detection.

4.6 Conclusions

We have shown that 2D S-PRESS is a viable alternative to the widelyused JPRESS method for prostate spectroscopy. The technique may helpto further establish MRS as a method for detecting metabolic changesassociated with prostate cancer, and has the potential to be useful forspectroscopy in other anatomical regions as well.

4.7 Appendix

The overall signal intensity of the strong coupling peaks can be calculatedas the integral over the 2D magnitude spectrum. An integration over acomplex spectrum f(ω) = fr(ω)+ i ·fi(ω) corresponds to the time domainsignal F (t) = Fr(t) + i · Fi(t) at the reference point (zero point) of theFFT:

F (t) =∫

f(ω) · exp(iωt) · dω ⇒ F (0) =∫

f(ω) · dω. (4.8)

However, this relation is no longer valid for the magnitude data, unless allresonances are in-phase with respect to each other. Only in this case allsignal can be transferred to the real part Fr(t) by rephasing the spectrum,which amounts to

|F (0)| = |Fr(0)| =∫|fr(ω)| · dω =

∫|f(ω)| · dω. (4.9)

Only the strong coupling terms (depending on TE1−TE2) in the in-phasemagnetization expectation values (Eqs. (4.2, 4.3)) are taken into accountfor the integration (ϕ = π/2 · Λ · (2TE1 − TE)):

〈Ascy 〉 = 〈Bsc

y 〉 = −2Jδ2/Λ3 · sin (πJ · TE) sin(π

2Λ · TE

)cos (ϕ)

= F1,y(TE) · cos (ϕ) (4.10)

〈Ascx 〉 = −〈Bsc

x 〉 = −2J2δ/Λ3 · cos (πJ · TE) sin(π

2Λ · TE

)· cos (ϕ)

−2Jδ/Λ2 · sin (πJ · TE) sin(π

2Λ · TE

)· sin (ϕ)

= F1,x(TE) · cos (ϕ) + F2,x(TE) · sin (ϕ) (4.11)


Because of 〈Ascx 〉 = −〈Bsc

x 〉 the signal intensities have to be calculated sep-arately for the two spins. Due to the factor sin(ϕ) in the x-magnetizationamplitude, the strong coupling peaks in the positive (f1 > 0) and negative(f1 < 0) half-plane are out of phase with respect to each other and needto be integrated separately. This requires splitting up the trigonometricfactors cos(ϕ) and sin(ϕ) into exponential expressions:

cos (ϕ) =12

(exp(iϕ) + exp(−iϕ)) (4.12)

sin (ϕ) =12i

(exp(iϕ)− exp(−iϕ)) (4.13)

All terms with exp(iϕ) give rise to resonances in the positive half-plane,while all terms with exp(−iϕ) give rise to resonances in the negative half-plane. Assuming quadrature detection, the overall signal turns out to bethe sum of four contributions:

Isc =∣∣⟨Asc

−⟩∣∣

f1>0+

∣∣⟨Asc−

⟩∣∣f1<0

+∣∣⟨Bsc

−⟩∣∣

f1>0+

∣∣⟨Bsc−

⟩∣∣f1<0

=∣∣∣∣12

(F1,x − i(F2,x + F1,y)) · exp(iϕ)∣∣∣∣

+∣∣∣∣12

(F1,x + i(F2,x − F1,y)) · exp(−iϕ)∣∣∣∣

+∣∣∣∣12

(−F1,x + i(F2,x − F1,y)) · exp(iϕ)∣∣∣∣

+∣∣∣∣12

(−F1,x − i(F2,x + F1,y)) · exp(−iϕ)∣∣∣∣

=√

F 21,x + (F2,x + F1,y)2 +

√F 2

1,x + (F2,x − F1,y)2. (4.14)

Chapter 5

Quantitative J-ResolvedProstate SpectroscopyUsing Two-DimensionalPrior-Knowledge Fitting

(submitted to: Magnetic Resonance in Medicine79)

5.1 Introduction

Digital rectal examination, prostate-specific antigen (PSA) blood test andtransrectal ultrasound (TRUS) guided biopsy are standard methods forthe diagnosis and staging of prostate cancer. Over the past two decadesmagnetic resonance imaging (MRI) has been established as a supplemen-tary non-invasive method, identifying pathological tissue on the basis ofrelaxometrical information. However, like the afore-mentioned tools, MRIlacks sensitivity as well as specificity in clinical diagnosis, especially when itcomes to distinguishing prostate cancer from benign prostate hyperplasia(BPH), which is very common in older men. In particular, since the ad-vent of endorectal receiver coils, magnetic resonance spectroscopy (MRS)has emerged as another promising method providing additional metabolic

79

80 Quantitative J-Resolved Prostate Spectroscopy Using 2D Fitting

information for a clinical diagnosis.52 Proton MR spectra of cancerousprostate tissue exhibit largely reduced citrate (Cit) levels and increasedlevels of choline-containing compounds (t-Cho) as compared to healthy tis-sue.52,53 Thus the Cit/t-Cho concentration ratio has been used as a markerfor prostate cancer in a large number of clinical studies. Furthermore, exvivo NMR studies have identified spermine (Spm) and myo-inositol (MI)as additional potential metabolic markers for prostate cancer, althoughtheir patho-physiological role is not yet well understood.38,55,73,74,80,81

Two-dimensional (2D) spectroscopy methods at least partially resolvethe spectral overlap usually observed in one-dimensional spectra and hencefacilitate peak assignment and improve metabolite quantification. Thefeasibility of localized 2D J-resolved spectroscopy (JPRESS) in the humanprostate with an endorectal/pelvic phased array coil system was shown inseveral works.38,39,64 Swanson et al. demonstrated that five metabolites(Cit, creatine (Cr), choline (Cho), Spm and lactate (Lac)) can be observedin vivo with this method.38 Yue et al. identified the characteristic 2Dmultiplet patterns of Cit and Spm in JPRESS spectra with a reasonableresolution.39 Kim et al. combined the JPRESS method with spectroscopicimaging to identify the region of abnormal metabolism in patients withprostate cancer.64 Recently 2D S-PRESS was presented as an alternativetwo-dimensional method for prostate spectroscopy, particularly well suitedfor Cit detection.51 However, since relatively long echo times are requiredto obtain a reasonable spectral resolution in the indirect dimension, 2DS-PRESS is not the method of choice for the detection of metabolites withvery short T2 relaxation constants like Spm.

When there is a considerable overlap of resonances in a 1D spectrum,conventional quantification methods such as peak integration or simpleline fitting fail to yield reliable results. Over the years, more advancedquantification methods have been developed, using time-domain as well asfrequency-domain fitting procedures, which enable the inclusion of priorknowledge to fit spectra with a physically reasonable model.82,83 An ob-jective and widely used frequency-domain fitting method is LCModel,76

which incorporates the entire prior knowledge by fitting spectra as linearcombinations of basis spectra, which can be either simulated19 or acquiredin phantom experiments. As an additional benefit, LCModel yields qualityestimates for the determined concentration ratios in the form of Cramer-Rao lower bounds (CRLBs).84

The vast majority of prostate MRS studies apply simple peak inte-

5.1. INTRODUCTION 81

gration to the postprocessed spectra for metabolite quantification. Un-fortunately, there is a strong overlap of detectable metabolites in thechemical shift range between 3.0 and 3.3 ppm (Cr, t-Cho, MI, taurine(Tau) and polyamines (PA, mainly Spm)), which aggravates quantifica-tion by peak integration considerably. Therefore, the concentration ratiosCit/(Cr+t-Cho)55,56,70,85,86 or even Cit/(Cr+t-Cho+PA)87 were used asa patho-physiological marker for prostate cancer in many studies. An-other difficulty for metabolite quantification in the prostate arises fromthe fact that resonances of strongly coupled metabolites (e.g. Cit, MI,Tau, PA) undergo a signal intensity modulation, which makes concentra-tion ratios determined via peak integration echo-time dependent. Thiscomplicates the comparison of quantitative results obtained in MRS mea-surements at different field strengths and with different sequences. Peakfitting procedures can increase the accuracy and robustness of metabo-lite quantification in the prostate. Scheenen et al. applied a basic fittingalgorithm assuming Gaussian line-shapes to the spectra prior to integra-tion.88 Kurhanewicz et al. quantified prostate spectra by means of amodified PIQABLE algorithm,56,89 while Garcıa-Segura et al. used a fit-ting routine based on least-squares optimization.55 Alternatively spectracan be fitted in the time domain, e.g., using the AMARES algorithm.86 Adetailed comparison of a frequency domain analysis based on peak integra-tion and time domain fitting based on AMARES for prostate spectroscopycan be found elsewhere, suggesting that the time domain fitting methodis more accurate.87 However, even with these 1D fitting methods metabo-lite quantification of in vivo prostate spectra is usually limited to Cit, Crand t-Cho. High-resolution ex vivo studies reported a much larger num-ber of detectable and quantifiable metabolites, also including Lac, alanine(Ala), glutamate (Glu), glutamine (Gln), scyllo-inositol (Scy), MI, Tauand PA.52,73,80,81,90–92

Since the extra information contained in 2D spectra promises greaterrobustness and higher quantification accuracy, several attempts have beenundertaken to adapt established 1D fitting methods to two dimensions. DeBeer et al.93 suggested a 2D Hankel singular value decomposition (HSVD)method, while Slotboom et al.94 developed a pseudo-2D technique basedon constrained 1D fits. Recently Schulte et al.49 proposed a 2D fittingprocedure dubbed ProFit (Prior-Knowledge Fitting), which fits JPRESSspectra as linear combinations of 2D basis spectra, thus incorporating themaximum prior knowledge available. Like LCModel, ProFit combines a


non-linear least-squares algorithm, which optimises the model parameters(chemical shifts, line widths, phases and line shapes), with a linear least-squares algorithm for the determination of the metabolite concentrations.The fitting method was validated for brain spectroscopy in the parietaland frontal lobe and proved to be more accurate than LCModel fitting of1D spectra acquired from the same volume with equal scan time.

In this work we adapt the ProFit procedure for JPRESS spectroscopyin the human prostate and show that via 2D prior knowledge fitting sixmetabolites can be quantified with reasonable precision. The method iscompared with LCModel fitting of a PRESS spectrum acquired from thesame volume and validated in ten repetitive JPRESS measurements in thesame subject as well as in a study on ten healthy subjects.

5.2 Materials and Methods

All prostate spectra were measured on a Philips Achieva 3T system (PhilipsMedical Systems, Best, The Netherlands) with a two-element ellipsoidalsurface coil (semi-major axis = 17 cm, semi-minor axis = 14 cm), whilethe body coil was used for signal excitation. The 2D J-resolved spectrawere acquired with a PRESS sequence (90◦x− [TE1/2]−180◦y− [TE1/2+TE2/2]− 180◦y − [TE2/2]−Acq), where the first echo time TE1 was heldconstant, while the second echo time TE2 and therewith the total echotime TE were varied for encoding the coupling information in the indirectdimension (f1). The water signal was suppressed with selective excitationin conjunction with gradient spoiling prior to the PRESS sequence. Amaximum-echo sampling scheme was applied, i.e., the acquisition starteddirectly after the second refocusing pulse for each encoding step.20 Apartfrom the SNR gain, this acquisition scheme leads to a tilt of the truncation-induced peak tails in the 2D spectrum, which prevents the residual waterpeak from spoiling the spectral region with the metabolites of interest. TEwas varied in the range between 49 and 447 ms, covering 100 equidistantencoding steps with ∆TE = 4 ms. This gives rise to a bandwidth of 250 Hzand a nominal spectral resolution of 2.5 Hz in f1. In the direct dimension(f2), 1024 samples were acquired covering a bandwidth of 2000 Hz. Withan eight-step phase-cycling scheme and a repetition time TR = 1.7 s, thisamounts to an overall scan duration of about 23 min. The voxel was placedin the middle of the prostate covering predominantly central gland tissue(Fig. 5.1). The voxel size had to be adjusted to the shape of the prostate


to exclude periprostatic lipid signal and was varied in the range from 5.3to 5.8 ml. Postprocessing of the JPRESS spectra prior to fitting includedfrequency shifts in f1 and f2 as well as zeroth-order phase correction.

The ProFit algorithm consists of a non-linear and a linear optimizationloop. The constrained non-linear algorithm optimizes the following globalparameters: zeroth-order phase ϕ0, Gaussian line-broadening σg in f2, afrequency shift ∆1 in f1 and a bi-exponential phase decay θ(t2) model-ing line shape distortions due to eddy currents. Individual fit parameters(specific for each metabolite m) in the non-linear optimization are: anexponential line broadening σe,m due to T2 decay in f1 and f2 and a fre-quency shift ∆2,m in f2. The metabolite concentrations cm are determinedby linear least-squares optimization based on the Moore-Penrose pseudoin-verse. The measured spectrum S is updated with ∆1, ϕ0 and θ(t2) in eachiteration of the non-linear optimization, yielding Su, while the metabolitebasis spectra Bm are updated with ∆2,m, σe,m, σg, yielding Bu

m. For fur-ther mathematical and algorithmical details the reader is referred to theoriginal ProFit publication.49 The cost function for the non-linear opti-

Figure 5.1: Experimental setup for the prostate experiments: Asingle voxel (measuring approximately 16 mm × 23 mm × 16 mm)was placed in the centre of the prostate, excluding periprostatic fattissue.


mization contains the fit residue as the difference between the measuredand the fitted spectrum and regularization terms constraining the expo-nential line broadenings and chemical shifts of the fitted metabolites tophysically reasonable values:

fC(x) =∑

k,l

(Re{Suk,l(∆1, ϕ0, θl)}

−∑m

cmRe{Buk,l,m(∆2,m, σe,m, σg)})2

+ρσe

∑m

(σe,m − σρe,m)2 + ρ∆2var(∆2,m). (5.1)

The indices k and l run over all data points in the fit region along t1 andt2, respectively, while the index m runs over all basis metabolites; ρσe andρ∆2 denote regularization factors for the exponential line broadenings andthe frequency shifts in f2. The cost function was optimised for the spec-tral region {(f1, f2)| − 28 Hz < f1 < 28 Hz ∧2.2 ppm < f2 < 4.1 ppm∧f2 − 2 · f1 < 4.3 ppm}, to exclude residual water and lipid signal andto reduce the numerical burden. Thus, the measured JPRESS spectrumand the basis spectra were cropped to a matrix size of 64 × 164. Sincethe T2 relaxation constants of the individual prostate metabolites varywithin one order of magnitude, the exponential line broadenings have tobe regularized with metabolite-specific values σρ

e,m = 1/πT2,m. This im-plies an extension to the original ProFit implementation where one singleregularization constant for all metabolites was used.49

The metabolite basis spectra can be either simulated or acquired frompure phantom solution. The advantage of measured basis spectra is thatthey account for most measurement imperfections in the spectrum underanalysis. However, J coupling constants and relative chemical shifts incoupled metabolites depend on the chemical environment and can thereforebe significantly different in vitro and in vivo, giving rise to a different spinevolution. Especially the coupling constant and chemical shift differencein the AB spin system of Cit exhibit a relatively strong dependence on theconcentration of divalent ions.68 This can be particularly severe for longecho times as used in JPRESS experiments and impair quantitation withbasis spectra considerably. Therefore, in this work, the basis spectra weresimulated numerically with the GAMMA library,95 using chemical shiftsand coupling constants from literature.51,96,97

Literature suggests that the following metabolites can be detected and


quantified ex vivo from healthy prostatic tissue:52,73,90,91 Lac, Ala, Cit,Cr, PA, free choline (Cho), phosphorylcholine (PCh) + glycerophosphoryl-choline (GPC), Scy, MI, Tau, Glu, Gln, leucine (Leu), acetic acid, N-acetylneuraminic acid. However, for the sake of robustness, the fit should havea restricted number of degrees of freedom. Therefore the basis set shouldnot be too large and contain too many similar metabolites. The resonancesfrom acetic acid and N-acetyl neuraminic acid are too weak to be detectedin vivo, while the Lac and Leu resonances lie outside the chosen fit re-gion. Ala gives rise to a doublet at 1.48 ppm (outside the fit region) and aquartet at 3.77 ppm, which failed to yield a significant contribution in aninitial ProFit test run. The PA signal contains contributions from severalmetabolites (mainly Spm, but also spermidine and putrescine), which giverise to very similar resonance patterns and frequencies.97 Therefore, inthis work, the PA signal was only fitted with a basis spectrum from Spm.Since PCh and GPC have almost identical resonance frequencies in thespectral region around 3.2 ppm, only a PCh basis spectrum was includedin the basis set. A set of ten basis metabolites turned out to be the mostsuitable choice for fitting JPRESS prostate spectra acquired in vivo: Cr,PCh, Cit, Spm, Cho, MI, Scy, Tau, Glu and Gln.

To enhance the efficiency and robustness of the non-linear fitting algo-rithm, it is iterated three times with increasing degrees of freedom. Theinitial iteration step only fits the dominant metabolites Cr, PCh, Cit andSpm. In the second iteration a line-shape distortion function is added,while in the final iteration the spectrum is fitted with all ten basis metabo-lites. The fit parameters of each iteration are initialized with the resultsfrom the previous iteration.

No absolute quantification was attempted, but metabolite concentra-tions were determined as ratios to Cr. Since relaxation effects were ne-glected in the simulation of the basis spectra the obtained metabolite con-centration ratios had to be corrected for T1 and T2 decay:

(cmet

cCr

)

corr

=(

cmet

cCr

)

fit

· 1− exp (−TR/T1,Cr)1− exp (−TR/T1,met)

· exp (−TEmin/T2,Cr)exp (−TEmin/T2,met)

. (5.2)

The T1 and T2 relaxation constants of prostate metabolites were obtainedfrom literature.91 Since the relaxation constants for Glu and Gln have notbeen measured in prostate tissue yet, they were estimated to be: T1 =


1200 ms, T2 = 600 ms. The T2 constants were also used for the regular-ization in the cost function (Eq. 5.1).

For comparison, measuring one subject, a short echo time PRESS spec-trum was acquired from the same volume with the same total scan timeand quantified using LCModel. The basis set used by LCModel containedthe same metabolites as the one used by ProFit and the basis spectra werealso simulated with the GAMMA library.

To investigate the reproducibility of the method, one subject was repet-itively measured in different sessions over a period of several weeks. Somespectra had to be discarded due to extremely poor spectral quality, butten data sets were fitted with ProFit and included in the statistical anal-ysis. Furthermore, average metabolite concentration ratios in the humanprostate were determined in a study on healthy subjects (between 20 and46 years old), who gave written informed consent according to the rules ofthe Local Ethics Committee (EK: 09/2006 (ETH)). Again, ten data setshad a sufficient spectral quality for 2D fitting and quantification.

5.3 Results

Figure 5.2 shows a typical JPRESS prostate spectrum, the fitted spec-trum and the fit residue. Besides the prominent Cit multiplet at 2.6 ppm,there is a large number of hardly distinguishable coupled and uncoupledresonances in the spectral region between 3.0 and 3.3 ppm. At 2.05 ppm,a broad resonance with a strong inter-subject amplitude fluctuation wasobserved. It was attributed to lipid signal from outside the prostate dueto imperfect volume selection and therefore not included in the fit region.However, the human prostate is a very difficult anatomical region for spec-troscopy experiments with surface coils. For some measured subjects, nei-ther PRESS nor JPRESS spectra showed any metabolite peaks apart fromwater and fat and therefore had to be excluded from the study. This ismainly due to considerable B0 and B1 inhomogeneities in the prostate,which can give rise to bad shimming and compromised power optimisa-tion. Apart from that, motion artifacts are far more severe than in otherorgans due to the small size of the prostate and can degrade the spectralquality considerably. The sensitivity of the method strongly depends onthe build of the subject, but also the positioning of the receiver coils turnedout to be extremely critical.

Figure 5.3 shows the PRESS spectrum acquired from the same volume

5.3. RESULTS 87

spectrum

22.533.54

−20

0

20

−0.5

0

0.5!t

22.533.54

−20

0

20

−0.5

0

0.5

residue

22.533.54

−20

0

20

−0.5

0

0.5

Cr

MI Cho/PCh Cr

Scy Spm Cit

f1 [Hz]

f2 [ppm]

f2 [ppm]

f2 [ppm]

f1 [Hz]

f1 [Hz]

Figure 5.2: JPRESS prostate spectrum (top), its fit (middle) andthe fit residue (bottom) as logarithmically scaled colour plots de-termined with ProFit. The white box defines the spectral region ofinterest where the cost function of the fit was minimised.


Chemical Shift (ppm) 4.2 4.0 3.8 3.6 3.4 3.2 3.0 2.8 2.6 2.4 2.2 2.0 1.8 1.6 1.4 1.2 1.0 0.80 0.60 0.40

0

0.20

Cr

Cit

Cr

Cho/PCh

Lipids and Macromolecules

Baseline

Figure 5.3: PRESS prostate spectrum fitted with LCModel. Thesame basis metabolites were used as for the ProFit analysis (Fig.5.2). The bold red line indicates the fit, while the fit residue isshown above the fitted spectrum.

5.3. RESULTS 89

JPRESS/ProFit PRESS/LCModel

Metabolite Conc. [/Cr] CRLB [%] Conc. [/Cr] CRLB [%]Cr 1.00 5.1 1.00 7.0

PCh 0.43 7.0 0.31 17.0Cit 9.78 1.2 5.42 3.0Spm 5.68 4.8 0.72 36.0Cho 0.15 14.2 0.04 102.0MI 1.94 8.0 1.14 18.0Scy 0.16 14.6 0.00 -Tau 0.56 24.9 0.05 195.0Gln 0.00 - 1.49 34.0Glu 0.84 18.8 0.00 -

t-Cho 0.58 3.7 0.35 7.0Glx 0.84 22.9 1.49 34.0

Table 5.1: Metabolite concentrations (as ratios to Cr) in theprostate of one healthy subject. Both JPRESS and PRESS spec-tra were acquired subsequently from the same volume with identicalscan time and quantified with ProFit and LCModel, respectively.Fit quality measures for individual metabolites were determined inthe form of CRLBs.

as the JPRESS spectrum (Fig. 5.2) and fitted with LCModel. Since thewater suppression is poor, the baseline of the PRESS spectrum is heav-ily distorted, which is accounted for by LCModel through fitting withspline functions. In contrast to that, the spectral region of interest in theJPRESS spectrum is hardly affected by the poor water suppression, dueto the maximum-echo sampling scheme. Table 5.1 shows the quantitativefit results obtained by ProFit and LCModel, fitting the JPRESS (Fig. 5.2)and the PRESS (Fig. 5.3) spectrum, respectively. Error margins were cal-culated as CRLBs from the Fisher information matrix.84 Due to theirmutual overlap, Cho and PCh as well as Glu and Gln have large CRLBsand can therefore be more reliably quantified as sums Glx = Glu + Glnand t-Cho = Cho + PCh.

The metabolite ratios turn out to be quite different (e.g. for Spm).Generally, ProFit yields much smaller CRLBs than LCModel, especially


Metabolite Conc. [/Cr] CRLB [%] Error [%] SD [%]Cr 1.00 6.5 - -

t-Cho 0.58 5.4 8.5 20.4Cit 10.48 1.7 6.7 21.8Spm 5.98 6.3 9.1 21.2MI 1.87 10.8 12.6 13.9Scy 0.18 16.2 17.4 11.7

Table 5.2: Metabolite concentrations in the prostate of one healthysubject (as ratios to Cr), determined with ProFit, as averagesover ten JPRESS measurements in different sessions. Average er-ror margins for the ratios to Cr were calculated from the averageCRLBs via error propagation. The standard deviations of the me-tabolite ratios are shown in the last column.

for coupled metabolites such as Cit and Spm. This is due to the in-creased information content and reduced peak overlap in the 2D spectrum.While the Spm resonance can hardly be recognized in the PRESS spectrum(Fig. 5.3), it appears as a characteristic multiplet pattern in the JPRESSspectrum (broad negative peaks at f1 ≈ ±6 Hz in Fig. 5.2) and can thusbe distinguished more easily from other uncoupled (Cr, Cho/PCh) andcoupled (MI, Tau) resonances in the spectral region around 3.15 ppm.This also explains the large discrepancy between the Spm/Cr ratios de-termined by ProFit and LCModel. Furthermore, the consistently lowerconcentration ratios obtained by LCModel suggest an overestimation ofthe Cr concentration due to the overlap with other resonances.

Table 5.2 shows the average concentration ratios (with average CRLBsfor individual metabolites) from ten JPRESS/ProFit experiments in thesame subject. Confidence intervals for the ratios were calculated from theCRLBs via error propagation. In the last column, the standard deviations(SDs) of the ratios are listed, indicating the reproducibility of the method.Only quantitation results with CRLBs below 20 % are commonly consid-ered reliable.98 Therefore Cho, Tau, Gln, Glu and Glx with huge CRLBsin most fits, were excluded from the list. Since Cho and PCh can be morereliably quantified together, only the t-Cho concentration is listed and ad-dressed in the further discussions. Six metabolites yielded average CRLBs

5.3. RESULTS 91

Metabolite Conc. [/Cr] CRLB [%] Error [%] SD [%]Cr 1.00 6.3 - -

t-Cho 0.61 4.6 7.8 14.7Cit 12.49 1.4 6.4 38.9Spm 6.88 5.1 8.2 37.4MI 2.11 9.3 11.2 22.6Scy 0.15 18.9 19.9 22.1

Table 5.3: Metabolite concentrations in healthy prostate tissue (asratios to Cr) determined with ProFit and averaged over ten healthysubjects. Average error margins for the ratios to Cr were calculatedfrom the average CRLBs via error propagation. The inter-subjectstandard deviations of the metabolite ratios are shown in the lastcolumn.

below 20 % (Cr, Cit, Spm, MI, Scy and t-Cho). For MI and Scy the SDsare comparable to the average error margins calculated from the CRLBs,whereas for Cit, Spm and t-Cho the SDs are considerably higher. Thisdiscrepancy suggests that the VOI placement might have been slightlydifferent for the singular measurements and there might have been nat-ural fluctuations in metabolite ratios between measurements. However,imperfections in the fitting model have to be taken into account as well.

Table 5.3 shows the determined metabolite concentration ratios aver-aged over ten subjects, with average CRLBs and average error margins.Additionally, the inter-subject SDs of the metabolite ratios are shown inthe last column. For Cit and Spm, the inter-subject SD is many timeslarger than the error estimate calculated from the CRLBs, while for t-Cho, MI and Scy error margins and SDs only differ by a factor of up to 2.This can be partly explained with varying amounts of glandular and stro-mal tissue in the VOI. The Cit and Spm concentrations in glandular tissueare many times larger than those found in stromal tissue, while the con-centrations of the other metabolites (Cr, t-Cho, MI, Scy) vary only slightlywith the tissue type.91 The calculated error margins in Table 5.3 are verysimilar to those in Table 5.2, but the inter-subject SDs are about twice aslarge as the individual SDs. This corroborates the finding of considerableinter-subject variations in metabolite concentration ratios.


5.4 Discussion

Quantitative prostate studies based on peak integration as carried out bymany groups over the past decade suffer from subjectiveness and a heavyoverlap of metabolite peaks in the spectral region between 3.0 and 3.3 ppm,rendering the comparison of quantitative results obtained at different siteswith different sequences difficult. Furthermore, strongly coupled reso-nances show a signal intensity modulation with echo time. This obviatesthe need for an objective, user-independent fitting method like LCModel,which includes prior knowledge in the form of metabolite basis spectra.However, especially for coupled metabolites (e.g. Cit, Spm and MI), thespectral information content can be considerably increased through 2Dspectroscopy methods like JPRESS, which enhance the orthogonality ofthe basis spectra and thus the accuracy of quantitation results. This isdemonstrated by comparing the quantitation results of JPRESS/ProFitwith those obtained using PRESS/LCModel from the same volume in thesame scan time. In particular for coupled metabolites, the CRLBs of theProFit results are much smaller than those of the LCModel results.

It is not surprising that there is a large difference in metabolite concen-trations determined in this work in comparison with other in vivo and invitro prostate studies, where simple peak integration was applied for quan-tification.52,55,81,86,90–92,99–102 Compared to results obtained by Swansonet al.91 through quantification of 1D in vitro spectra, the proposed methodyields similar concentration ratios for t-Cho/Cr (this work: 0.61, Swansonet al.: 0.78) and Scy/Cr (this work: 0.15, Swanson et al.: 0.2), whereasCit/Cr (this work: 12.49, Swanson et al.: 4.75), MI/Cr (this work: 2.11,Swanson et al.: 3.77) and PA/Cr (this work: 6.88, Swanson et al.: 2.03)are completely different. This demonstrates that for uncoupled resonances(Cr, t-Cho, Scy) quantification via simple line fitting and peak integra-tion works reasonably well, whereas for coupled metabolites the resultsare strongly distorted by J modulation, which leads to considerably de-creased signal intensity at certain echo times. J modulation is taken intoaccount when the spectrum is fitted using metabolite basis spectra (asdone in this work), but it is not considered when quantifying via simpleconstrained peak fitting, which renders quantification sequence and echotime dependent. Consequently, the Cit/t-Cho concentration ratio, whichwas found to be around 20 in healthy glandular tissue, is also much larger.

However, the regularisation parameters used in the cost function (Eq.

5.4. DISCUSSION 93

5.1), in particular ρσe , have a considerable influence on the fitting results.The regularization constants ρσe

and ρ∆2 were chosen such that the devi-ation of T2 relaxation constants and chemical shifts from literature valuesdid not become excessively large. Chemical shifts and coupling constantshave been well investigated in brain, but not in prostate tissue. There-fore literature values determined in brain tissue96 have been taken for thesimulation of the basis spectra except for Cit and Spm. For this reason,a divergence of a few Hz in the ∆2,m values was tolerated in the fit. Theexponential broadening factors σe,m were regularised such that their devi-ation from σρ

e,m stayed approximately in the range of the error estimatesin.91 However, the regularization underlies a certain subjectiveness. Thiscan give rise to systematic quantification errors beyond the range of theCRLBs, which assume a perfect fitting model.

Since prostate cancer usually occurs in the peripheral zone of theprostate, single voxel measurements as performed in this work are not ap-propriate for clinical diagnosis. However, the signal-to-noise ratio can beraised considerably by using an endorectal/pelvic phased array coil systemfor signal reception,39 hence enabling a further increase in quantificationaccuracy. This should enable much smaller voxels or a combination ofJPRESS with spectroscopic imaging64 in order to obtain spatial distribu-tions of metabolites over the whole prostate.

Conclusions and Outlook

The objective of this thesis was to improve the detectability of J-coupledmetabolites and to enhance the accuracy of quantitative measurements.Due to their complicated coupling evolution, strongly coupled spin systemsare usually a greater challenge for spectroscopists than weakly coupledones. However, anomalous J modulation, discussed in Chapter 2, is mostsevere for weakly coupled protons with large chemical shift differences suchas in the lactate molecule. The resulting signal loss approximately riseswith B2

0 , becoming severe for B0 ≥ 3 T. Whole-body 3T MR systemsare no longer limited to research facilities, but have also arrived in manyhospitals where clinical studies are carried out with standard protocolsand without access to more technical sequence parameters like selectiongradients as well as pulse shapes and bandwidths. For lactate detection inclinical environments, the most obvious remedy is therefore the adaptationof the echo time, which should be either 288 ms or as short as possible.

Among the 2D spectroscopy methods used on human MR systems,JPRESS has proven to be most robust and versatile. Its applicability wasdemonstrated in Chapter 3 for brain spectroscopy and in Chapter 5 forprostate spectroscopy. The proposed improvements such as the maximum-echo sampling scheme, which increases the SNR and reduces baseline dis-tortions through imperfect water suppression, are essential for spectro-scopic applications in difficult anatomical regions like prostate, kidney orliver, where sensitivity and residual nuisance signal are a major concern.Such an optimised JPRESS sequence is also an excellent experimental ba-sis for fitting and metabolite quantification, as has been demonstrated forbrain spectroscopy49 and for prostate spectroscopy (Chapter 5). The addi-tional information for J-coupled metabolites contained in JPRESS spectracan best be exploited by fitting and quantification with 2D basis spec-

95

96 Conclusions and Outlook

tra incorporating the whole prior knowledge available. This enables thedetermination of objective metabolite concentration ratios in the humanprostate where quantitative measurements are normally based on simplepeak integration. Furthermore, the confidence intervals provided by ProFitin the form of CRLBs are an asset for the objective evaluation of quanti-tative studies. The introduction of such a method as a golden standard inclinical prostate studies could facilitate the comparison of results obtainedat different sites and with different measurement parameters. The appli-cation of the ProFit algorithm for the quantification of other 2D spectrasuch as COSY or 2D S-PRESS is readily feasible with minor adjustmentsof the fitting model (e.g. the line-shape in the indirect dimension).

While JPRESS is the method of choice when the simultaneous de-tection and quantification of a large number of J-coupled metabolites isaspired, 2D S-PRESS has been designed for the detection and investiga-tion of strongly coupled metabolites such as citrate. The more clearlyarranged spectral pattern of citrate in 2D S-PRESS spectra as comparedto JPRESS spectra can be exploited to determine the spectral parame-ters of citrate (chemical shift difference δ and coupling constant J) in vivowith reasonable accuracy (Chapter 4). But MRS not only reveals infor-mation about the concentrations of MR-active metabolites, but also abouttheir chemical environment (e.g. the solvent or certain ion concentrations).Therefore the spectral evolution and appearance of the citrate multipletcan be quite different in vitro and in vivo, as was demonstrated with the2D S-PRESS method in Chapter 4. Hence citrate basis spectra for meta-bolite quantification using LCModel or ProFit should be simulated ratherthan measured in phantom solution as usually recommended.

In most clinical applications of MRS, in particular for prostate exam-inations, the spatial distribution of certain metabolites is of interest, e.g.to identify cancerous or necrotic tissue in a tumor region. Therefore thecombination of 2D spectroscopy with spectroscopic imaging methods mightplay a major role in future high-field applications.42,64,103 To keep scantime within acceptable limits, fast SI sequences such as parallel spectro-scopic imaging,104,105 echo planar spectroscopic imaging (EPSI),106 spec-troscopic GRASE107 and Spectroscopic RARE108 are potential methodsof choice for such a combination.

The presented methods focused on 1H spectroscopy at a field strengthof 3 T, which has become more and more available for clinical purposesover the past few years. Human MR systems operating at 7 T are cur-

Conclusions and Outlook 97

rently being installed at many research sites. This opens up a wide rangeof new applications benefiting from the increased SNR and spectral reso-lution, but also exacerbates common problems encountered on human MRsystems such as B0 and B1 inhomogeneities, SAR limitations and chem-ical shift displacement artifacts. The latter can become so severe thatthe selected volumes for spins with Larmor frequencies which are furtherapart in the spectrum hardly overlap any more in PRESS experiments.Consequently, signal loss due to anomalous J-modulation, as discussedin Chapter 2 for the lactate resonance, will result in considerable signalmisregistration for other weakly coupled metabolites with much smallerchemical shift differences than observed in lactate. This also has to betaken into account for 2D spectroscopy methods, which are commonlybased on PRESS localisation, e.g. JPRESS and 2D S-PRESS. Apart fromover-prescribed PRESS,109 LASER12(or SADLOVE13) and STEAM57 ba-sically turn out to be the most promising localisation methods at ultra-high field strengths. In a LASER sequence, refocusing and slice selectionare achieved with two consecutive adiabatic full passage (AFP) pulses in-stead of one non-adiabatic pulse. To achieve localisation with adiabaticpulses only, a LASER sequence applies a spatially non-selective excita-tion pulse and three pairs of slice-selective AFP pulses. Apart from thefact that the signal is refocused three instead of two times, the coherencepathway is identical to that of a PRESS sequence so that PRESS canbe readily replaced by LASER localisation in JPRESS and 2D S-PRESSsequences. However, the six adiabatic inversion pulses entail a consider-able RF power deposition, which can necessitate longer repetition times.In a STEAM sequence, consisting of three 90◦ pulses, a stimulated echois acquired, which has a different coherence history than ordinary spinechoes.7 For J-encoding the indirect spectral dimension, the second andthird pulse would have to be shifted simultaneously, therewith varyingTE, but keeping the mixing time (TM) constant. However, zero-quantumcoherences during the mixing time, which are not filtered out by the selec-tion gradients, can give rise to a strong TM -dependent signal modulationfor J-coupled metabolites, which would have to be eliminated by addi-tional pulse/gradient elements during TM .7,110 Constant-time PRESS(CT-PRESS)43,44 might also gain popularity at field strengths above 3 T.It is based on the same sequence as JPRESS, but is reconstructed dif-ferently. While in a JPRESS reconstruction the acquired echo signals arealigned along the echo top in t1 before Fourier transformation, a fixed time

98 Conclusions and Outlook

point with respect to the excitation is used as reference for a CT-PRESSreconstruction. Thus the indirect dimension primarily displays the chem-ical shift instead of the coupling information. While the J-dispersion inthe indirect dimension of a JPRESS spectrum is independent of B0, CT-PRESS benefits from high field strengths through the increased chemicalshift range of resonances in both spectral dimensions. It remains to beseen which of these prospective 2D high-field applications prevail.

References

1 A.H. Trabesinger, D. Meier, U. Dydak, R. Lamerichs, and P. Boesiger.Optimizing PRESS localized citrate detection at 3 Tesla. Magn ResonMed, 54(1):51–58, 2005.

2 P.S. Allen, R.B. Thompson, and A.H. Wilman. Metabolite-specificNMR spectroscopy in vivo. NMR Biomed, 10(8):435–444, 1997.

3 A.H. Trabesinger, D. Meier, and P. Boesiger. In vivo H-1 NMRspectroscopy of individual human brain metabolites at moderate fieldstrengths. Magn Reson Imaging, 21(10):1295–1302, 2003.

4 R.R. Ernst, B.G., and A. Wokaun. Principles of nuclear magneticresonance in one and two dimensions. Clarendon Press, Oxford, 1987.

5 O.W. Sørensen, G.W. Eich, M.H. Levitt, G. Bodenhausen, and R.R.Ernst. A product operator-formalism for the description of NMR pulseexperiments. Prog Nucl Magn Reson Spectrosc, 16:163–192, 1983.

6 T. Lange, U. Dydak, T.P.L. Roberts, H.A. Rowley, M. Bjeljac, andP. Boesiger. Pitfalls in lactate measurements at 3T. Am J Neuroradiol,27(4):895–901, 2006.

7 R.A. de Graaf. In Vivo NMR Spectroscopy: Principles and Techniques.John Wiley, Chichester, 1998.

8 D.A. Yablonskiy, J.J. Neil, M.E. Raichle, and J.J.H. Ackerman.Homonuclear J coupling effects in volume localized NMR spectros-copy: Pitfalls and solutions. Magn Reson Med, 39(2):169–178, 1998.

99

100 REFERENCES

9 D.A.C. Kelley, L.L. Wald, and J.M. Star-Lack. Lactate detection at3T: Compensating J coupling effects with BASING. J Magn ResonImaging, 9(5):732–737, 1999.

10 T.K.C. Tran, D.B. Vigneron, N. Sailasuta, J. Tropp, P. Le Roux,J. Kurhanewicz, S. Nelson, and R. Hurd. Very selective suppressionpulses for clinical MRSI studies of brain and prostate cancer. MagnReson Med, 43(1):23–33, 2000.

11 J. Star-Lack, D. Spielman, E. Adalsteinsson, J. Kurhanewicz, D.J.Terris, and D.B. Vigneron. In vivo lactate editing with simultaneousdetection of choline, creatine, NAA, and lipid singlets at 1.5 T usingPRESS excitation with applications to the study of brain and headand neck tumors. J Magn Reson, 133(2):243–254, 1998.

12 M. Garwood and L. DelaBarre. The return of the frequency sweep:designing adiabatic pulses for contemporary NMR. J Magn Reson,153(2):155–177, 2001.

13 J. Slotboom, A.F. Mehlkopf, and W.M.M.J. Bovee. A single-shot lo-calization pulse sequence suited for coils with inhomogeneous RF fieldsusing adiabatic slice-selective RF pulses. J Magn Reson, 95(2):396–404, 1991.

14 F. Schick, T. Nagele, U. Klose, and O. Lutz. Lactate quantificationby means of PRESS spectroscopy: influence of refocusing pulses andtiming scheme. Magn Reson Imaging, 13(2):309–319, 1995.

15 W.I. Jung, M. Bunse, and O. Lutz. Quantitative evaluation of thelactate signal loss and its spatial dependence in PRESS localized H-1NMR spectroscopy. J Magn Reson, 152(2):203–213, 2001.

16 J. Wild I. Marshall. A systematic study of the lactate line-shape inPRESS-localized proton spectroscopy. Magn Reson Med, 40(1):72–78,1998.

17 R.B. Thompson and P.S. Allen. Sources of variability in the responseof coupled spins to the PRESS sequence and their potential impact onmetabolite quantification. Magn Reson Med, 41(6):1162–1169, 1999.

18 J. Slotboom, A.F. Mehlkopf, and W.M.M.J. Bovee. The effects offrequency-selective RF pulses on J-coupled spin-1/2 systems. J MagnReson, 108(1):38–50, 1994.

REFERENCES 101

19 A.A. Maudsley, V. Govindaraju, K. Young, Z.K. Aygula, P.M. Pat-tany, B.J. Soher, and G.B. Matson. Numerical simulation of PRESSlocalized MR spectroscopy. J Magn Reson, 173(1):54–63, 2005.

20 R.F. Schulte, T. Lange, J. Beck, D. Meier, and P. Boesiger. Improvedtwo-dimensional J-resolved spectroscopy. NMR Biomed, 19(2):264–270, 2006.

21 W.P. Aue, E. Bartholdi, and R.R. Ernst. Two-dimensional spec-troscopy - application to nuclear magnetic resonance. J Chem Phys,64(5):2229–2246, 1976.

22 S. Macura and R.R. Ernst. Elucidation of cross relaxation in liquidsby two-dimensional NMR spectroscopy. Mol Phys, 41(1):95–117, 1980.

23 R. Kreis and C. Boesch. Liquid-crystal-like structures of human mus-cle demonstrated by in vivo observation of direct dipolar coupling inlocalized proton magnetic resonance spectroscopy. J Magn Reson SerB, 104(2):189–92, 1994.

24 L.N. Ryner, J.A. Sorenson, and M.A. Thomas. Localized 2D J-resolved H-1 MR spectroscopy - strong-coupling effects in vitro andin vivo. Magn Reson Imaging, 13(6):853–869, 1995.

25 G. C. McKinnon and P. Boesiger. Localized double-quantum filter andcorrelation spectroscopy experiments. Magn Reson Med, 6(3):334–343,1988.

26 W.P. Aue, J. Karhan, and R.R. Ernst. Homonuclear broad-band de-coupling and 2-dimensional J-resolved NMR-spectroscopy. J ChemPhys, 64(10):4226–4227, 1976.

27 P.A. Bottomley. Spatial localization in NMR spectroscopy in vivo.Ann N Y Acad Sci, 508:333–348, 1987.

28 P. Meric, G. Autret, B.T. Doan, B. Gillet, C. Sebrie, and J.C. Beloeil.In vivo 2D magnetic resonance spectroscopy of small animals. MagnReson Mater Phy Biol Med, 17(3-6):317–338, 2004.

29 L.N. Ryner, J.A. Sorenson, and M.A. Thomas. 3D localized 2D NMRspectroscopy on an MRI scanner. J Magn Reson Ser B, 107(2):126–37,1995.

102 REFERENCES

30 M.A. Thomas, L.N. Ryner, M.P. Mehta, P.A. Turski, and J.A. Soren-son. Localized 2D J -resolved 1H MR spectroscopy of human braintumors in vivo. J Magn Reson Imaging, 6(3):453–459, 1996.

31 R.E. Hurd, D. Gurr, and N. Sailasuta. Proton spectroscopy withoutwater suppression: The oversampled J -resolved experiment. MagnReson Med, 40(3):343–347, 1998.

32 Y. Ke, B.M. Cohen, J.Y. Bang, M. Yang, and P.F. Renshaw. Assess-ment of GABA concentration in human brain using two-dimensionalproton magnetic resonance spectroscopy. Psychiatry Res, 100(3):169–178, 2000.

33 L.M. Levy and M. Hallett. Impaired brain GABA in focal dystonia.Ann Neurol, 51(1):93–101, 2002.

34 M.A. Thomas, N. Hattori, M. Umeda, T. Sawada, and S. Naruse.Evaluation of two-dimensional L-COSY and JPRESS using a 3 TMRI scanner: From phantoms to human brain in vivo. NMR Biomed,16(5):245–51, 2003.

35 R. Hurd, N. Sailasuta, R. Srinivasan, D.B. Vigneron, D. Pelletier,and S.J. Nelson. Measurement of brain glutamate using TE-averagedPRESS at 3T. Magn Reson Med, 51(3):435–440, 2004.

36 Y. Ke, C.C. Streeter, L.E. Nassar, O. Sarid-Segal, J. Hennen, D.A.Yurgelun-Todd, L.A. Awad, M.J. Rendall, S.A. Gruber, A. Nason,M.J. Mudrick, S.R. Blank, A.A. Meyer, C. Knapp, D.A. Ciraulo, andP.F. Renshaw. Frontal lobe GABA levels in cocaine dependence: Atwo-dimensional, J -resolved magnetic resonance spectroscopy study.Psychiatry Res, 130(3):283–93, 2004.

37 R. Kreis and C. Boesch. Spatially localized, one- and two-dimensionalNMR spectroscopy and in vivo application to human muscle. J MagnReson Ser B, 113(2):103–18, 1996.

38 M.G. Swanson, D.B. Vigneron, T.K.C. Tran, N. Sailasuta, R.E. Hurd,and J. Kurhanewicz. Single-voxel oversampled J-resolved spectroscopyof in vivo human prostate tissue. Magn Reson Med, 45(6):973–980,2001.

REFERENCES 103

39 K. Yue, A. Marumoto, N. Binesh, and M.A. Thomas. 2D JPRESS ofhuman prostates using an endorectal receiver coil. Magn Reson Med,47(6):1059–1064, 2002.

40 P.J. Bolan, L. DelaBarre, E.H. Baker, H. Merkle, L.I. Everson, D. Yee,and M. Garwood. Eliminating spurious lipid sidebands in 1H MRS ofbreast lesions. Magn Reson Med, 48(2):215–222, 2002.

41 B.Kuhn, W. Dreher, D. Leibfritz, and M. Heller. Homonuclear uncou-pled 1H-spectroscopy of the human brain using weighted accumulationschemes. Magn Reson Imaging, 17(8):1193–201, 1999.

42 E. Adalsteinsson and D.M. Spielman. Spatially resolved two-dimensional spectroscopy. Magn Reson Med, 41(1):8–12, 1999.

43 W. Dreher and D. Leibfritz. Detection of homonuclear decoupled invivo proton NMR spectra using constant time chemical shift encoding:CT-PRESS. Magn Reson Imaging, 17(1):141–150, 1999.

44 R.F. Schulte, A.H. Trabesinger, and P. Boesiger. Chemical-shift-selective filter for the in vivo detection of J-coupled metabolites at3T. Magn Reson Med, 53(2):275–281, 2005.

45 S. Macura and L.R. Brown. Improved sensitivity and resolution intwo-dimensional homonuclear J-resolved NMR-spectroscopy of macro-molecules. J Magn Reson, 53(3):529–535, 1983.

46 G. Herigault, S. Zoula, C. Remy, M. Decorps, and A. Ziegler. Multi-spin-echo J-resolved spectroscopic imaging without water suppression:application to a rat glioma at 7 T. Magn Reson Mater Phy Biol Med,17(3-6):140–148, 2004.

47 K. Lymer, K. Haga, I. Marshall, and J. Wardlaw. Reproducibility of invitro GABA measurements using 2D J-resolved MRS. In Proceedingsof the 12th Annual Meeting of ISMRM, page 686, Kyoto, Japan, 2004.

48 U. Klose. In vivo proton spectroscopy in presence of eddy currents.Magn Reson Med, 14(1):26–30, 1990.

49 R.F. Schulte and P. Boesiger. Profit: two-dimensional prior-knowledgefitting of J-resolved spectra. NMR Biomed, 19(2):255–263, 2006.

104 REFERENCES

50 R.M. Dixon and J. Frahm. Localized proton MR spectroscopy of thehuman kidney in vivo by means of short echo time STEAM sequences.Magn Reson Med, 31(5):482–487, 1994.

51 T. Lange, A.H. Trabesinger, R.F. Schulte, U. Dydak, and P. Boesiger.Prostate spectroscopy at 3 Tesla using two-dimensional S-PRESS.Magn Reson Med, 56(6):1220–1228, 2006.

52 J. Kurhanewicz, M.G. Swanson, S.J. Nelson, and D.B. Vigneron. Com-bined magnetic resonance imaging and spectroscopic imaging approachto molecular imaging of prostate cancer. J Magn Reson Imaging,16(4):451–463, 2002.

53 J. Scheidler, H. Hricak, D.B. Vigneron, K.K. Yu, D.L. Sokolov, L.R.Huang, C.J. Zaloudek, S.J. Nelson, P.R. Carroll, and J. Kurhanewicz.Prostate cancer: Localization with three-dimensional proton MR spec-troscopic imaging - clinicopathologic study. Radiology, 213(2):473–480,1999.

54 L.C. Costello, R.B. Franklin, and P. Narayan. Citrate in the diagnosisof prostate cancer. Prostate, 38(3):237–245, 1999.

55 J.M. Garcia-Segura, M. Sanchez-Chapado, C. Ibarburen, J. Viano,J.C. Angulo, J. Gonzalez, and J.M. Rodriguez-Vallejo. In vivo protonmagnetic resonance spectroscopy of diseased prostate: Spectroscopicfeatures of malignant versus benign pathology. Magn Reson Imaging,17(5):755–765, 1999.

56 J. Kurhanewicz, D.B. Vigneron, H. Hricak, P. Narayan, P. Carroll, andS.J. Nelson. Three-dimensional H-1 MR spectroscopic imaging of thein situ human prostate with high (0.24-0.1-cm(3)) spatial resolution.Radiology, 198(3):795–805, 1996.

57 J. Frahm, K.D. Merboldt, and W. Hanicke. Localized proton spectros-copy using stimulated echoes. J Magn Reson, 72(3):502–508, 1987.

58 R.V. Mulkern, J.L. Bowers, S. Peled, R.A. Kraft, and D.S. Williamson.Citrate signal enhancement with a homonuclear J-refocusing modifi-cation to double-echo PRESS sequences. Magn Reson Med, 36(5):775–780, 1996.

REFERENCES 105

59 F. Schick, H. Bongers, S. Kurz, W.I. Jung, M. Pfeffer, and O. Lutz.Localized proton MR spectroscopy of citrate in vitro and of the humanprostate in vivo at 1.5T. Magn Reson Med, 29(1):38–43, 1993.

60 F. Schick, K. Straubinger, J. Machann, T. Nagele, M. Bunse, U. Klose,and O. Lutz. Sequence parameters of double spin-echo sequences affectquantification of citrate. Magn Reson Imaging, 14(6):663–672, 1996.

61 K. Straubinger, F. Schick, and O. Lutz. Pulse angle dependence ofdouble-spin-echo proton NMR spectra of citrate - theory and experi-ments. J Magn Reson Ser B, 110(2):188–194, 1996.

62 A.H. Wilman and P.S. Allen. The response of the strongly coupled ABsystem of citrate to typical H-1 MRS localization sequences. J MagnReson Ser B, 107(1):25–33, 1995.

63 M. van der Graaf, G.J. Jager, and A. Heerschap. Removal of the outerlines of the citrate multiplet in proton magnetic resonance spectra ofthe prostatic gland by accurate timing of a point-resolved spectroscopypulse sequence. Magn Reson Mater Phy Biol Med, 5(1):65–69, 1997.

64 D.H. Kim, D. Margolis, L. Xing, B. Daniel, and D. Spielman. Invivo prostate magnetic resonance spectroscopic imaging using two-dimensional J-resolved PRESS at 3 T. Magn Reson Med, 53(5):1177–1182, 2005.

65 M.J. Thrippleton, R.A.E. Edden, and J. Keeler. Suppression of strongcoupling artefacts in J-spectra. J Magn Reson, 174(1):97–109, 2005.

66 G. Gambarota, M. van der Graaf, D. Klomp, R.V. Mulkern, andA. Heerschap. Echo-time independent signal modulations usingPRESS sequences: A new approach to spectral editing of stronglycoupled AB spin systems. J Magn Reson, 177(2):299–306, 2005.

67 G. Gambarota, M. Van der Graaf, D. Klomp, J. van Asten, andA. Heerschap. S-PRESS: a novel approach to difference spectroscopyediting. In Proceedings of the 12th Annual Meeting of ISMRM, page307, Kyoto, Japan, 2004.

68 M. van der Graaf and A. Heerschap. Effect of cation binding on theproton chemical shifts and the spin-spin coupling constant of citrate.J Magn Reson Ser B, 112(1):58–62, 1996.

106 REFERENCES

69 L.E. Kay and R.E.D. McClung. A product operator description of ABand ABX spin systems. J Magn Reson, 77(2):258–273, 1988.

70 R.G. Males, D.B. Vigneron, O. Star-Lack, S.C. Falbo, S.J. Nelson,H. Hricak, and J. Kurhanewicz. Clinical application of BASING andspectral/spatial water and lipid suppression pulses for prostate can-cer staging and localization by in vivo 3D H-1 magnetic resonancespectroscopic imaging. Magn Reson Med, 43(1):17–22, 2000.

71 J. Star-Lack, S.J. Nelson, J. Kurhanewicz, L.R. Huang, and D.B. Vi-gneron. Improved water and lipid suppression for 3D PRESS CSI usingRF band selective inversion with gradient dephasing (BASING). MagnReson Med, 38(2):311–321, 1997.

72 T.W.J. Scheenen, G. Gambarota, E. Weiland, D.W.J. Klomp, J.J.Futterer, J.O. Barentsz, and A. Heerschap. Optimal timing for in vivoH-1-MR spectroscopic imaging of the human prostate at 3T. MagnReson Med, 53(6):1268–1274, 2005.

73 M. van der Graaf, R.G. Schipper, G.O. Oosterhof, J.A. Schalken, A.A.Verhofstad, and A. Heerschap. Proton MR spectroscopy of prostatictissue focused on the detection of spermine, a possible biomarker ofmalignant behavior in prostate cancer. Magn Reson Mater Phy BiolMed, 10(3):153–9, 2000.

74 M.J. Lynch and J.K. Nicholson. Proton MRS of human prostaticfluid: Correlations between citrate, spermine, and myo-inositol levelsand changes with disease. Prostate, 30(4):248–255, 1997.

75 G.J. Moore and L.O. Sillerud. The ph-dependence of chemical-shiftand spin-spin coupling for citrate. J Magn Reson Ser B, 103(1):87–88,1994.

76 S.W. Provencher. Estimation of metabolite concentrations from local-ized in-vivo proton NMR-spectra. Magn Reson Med, 30(6):672–679,1993.

77 A.H. Trabesinger, D.C. Mueller, and P. Boesiger. Single-quantum co-herence filter for strongly coupled spin systems for localized H-1 NMRspectroscopy. J Magn Reson, 145(2):237–245, 2000.

REFERENCES 107

78 G. Gambarota, V. Mlynarik, R. Gruetter, and T. Liptaj. A novelapproach to spectral editing of glutathione at 7 Tesla using echo-timeindependent signal modulations in PRESS sequence. In Proceedings ofthe 14th Annual Meeting of ISMRM, page 3058, Seattle, US, 2006.

79 T. Lange, R.F. Schulte, and P. Boesiger. Quantitative J-resolvedprostate spectroscopy using two-dimensional prior-knowledge fitting.submitted to Magn Reson Med, 2007.

80 L.L. Cheng, C.L. Wu, M.R. Smith, and R.G. Gonzalez. Non-destructive quantitation of spermine in human prostate tissue sam-ples using HRMAS H-1 NMR spectroscopy at 9.4 T. Febs Letters,494(1-2):112–116, 2001.

81 J. Kurhanewicz, R. Dahiya, J.M. Macdonald, L.H. Chang, T.L. James,and P. Narayan. Citrate alterations in primary and metastatic humanprostatic adenocarcinomas - H-1 magnetic-resonance spectroscopy andbiochemical-study. Magn Reson Med, 29(2):149–157, 1993.

82 S. Mierisova and M. Ala-Korpela. MR spectroscopy quantitation: areview of frequency domain methods. NMR Biomed, 14(4):247–259,2001.

83 L. Vanhamme, T. Sundin, P. Van Hecke, and S. Van Huffel. MRspectroscopy quantitation: a review of time-domain methods. NMRBiomed, 14(4):233–246, 2001.

84 S. Cavassila, S. Deval, C. Huegen, D. van Ormondt, and D. Graveron-Demilly. Cramer-Rao bounds: an evaluation tool for quantitation.NMR Biomed, 14(4):278–283, 2001.

85 A. Kumar and R.R. Ernst. Influence of nonresonant nuclei on NMRspin echoes in liquids and in solids. J Magn Reson, 24(3):425–447,1976.

86 F.A. van Dorsten, M. van der Graaf, M.R.W. Engelbrecht, G. vanLeenders, A. Verhofstad, M. Rijpkema, J. de la Rosette, J.O. Barentsz,and A. Heerschap. Combined quantitative dynamic contrast-enhancedMR imaging and H-1 MR spectroscopic imaging of human prostatecancer. J Magn Reson Imaging, 20(2):279–287, 2004.

108 REFERENCES

87 P. Pels, E. Ozturk-Isik, M.G. Swanson, L. Vanhamme, J. Kurhanewicz,S.J. Nelson, and S. Van Huffel. Quantification of prostate MRSI databy model-based time domain fitting and frequency domain analysis.NMR Biomed, 19(2):188–197, 2006.

88 T.W.J. Scheenen, D.W.J. Klomp, S.A. Roll, J.J. Futterer, J.O. Bar-entsz, and A. Heerschap. Fast acquisition-weighted three-dimensionalproton MR spectroscopic imaging of the human prostate. Magn ResonMed, 52(1):80–88, 2004.

89 S.J. Nelson and T.R. Brown. A method for automatic quantification ofone-dimensional spectra with low signal-to-noise ratio. J Magn Reson,75(2):229–243, 1987.

90 M.L. Schiebler, K.K. Miyamoto, M. White, S.J. Maygarden, andJ.L. Mohler. Invitro high-resolution H-1-spectroscopy of the humanprostate - benign prostatic hyperplasia, normal peripheral zone andadenocarcinoma. Magn Reson Med, 29(3):285–291, 1993.

91 M.G. Swanson, A.S. Zektzer, Z.L. Tabatabai, J. Simko, S. Jarso, K.R.Keshari, L. Schmitt, P.R. Carroll, K. Shinohara, D.B. Vigneron, andJ. Kurhanewicz. Quantitative analysis of prostate metabolites usingH-1 HR-MAS spectroscopy. Magn Reson Med, 55(6):1257–1264, 2006.

92 A.H. Fowler, A.A. Pappas, J.C. Holder, A.E. Finkbeiner, G.V. Dal-rymple, M.S. Mullins, J.R. Sprigg, and R.A. Komoroski. Differenti-ation of human prostate-cancer from benign hypertrophy by in vitroH-1-NMR. Magn Reson Med, 25(1):140–147, 1992.

93 R. de Beer, D. van Ormondt, and W.W.F. Pijnappel. Quantificationof 1-D and 2-D magnetic-resonance time domain signals. Pure ApplChem, 64(6):815–823, 1992.

94 J. Slotboom, C. Boesch, and R. Kreis. Versatile frequency domainfitting using time domain models and prior knowledge. Magn ResonMed, 39(6):899–911, 1998.

95 S.A. Smith, T.O. Levante, B.H. Meier, and R.R. Ernst. Computer-simulations in magnetic-resonance - an object-oriented programmingapproach. J Magn Reson Ser A, 106(1):75–105, 1994.

REFERENCES 109

96 V. Govindaraju, K. Young, and A.A. Maudsley. Proton NMR chemicalshifts and coupling constants for brain metabolites. NMR Biomed,13(3):129–153, 2000.

97 W. Willker, U. Flogel, and D. Leibfritz. A H-1/C-13 inverse 2D methodfor the analysis of the polyamines putrescine, spermidine and sperminein cell extracts and biofluids. NMR.

98 S.W. Provencher. LCModel Manual. URL: http://www.s-provencher.com/pages/lcmodel.shtml, 2005.

99 A. Heerschap, G.J. Jager, M. vanderGraaf, J.O. Barentsz, and S.H.J.Ruijs. Proton MR spectroscopy of the normal human prostate withan endorectal coil and a double spin-echo pulse sequence. Magn ResonMed, 37(2):204–213, 1997.

100 R. Kumar, M. Kumar, N.R. Jagannathan, N.P. Gupta, and A.K.Hemal. Proton magnetic resonance spectroscopy with a body coil inthe diagnosis of carcinoma prostate. Urol Res, 32(1):36–40, 2004.

101 G.P. Liney, L.W. Turnbull, and A.J. Knowles. In vivo magnetic res-onance spectroscopy and dynamic contrast enhanced imaging of theprostate gland. NMR Biomed, 12(1):39–44, 1999.

102 G.P. Liney, L.W. Turnbull, M. Lowry, L.S. Turnbull, A.J. Knowles,and A. Horsman. Proton MR T-2 maps correlate with the citrateconcentration in the prostate. Magn Reson Imaging, 15(10):1177–1186,1997.

103 J.E. Jensen, B. deB. Frederick, L. Wang, J. Brown, and P.F. Renshaw.Two-dimensional J-resolved spetroscopic imaging of GABA at 4 Teslain the human brain. Magn Reson Med, 54(4):783–788, 2005.

104 U. Dydak, M. Weiger, K.P. Pruessmann, D. Meier, and P. Boe-siger. Sensitivity-encoded spectroscopic imaging. Magn Reson Med,46(4):713–22, 2001.

105 U. Dydak, K.P. Pruessmann, M. Weiger J. Tsao, D. Meier, and P. Boe-siger. Parallel spectroscopic imaging with spin-echo trains. Magn Re-son Med, 50(1):196–200, 2003.

106 R.V. Mulkern and L.P. Panych. Echo planar spectroscopic imaging.Concepts Magn Reson, 13(4):213–237, 2001.

110 REFERENCES

107 W. Dreher and D. Leibfritz. A new method for fast proton spectro-scopic imaging: spectroscopic GRASE. Magn Reson Med, 44(5):668–672, 2000.

108 W. Dreher and D. Leibfritz. Fast proton spectroscopic imaging withhigh signal-to-noise ratio: spectroscopic RARE. Magn Reson Med,47(3):523–528, 2002.

109 R.A. Edden, M. Schar, A.E. Hillis, and P.B. Barker. Optimized de-tection of lactate at high fields using inner volume saturation. MagnReson Med, 56(4):912–917, 2006.

110 K.E. Cano, M.J. Thrippleton, J. Keeler, and A.J. Shaka. Cascaded z-filters for efficient single-scan suppression of zero-quantum coherence.J Magn Reson, 167(2):291–297, 2004.

Acknowledgements

Many people have contributed to this thesis and supported my work indifferent ways. My special thanks go to...

• Prof. Dr. Peter Boesiger, head of the Biophysics Division at theInstitute of Biomedical Engineering, for the opportunity to workin an excellent scientific environment. His confidence and carefulguidance provided the basis for the present work.

• Rolf Schulte, Andreas Trabesinger and Ulrike Dydak for the con-structive collaboration in the projects published in this thesis. Withtheir steady scientific and personal support, their knowledge andexperience they contributed considerably to the success of my dis-sertation.

• Dieter Meier and Urs Sturzenegger for the technical support theyprovided over all these years, constantly updating our MR scannerswith the newest hardware and software releases.

• Roger Luchinger and Bruno Willi for their help with numerous com-puter-related problems.

• Dr. Peter Sandor from the Neurology Department of the UniversityHospital Zurich for the collaboration in several clinical projects.

• Marianne Berg for helping me with many administrative issues.

• Prof. Dr. Roland Kreis for the co-examination of this thesis.

• all current and former colleagues, Alex, Andrea, Anke, Basti, Berti,Betina, Carolin, Christian, Christof, Christoph, Conny, David, Diane-Elise, Florian, Gerard, Hendrik, Jeff, Jochen, Jurek, Klaas, Malex,

111

112 Acknowledgements

Marco J., Marco P., Martin, Michi H., Michi S., Mike, Moe, Nicola,Peter S., Philipp, Reto, Rudolf, Salome, Stephan, Stigi, Susanne,Thomas, Urs G., Viton, Xinhui, for the friendly and inspiring work-ing atmosphere.

• my parents for their steady support, particularly in hard times.

Curriculum Vitae

I was born on August 22nd 1974 in Krefeld (Germany) as the first childof Gisela and Friedrich-Georg Lange. In the same place, I attended theArndt-Gymnasium (high school), from where I graduated in May 1994(Abitur). After one year of mandatory military service, I started my stud-ies in Physics at the technical university RWTH Aachen (Germany) inOctober 1995. Between 1997 and 1998, I spent one year as an exchangestudent at the University of Edinburgh (Scotland). In 2000, I worked fortwo months as a student trainee in the Corporate Research Departmentof Infineon Technologies (Munich). After my diploma thesis at the In-stitute of Solid State Research (Forschungszentrum Julich, Germany), Igraduated from the RWTH Aachen with a physics diploma in September2001. In December 2001, I joined the bio-physics group of Prof. P. Bosigerat the Institute for Biomedical Engineering, University and ETH Zurich(Switzerland), as a PhD student and research assistant. My research wasfocused on various aspects of magnetic resonance spectroscopy at higherfield strengths, in particular the detection of J-coupled metabolites in thehuman brain and prostate as well as two-dimensional spectroscopy tech-niques.

113

rights / license: research collection in copyright - non …29643/et… · diss. eth no. 17140...

Documents