Optimized quantitative magnetic resonance spectroscopy for clinical routine



Several practical obstacles in data handling and evaluation complicate the use of quantitative localized magnetic resonance spectroscopy (qMRS) in clinical routine MR examinations. To overcome these obstacles, a clinically feasible MR pulse sequence protocol based on standard available MR pulse sequences for qMRS has been implemented along with newly added functionalities to the free software package jMRUI-v5.0 to make qMRS attractive for clinical routine. This enables (a) easy and fast DICOM data transfer from the MR console and the qMRS-computer, (b) visualization of combined MR spectroscopy and imaging, (c) creation and network transfer of spectroscopy reports in DICOM format, (d) integration of advanced water reference models for absolute quantification, and (e) setup of databases containing normal metabolite concentrations of healthy subjects. To demonstrate the work-flow of qMRS using these implementations, databases for normal metabolite concentration in different regions of brain tissue were created using spectroscopic data acquired in 55 normal subjects (age range 6–61 years) using 1.5T and 3T MR systems, and illustrated in one clinical case of typical brain tumor (primitive neuroectodermal tumor). The MR pulse sequence protocol and newly implemented software functionalities facilitate the incorporation of qMRS and reference to normal value metabolite concentration data in daily clinical routine. Magn Reson Med, 2013. © 2012 Wiley Periodicals, Inc.

Techniques for quantitative localized in vivo magnetic resonance spectroscopy (qMRS) have been available for 30 years and were used in countless research projects (1–4). The potential value of qMRS in a clinical setting has especially been demonstrated in the characterization of pathologic tissue changes, where the diagnosis is not evident in MR imaging and in tissues that are not readily available to biopsy (e.g., intracerebral tumors) (5). Despite the fact that localized in vivo MR spectroscopy is noninvasive, and produces quantitative biochemical information (6), its use in clinical routine examinations is still very limited. There are several reasons for this:

First, as MR images are closely linked to anatomical structures, their interpretation is for clinical radiologists much more intuitive than that of MR spectra. To interpret MR spectra correctly, knowledge about the signal acquisition and signal artifacts is mandatory. To obtain this knowledge, training is required with experienced spectroscopists, which is not possible in most clinical centers. Furthermore, background knowledge about cellular metabolism is also necessary in order to meaningfully interpret the limited number of metabolites in MR spectra. Compared to the immense amount of different structural proteins, enzymes, and metabolites of a human cell, the very few substances that can be measured using MR spectroscopy yield a low specificity and sensitivity in the context of pathologic tissue changes.

Second, clinical MR scanners and manufacturer's spectroscopy postprocessing software have greatly improved and became user-friendly in the recent years. However, most of them do not offer the possibility to quantify the measured metabolites in units of amount (expressed in mole) per unit volume (L−1) or per kilogram wet weight (kg−1 ww). This implies that the user can only interpret peak area ratios, which is for some applications sufficient, but not for all. The inherent weaknesses of using peak area ratios (6), is that the total number of quantifiable metabolites decreases by one, and alterations in the peak ratio cannot be assigned unambiguously to changes in the concentration of the numerator or denominator. More serious are the indications that peak area ratios may introduce larger errors than qMRS (7, 8). In order to calculate “absolute” quantities of measured substances (mol L−1 or mol kg−1 ww) in a clinical setting, freely available quantification methods like AMARES, AQSES, QUEST, TDFDFIT (9–13), the free software package jMRUI (14), or commercially available software like LC-Model can be used (15). Most of these applications have no, or only very rudimentary image display possibilities, preventing the study of MR spectra together with MR images of the same examination. As the currently available offline applications for qMRS do not support DICOM network transfer functionality, the acquired spectral data need to be exported manually from the main MR console (or satellite workstation), to a computer for quantification having one of the above mentioned programs installed. Additionally, none of the applications for qMRS offer any DICOM-typed reporting functionality, and therefore, reporting of quantitative results is normally done by the creation of text/html files and/or images (e.g., jpg, gif), or by manual insertion of screenshots of reports into a picture archiving and communication system (PACS). Overall, in order to do clinical qMRS, most data handling has to be done manually, which is time consuming, and prone to errors. The time and financial cost of additional personnel makes qMRS unattractive in centers with intense workload, and prevents its use on a larger scale.

Third, qMRS of pathologic tissues requires the availability of normal values for comparison. Normal values, even within an organ (e.g., brain tissue), depend on voxel localization and subject age (16–18). Moreover, as large voxel sizes are needed for qMRS due to low signal-to-noise ratio, there is always partial volume effect of tissues with different MR spectroscopic profiles. The high variability of MR spectroscopic results further decrease the specificity and sensitivity of this method for the use in clinical routine. To minimize this variability, patient MRS results have to be compared to valid reference concentrations, and multiple databases per organ and age must be available. The currently available software packages do not offer any possibility of building up and using normal value databases, which prevents simple and easy to use comparison of patient data with reference data.

In order to make qMRS a more valuable and attractive tool for daily routine use, we present a clinically feasible MR pulse sequence protocol (using standard MR pulse sequences) along with newly developed software functionalities extending the functionality of the free software package jMRUI-v5.0 to overcome all of the obstacles described above (14).

The new software functionalities of jMRUI-v5.0 described in this article were developed in the framework of the FAST project supported by the EU (EU Marie Curie Research Network: MRTNCT-2006-035801, 2006–2010) and parts of it were presented at the 19th annual meeting of the International Society for Magnetic Resonance in Medicine in 2011 (19).


Following requirements need to be met for an optimized work-flow in clinical qMRS, which will be elaborated in detail below: (1) adequate and clinically feasible MR pulse sequence protocol, (2) network capabilities for automatic transfer of qMRS data to personal computer, (3) quantification algorithms using water reference data, (4) creation of institutional normal metabolite concentration databases, (5) report generation and storage capabilities, and (6) network capabilities for automatic transfer of qMRS results/reports to a PACS.

General Software Implementation Details

The main idea behind the software developments is to minimize the time needed by the clinician to perform a qMRS evaluation, while maximizing the visibility of MRS results among MR imaging results to other clinicians. A JAVA-library named SCANalyze has been developed that provides easy to use graphical controls for DICOM file manipulation over the network (DICOM file transfer from MR console to personal computer with jMRUI-v5.0 installed, and DICOM-formatted spectroscopy report file transfer from personal computer to MR console or PACS for long time storage), image and spectrum display within the same qMRS application, qMRS data processing for metabolite quantification, and normal metabolite database creation. Two types of widely used water referencing methods for qMRS were implemented, one that requires only one water reference signal, the other that requires multiple water reference signals (20, 21). The latter signals enable computation of relative volumes of the intracellular water compartment (ICW) and the extracellular water compartment (ECW). ICW contains most of the metabolites whereas ECW contains mainly cerebrospinal fluid (CSF) or pathologic extracellular edema. Based on a two-compartment model (tcm) described below, concentration corrections caused by increased partial volume of extracellular water can be handled.

This SCANalyze library has been developed in-house from scratch in the programming language JAVA. The DICOM network functionality, however, was obtained by integrating specific functionality from the dcm4che2 DICOM Toolkit (see Evans D, Zeilinger G, Cappellini U. dcm4che2 DICOM Toolkit. 2010. Available at: http://www.dcm4che.org/confluence/display/d2/dcm4che2+DICOM+Toolkit.) into our SCANalyze library. The new functionality is available in jMRUI-v5.0 and is incorporated into five custom plug-ins, which are described in more detail in the Supporting Information of this article (Figs. S1–S5).

MR Scanner Hardware

The pulse sequence protocols (see below) were tested on all three clinical MR scanners of the institution, Trio™ and Verio™ running at 3.0 T, and Avanto™ running at 1.5 T (MAGNETOM, Siemens® Healthcare Solutions, Erlangen, Germany) equipped with standard 16 or 32 multichannel head coils.

Required Pulse Sequence Protocol

As the goal of this work is to provide a solution for clinical routine, the scanner measurement protocol consists solely of standard, commercially available pulse sequences. We implemented a protocol for brain tissue MRS (see Table 1) consisting of a concatenation of 12 point resolved spectroscopy (PRESS) pulse sequences for single voxel spectroscopy (svs) (1), all 12 derived from the same Siemens® product “svs_se”.

Table 1. Displayed are the Required 12 MR Pulse Sequences and Settings for Performing qMRS in Clinical Routine
No.Protocol nameTE (ms)Repetition time (ms)NEXNEX (dummy scans)Total scan time (minutes:seconds)Details
  1. All sequences are derived from the svs_se sequence (i.e., standard Siemens® PRESS). The total measurement time is 380 s (6 min 20 s).

  2. tcm, two-compartment model.

  3. Dummy scans are required to reach steady state magnetization.

1.svs_se_30_ws30150019244:48Water-suppressed (ws) signal
2.svs_se_30_nws301500440:12Non-water-suppressed (nws) signal
3.svs_se_tcm_dummy308000100:08Dummy acquisition for steady state
4.svs_se_tcm_30308000100:08Total nws signal for tcm at TE 30 ms
5.svs_se_tcm_40408000100:08Total nws signal for tcm at TE 40 ms
6.svs_se_tcm_50508000100:08Total nws signal for tcm at TE 50 ms
7.svs_se_tcm_80808000100:08Total nws signal for tcm at TE 80 ms
8.svs_se_tcm_1001008000100:08Total nws signal for tcm at TE 100 ms
9.svs_se_tcm_2002008000100:08Total nws signal for tcm at TE 200 ms
10.svs_se_tcm_5005008000100:08Total nws signal for tcm at TE 500 ms
11.svs_se_tcm_100010008000100:08Total nws signal for tcm at TE 1000 ms
12.svs_se_tcm_150015008000100:08Total nws signal for tcm at TE 1500 ms

The programming of the measurement protocol itself (Table 1) on the main MR console takes less than twenty minutes. Starting from the Siemens® product “svs_se,” the sequence parameters are set such that the echo time (TE) is 30 ms, the repetition time is 1500 ms, applying water suppression (ws), using 192 averages for a 15 × 15 × 15 mm3 voxel and renaming the pulse sequence to “svs_se_30_ws.” By linking the other 11 protocol steps with the first “svs_se_30_ws” protocol step using the option “Copy measurement parameters,” users have to define the localization and size of the spectroscopic voxel only once, and the other protocol steps are executed automatically accordingly. The second measurement (svs_se_30_nws) with no water suppression (nws) is required for the first type of quantification, the so-called single compartmental quantification (see next section). Additionally, this signal can be used for the so-called QUALITY preprocessing when the line shape is severely distorted by eddy currents or insufficient shimming (22, 23).

The following 10 measurements (No. 3–12) are single-shot non-water-suppressed acquisitions (repetition time of 8000 ms) to obtain the relative sizes of the two fully relaxed water components: (1) ICW, and (2) ECW (e.g., CSF or extracellular edema in brain 1H MR spectroscopy), which can be discriminated based on their differences in T2-relaxation. As ECW contains hardly any in vivo MRS visible metabolites, knowledge of the size of this compartment enables more precise metabolite concentration estimation in the case of a non negligible ECW partial volume. The compartment model and the computation of absolute concentrations were first proposed by Kreis et al. and Ernst et al. in 1993 (20, 21). In contrast to their approach, which requires an additional water-reference signal from a water containing phantom located within the same radiofrequency coil, our plug-ins assume the availability of only the above described signals, and apply a two-compartment model only (see below). In the case of brain MR spectroscopy, ECW and ICW are the two compartments which can be discriminated. The reason for this is that we consider an additional water reference measurement to be unfeasible in a clinical routine setting.

The total duration of the spectroscopy protocol of Table 1, excluding power optimization and shimming, is 192 × 1.5 s + 8 × 1.5 s + 10 × 8 s = 380 s (6 min 20 s). Normally, we perform one short (30 ms) and one intermediate (135 ms) TE measurement from the same voxel location in brain MR spectroscopy, in order to facilitate the lactate concentration determination in the presence of an intense macromolecular signal at 1.3 ppm. In this case, the measurement time is not completely doubled, but 680 s (11 min 20 s), as the protocol steps for determination of the two-compartment model (i.e., No 3–12 in Table 1) must be determined only once.

Quantified Metabolites Using QUEST

Following metabolites were quantified using QUEST algorithm supported by jMRUI-v5.0: N-acetyl aspartate (NAA), choline (Cho), creatine (Cr), myo-inositol (mI), glutamine (Gln), glutamate (Glu), and lactate (Lac).

Quantification Using One Single Water Reference Signal

For this type of quantification–implemented in jMRUI-v3.x and v4.0–which we call single water compartmental quantification, the metabolite concentrations Cmetabolite are estimated from the localized water suppressed signal (svs_se_30_ws) and one nonsuppressed water reference signal (svs_se_30_nws) from the same voxel, and can be estimated using the following equation (24):

equation image(1)

in which Smetabolite is the fitted signal amplitude obtained by jMRUI, Swater is the signal amplitude of the unsuppressed water reference measurement obtained by jMRUI, Nmetabolite is the number of protons of the metabolite, Mwater is the molarity of water (which is 55.3 mol L−1 at 25°C), Ftissue the tissue correction factor accounting for the tissue water content (24), and finally Fwater the water correction factor accounting for differences in T1- and T2-relaxation times between the water signal and the metabolite signals. If the user provides only one water reference signal, Fwater cannot be determined, and this factor is normally assumed to be 1.0. For the factors Ftissue the user can either use predefined values for brain tissue, or set this factor manually on a subpanel of the Quantitative Result Analysis & DICOM Reporting Plug-In (for details see Supporting Information, Fig. S5). If the clinical user provides only one water reference signal Fwater and Ftissue are often assumed to be 1.0.

Quantification Using Multiple Water Reference Signals

In case the user provides all water reference signals of Table 1, the estimated metabolite concentrations can be corrected for ECW compartment size (CSF and extracellular edema in case of brain 1H MRS). The total water peak areas of the eight water reference signals Wtotal(ti) (see protocol in Table 1, rightmost column) are determined. The relative water compartment sizes A0,ECW and A0,ICW, and their corresponding relaxation times T2,ECW and T2,ICW are computed from Wtotal(ti) by nonlinear least squares fitting with the biexponential model function:

equation image(2)

using a constrained nonlinear least square fitting algorithm (25). The fitting of [Eq. 2] has been automated within one dedicated plug-in (for details see Supporting Information, Fig. S4). The water correction factor Fwater is then computed by:

equation image(3)

The factor Ftissue in [Eq. 1] can be set in a subpanel of the Quantitative Result Analysis & DICOM Reporting Plug-In (for details see Supporting Information, Fig. S5). In the current implementation users can select water correction values for white matter and gray matter, and for adults and infants which were obtained from Hofmann 2000 and references therein (26).

Creation of Normal Value Databases

An absolute quantitative estimation of metabolite concentrations in clinical routine qMRS makes sense only if the concentrations obtained for the patient can, with minimal effort, be compared to concentrations obtained from healthy controls for the same anatomical region, under comparable experimental conditions. We created a database for our 3.0T Verio™ and Trio™ scanner, by analyzing retrospectively signals from 19 healthy adults: 10 males, 9 females, average age: 32.7 ± 10.6 years (range 20–61 years, median: 31 years). All required signals for the protocol used (see Table 1) for TE = 30 ms and TE = 135 ms svs spectra in five different localizations have been acquired: i.e., (1) semioval center, (2) frontal white matter, (3) parietal white matter, (4) basal ganglia, and (5) cerebellar vermis. The voxel size was 15 × 15 × 15 mm3. The signals were acquired with a 1.2 kHz bandwidth sampling 1024 points. To improve signal-to-noise ratio for display, the spectra were preprocessed with a 5 Hz Lorentzian apodization. Hankel-Lanczos Singular Value Decomposition was used for residual water signal removal (27). Some of the spectra were zero-order phased, in order to have well phased data in the database. The spectra were quantified in the time domain with the QUEST algorithm using a 20 data-points “in base” macromolecular baseline determination for the TE = 30 ms svs spectra (10, 11). No macromolecular baseline estimation was used for the TE = 135 ms spectra.

We also created a pediatric database for our 3.0T Trio™ scanner, acquiring all required signals (Table 1) from healthy children (21 females, 15 males, average age 12.0 ± 2.6 years; median 13 years, range 6–16 years). Only two anatomical regions were examined from a TE = 135 ms svs experiment: (1) semioval center, and (2) basal ganglia. The spectra were acquired with a 2 kHz bandwidth, sampling a total of 2048 points. Again, to improve the signal-to-noise ratio for display, the spectra were apodized with a 5 Hz Lorentzian filter, and Hankel-Lanczos Singular Value Decomposition was used to remove the water signal before quantification (27). In some individual cases, the spectra were zero-order phased before fitting, in order to have well phased spectra into our database. All spectra were quantified using the QUEST algorithm (10, 11). As only long TE spectra were available, there was no need to estimate the macromolecular baseline.

This study was approved by the local ethics committee and the internal review board of our hospital.

Errors in Concentrations

As the MRS signals have only a limited signal-to-noise ratio, the concentrations computed from them contain errors. These errors can be estimated from the noise level and the applied model, and are called the Cramér-Rao minimum variance bounds (CRB). Theoretically, the CRB is the lowest possible error, and is computed by jMRUI. In in vivo spectroscopy it has become practice to provide the CRB along with their corresponding concentrations values; jMRUI and this article adhere to this.

For the errors in the concentration values averaged over several healthy control persons in the normal-value databases, the standard deviation is indicated.


The average metabolite concentrations obtained from our signals of children for two different anatomical localizations are listed in Table 2; the values obtained for the five localizations in adult brain are displayed in Table 3.

Table 2. Estimated Normal Metabolite Concentrations Obtained from Signals of 36 Children
Metabolite nameBG; N = 26 (mmol/kg−1 ww)CSO; N = 31 (mmol/ kg−1 ww)
  1. BG, basal ganglia; CSO, semiovale center; Cho, choline; Cr, creatine; Glu, glutamate; Gln, glutamine; Lac, lactate; mI, myo-inositol; NAA, N-acetyl aspartate.

Cho1.3 ± 0.11.5 ± 0.1
Cr6.4 ± 0.25.2 ± 0.1
Glu4.4 ± 0.23.7 ± 0.2
Gln2.9 ± 0.22.2 ± 0.1
Lac0.4 ± 0.10.3 ± 0.1
mI8.2 ± 0.77.8 ± 0.4
NAA7.1 ± 0.27.5 ± 0.2

We have tested the MR pulse sequence protocol and the developed software including above-mentioned databases on a number of clinical routine cases, and will now describe the obtained results in one illustrative case of a primitive neuroectodermal tumor (PNET).

Figure 1a shows the measured brain spectrum (TE = 135 ms) of a 10 years old patient with a supratentorial primitive neuroectodermal tumor (PNET) confirmed histologically, and its best fit computed with jMRUI-v5.0 (QUEST algorithm). Figure 1b shows the corresponding numerical report of this PNET and the localization of the voxel within the lesion. The spectral characteristics match very well with PNET spectra published previously (38). Interestingly, this PNET spectrum shows a 5.9 times increase in choline concentration (8.8 ± 0.1 mmol kg−1 ww), a 1.5 times increase in creatine (7.7± 0.2 mmol kg−1 ww), a 1.5 times reduction in N-acetyl aspartate (4.9 ± 0.2 mmol kg−1 ww), a 2 fold increase in myo-inositol (20.7 ± 1.2 mmol kg−1 ww), and a 1.9 times increase in the sum of glutamate and glutamine, i.e., Glx (11.0 ± 0.4 vs. 5.7 mmol kg−1 ww) compared to our normal database. Additionally, the spectrum shows pathologic lactate (1.2 ± 0.1 mmol kg−1 ww), normally 0.3 ± 0.1 mmol kg−1 ww.

Figure 1.

Reporting of qMRS results in clinical routine illustrated in a primitive neuroectodermal tumor. a: Fitted, measured and residual results of the MR spectrum at TE 135 ms (QUEST algorithm) of a primitive neuroectodermal tumor (PNET). b: DICOM-typed report created by jMRUI with concentrations obtained for the patient in the second column (including Cramér-Rao minimum variance bounds) and the average normal concentrations (including standard deviation) of the control group for the semioval center. Cho choline, Cr creatine, Glx (glutamine and glutamate), Lac lactate, mI myo-inositol, NAA N-acetyl aspartate.


Absolute Quantification for Diagnosis in Clinical Routine

For correct interpretation of the absolute concentrations reported in this article using the described MR pulse sequence protocol and newly developed software functionalities, one should be aware of the following assumptions/restrictions, made in most of the quantification algorithms:

  • 1)In the described approach, the “absolute” concentrations are computed relative to a “known” concentration, in this case the concentration of water in the tissue. The “Water Reference Information” tab (for details see Supporting Information, Fig. S4) enables to make further specifications on the water content of the tissue under investigation. In the literature, concentration values can be found for predominantly healthy tissues e.g., Ref.20,21. For most pathologic tissues, however, exact water content is often not known, and therefore, the exact reference is missing. In the examples given above, we assumed the same value for healthy brain tissue as well as for pathologic tissue.
  • 2)The two-compartment model that was implemented, enables to make concentration corrections that depend on the fraction size of ECW in the voxel (e.g., CSF or extracellular edema in brain), and is the largest correction factor. This means that the voxel can be placed freely within the brain, as the concentrations are corrected for ECW partial volume.
  • 3)Apart from the repetition time and TE, the T1- and T2-relaxation of all spins in the metabolites must be known for correct computation of concentrations from the fitted peak areas. For J-coupled spin systems, which most of the metabolites are, the measurement of T1- and T2-relaxation is time-consuming and far from trivial, and therefore, knowledge about them is normally limited to nonpathological tissues (39–41). For pathological tissues, however, the T1- and T2-relaxation are not known. On the “Relaxation Time Database”-tab of our “Quantitative Result Analysis & DICOM Reporting” plug-in (for details see Supporting Information, Fig. S5), the user can define one (T1, T2)-pair per metabolite. As the relaxation times T1 and T2 depend on the strength of the magnetic field B0, we generated a separate database for each commercially available field strength, namely 0.5, 1.0, 1.5, 2.0, 3.0, 4.0, and 7.0 T. These databases can be adapted by the user.
  • 4)A further restriction is that in the current implementation, one can define only (T1, T2)-pair per metabolite, implying spin independent relaxation constants, which is, strictly speaking, not correct. In our 3T databases, the effective relaxation time for all spins belonging to one metabolite were determined such that the average obtained concentrations matches well with literature values.

Due to these four uncertainties and assumptions, and the fact that different assumptions were made by researchers in the past, a wide variety of normal metabolite concentrations is published in the literature (17, 28–34, 42). The qMRS results obtained in this study are in good accordance to the metabolite concentration range reported in the referred literature (Table 3), with the exception of high values of myo-inositol in our study. The differences in observed concentrations not only stem from differences in assumptions made on relaxation times, water content etc. described above, but also from differences in hardware used like available gradient, radiofrequency-power, the radiofrequency-pulse shape, the receiver gain (43, 44), which lead to systematic differences in the obtained results. Another factor, which may not be forgotten in the comparison, is the average age of the examined subjects and the exact size and localization of spectroscopic voxels used. Additionally, to these factors that account for the variability of all metabolite concentrations in different studies, complicated J-coupled multiplets which have low spectrum intensity (such as myo-inositol at TE 135 ms) are more difficult to quantify in comparison to strong spectrum pattern dominating singlets (such as creatine, N-acetyl-aspartate, or choline), and may thus further increase the variability of the obtained metabolite concentration values.

Table 3. Estimated Normal Metabolite Concentrations Obtained from Signals of 19 Adults, and Comparison of Metabolite Concentration to Literature (in round brackets; 28–36)
Metabolite nameFWM; N = 16 (mmol/kg−1 ww)PWM; N = 16 (mmol/kg−1 ww)CSO; N = 15 (mmol/ kg−1 ww)BG; N = 16 (mmol/ kg−1 ww)CV; N = 17 (mmol/ kg−1 ww)
  1. Note that the voxel localization was thalamus for the concentrations from literature (28,36,37), whereas the voxel localization was striatum/globus pallidus in the present study.

  2. FWM, frontal white matter; PWM, parietal white matter; CSO, semioval center; BG, basal ganglia; CV, cerebellar vermis; Cho, choline; Cr, creatine; Glx, sum of glutamate (Glu); glutamine (Gln); Lac, lactate; mI, myo-inositol; NAA, N-acetyl aspartate.

Cho2.0 ± 0.1 (1.8–2.4)2.0 ± 0.2 (1.6–2.4)1.9 ± 0.1 (1.7–1.8)1.3 ± 0.1 (1.8–1.9)2.6 ± 0.1 (2.1–3.0)
Cr6.2 ± 0.3 (5.7–10.6)7.1 ± 0.5 (4.9–8.8)6.9 ± 0.2 (5.7–6.7)6.9 ± 0.2 (6.8–9.2)10.8 ± 0.3 (9.0–12.8)
Glu4.6 ± 0.23.9 ± 0.34.8 ± 0.34.6 ± 0.25.7 ± 0.3
Gln3.1 ± 0.32.5 ± 0.23.6 ± 0.23.1 ± 0.22.9 ± 0.3
Glx (Glu + Gln)7.7 ± 0.3 (8.4–8.8)6.4 ± 0.3 (6.5–8.2)8.4 ± 0.3 (6.8–8.2)7.7 ± 0.2 (9.8–10.3)8.6 ± 0.3 (12.9)
Lac0.4 ± 0.10.4 ± 0.10.3 ± 0.10.3 ± 0.10.3 ± 0.1
mI9.1 ± 0.8 (2.7–3.8)8.6 ± 0.7 (3.3–5.2)9.3 ± 0.6 (2.9–4.7)12.3 ± 1.1 (3.5–4.9)13.2 ± 0.8 (4.2–5.6)
NAA8.0 ± 0.2 (9.6–16.6)10.0 ± 0.4 (8.2–13.4)10.1 ± 0.2 (9.5–12.1)6.9 ± 0.3 (10.4–13.6)8.2 ± 0.2 (8.7–11.1)

Due to all the issues mentioned above, the term “absolute quantification” as it is used frequently in MRS should be put into perspective, as the obtained “absolute” concentrations are only valid under certain assumptions, and may contain systematic errors due to hardware characteristics. This means that the SI-unit [mol kg−1 ww] is only conditionally valid, and we cannot claim that our results are more accurate than those of others (17, 28–34, 42). For clinical application, however, this should not be a problem as long as we have reference concentrations from healthy subjects of the same anatomical region, measured with the same protocols, postprocessed, and quantified in the same manner. In other words, the units resulting are not SI-units in a strict sense, but are “institutional units.” The developed software functionalities offer the clinical user an easy way to create own reliable normal metabolite concentrations databases, to which concentrations obtained from patients can easily and reliably be compared.

The time needed to perform an evaluation of qMRS data with the described work-flow and software is approximately 3–4 min, and includes the following nine user steps (a) send the study DICOM files to jMRUI-computer (can be automated); (b) start jMRUI and mount the automatically renamed study; (c) select and load the spectroscopy data; (d) water signal removal with Hankel-Lanczos Singular Value Decomposition; (e) apodization; (f) quantification with QUEST; (g) loading water references; (h) determination of ICW and ECW compartment sizes from water reference signals; (i) result viewing and reporting to a PACS (DICOM send) enabled after selection of a tissue reference database.

Limitations of the Developed Software Functionalities

The developed software functionalities presented in this article speed up the process of clinical qMRS massively, and also speed up the reporting process to the PACS of the results, which is vital in a clinical environment. What is not solved by the plug-ins is the problem of discriminating bad quality signals (e.g., due to signal artifacts), i.e., whether the spectrum is reliable enough to be used for clinical diagnosis. Another plug-in, not described here, which quantifies the MRS signal reliability, is available in jMRUI-v5.0. However, this plug-in only tests whether the steady state measurement condition is satisfied, and warns for scanner instabilities or patient motion (45). As this covers only a small number of potential artifacts, MRS in a clinical environment still requires some specialized spectroscopic knowledge.

The SCANalyze library has been developed to optimally process MRS data from Siemens scanners, and it has only been tested on these systems so far. It will be extended to accept the official standard DICOM MR spectroscopy file format in the next release as well (jMRUI-v5.0 already supports this format). The workflow described in our study is based on the SCANalyze library and a concatenation of standard PRESS sequences. If other manufacturer's single-voxel spectroscopy sequences support the standard DICOM MR spectroscopy file format and the sequence parameters described in our study (especially variable TE and possibility to use the MR pulse sequence with and without water suppression) the described workflow will work as well on these MR scanners.


Recently, based on the work of Chong et al. (46), Bolliger et al. (47) have presented a new measurement method/postprocessing method in which the concentration parameters as well as T1- and T2-relaxation time parameters can be estimated from one combined 2DJ/inversion recovery experiment in less than 20 min acquisition time. This method is another important step toward true qMRS as no assumptions need be made on metabolite relaxation times.

The databases, which can be made in jMRUI with the current plug-ins, are very simple and store only the most essential data, in simple files. In future, we would like to store patient spectra together with relevant clinical parameters (e.g., histology) in sophisticated relational or object databases. A major requirement for this system is that data can be added, very time-efficiently, during routine examination. This would enable the sophisticated querying of the data, enabling not only comparisons to healthy subject data, but also comparisons to databases of tumors specific types, in order to support the clinician in more accurate diagnosis, or as data input sources for automated spectrum classification as described by Garcia-Gomez et al. (48, 49) or Luts et al. (50).

Currently, a successor training and research project of the EU FAST project is submitted, which will exactly focus on this issue, i.e., eliminating the last hurdles that block wide application of qMRS truly applicable in a purely (routine) clinical environment.


We have described a clinically feasible work-flow for qMRS using a MR pulse sequence protocol based on standard MR pulse sequences and newly developed software functionalities for the free software jMRUI-v5.0. By DICOM network transfer protocol, users can send DICOM images and spectroscopy files to jMRUI, which are automatically renamed in user understandable file names. In contrast to other applications for qMRS, jMRUI now enables spectra to be analyzed together with structural images in one single application. A two-compartment water reference model has been implemented, in order to be able to make corrections for variable amounts of CSF or extracellular edema in the voxel. The software functionalities further enable the simple creation of databases of normal metabolite concentration values by the user, which can afterwards be used to compare concentrations obtained from patients, an utmost important issue in qMRS. To illustrate the performance of the MR pulse sequence protocol and the new software functionalities, a clinical brain tumor case (PNET), which closely match observations described in previous publications, is reported. The current implementation was made for spectroscopy data of Siemens® MR scanners; a generalized version for all scanners is under development.


The authors thank Ali Rahman for carefully reading through the manuscript.