In Vivo Short Echo Time 1H-Magnetic Resonance Spectroscopic Imaging (MRSI) of the Temporal Lobes

NeuroImage 14, 501–509 (2001) doi:10.1006/nimg.2001.0827, available online at http://www.idealibrary.com on

In Vivo Short Echo Time 1H-Magnetic Resonance Spectroscopic Imaging (MRSI) of the Temporal Lobes M. A. McLean,* F. G. Woermann,* R. J. Simister,* G. J. Barker,† and J. S. Duncan* *The MRI Unit, National Society for Epilepsy; and Epilepsy Research Group, University Department of Clinical Neurology, Institute of Neurology, Queen Square, London, United Kingdom; and †Multiple Sclerosis NMR Research Unit, Institute of Neurology, University College London, Queen Square, London, United Kingdom Received August 31, 2000

Two different methodologies for obtaining PRESSlocalized magnetic resonance spectroscopic imaging (MRSI) data from the mesial and lateral temporal lobes were investigated. The study used short echo times (30 ms) and long repetition times (3000 ms) to minimize relaxation effects. Inhomogeneity and spectral distortions from the proximity of the temporal bones precluded the attainment of consistently goodquality data from both temporal lobes at once. Even when the right and left temporal lobes were studied separately, distortions often disturbed spectra from the anterior lateral temporal lobe. Quantitative analysis using LCModel was therefore performed only on the posterior lateral temporal lobe, and the posterior, middle, and anterior mesial temporal lobe. No significant left-right differences in metabolite content were found in a series of 10 controls. Significantly higher concentrations of myoinositol and choline were found in the anterior mesial temporal lobe, even when grey matter content was included as a covariate. The concentration of N-acetyl aspartate plus N-acetyl aspartyl glutamate (NAc) was not found to vary significantly along the length of the hippocampus. The previously observed lower anterior ratios of NAA to creatine plus choline (NAA/(Cr ⴙ Cho) may instead have been due to higher anterior choline. Large differences in metabolite concentrations were seen between posterior lateral temporal lobe (predominantly subcortical white matter) and the posterior mesial temporal lobe, most notably lower creatine, glutamate/glutamine, and myo-inositol, and higher NAA/(Cr ⴙ Cho) in the lateral than mesial temporal lobe. This pattern was similar to that previously seen for grey/white matter differences in the frontal, parietal and occipital regions. © 2001 Academic Press

Key Words: 1H-MRSI; human brain; temporal lobes.

INTRODUCTION Magnetic resonance spectroscopic imaging (MRSI) provides spectroscopic information from a number of

voxel locations simultaneously. A number of studies have suggested that MRSI can be a useful tool in lateralizing the epileptic focus in temporal lobe epilepsy (TLE) patients. Almost all have used long echo times (⬎130 ms), where the only peaks reliably visible are singlets arising from N-acetyl aspartate ⫹ N-acetyl aspartyl glutamate (NAc), creatine ⫹ phosphocreatine (Cr), and choline-containing compounds (Cho). Some groups have found the most useful markers of damage to be ratios such as NAA/Cr (Cendes et al., 1997) or NAA/Cho (Xue et al., 1997). Some have instead performed qualitative analysis of metabolite maps formed by integration or peak fitting of the singlet resonances (Constantinidis et al., 1996; Chu et al., 2000). One group has applied quantification relative to internal water in their evaluation (Ende et al., 1997). Long-echo times have several advantages over short echo times: they avoid the artifacts of lipid contamination and incomplete water suppression common to short-echo times, and typically produce spectra with smoother baselines and fewer peaks, which are easier to analyze. However, long echo time spectra lack information which is potentially of great interest in epilepsy, namely resonances due to the neurotransmitter glutamate (more readily measured as glutamate ⫹ glutamine, or Glx), and myoinositol (Ins), which is thought may be predominantly localized in glial cells (Brand et al., 1993). Our single-voxel short echo time studies have demonstrated alterations in these metabolites as well as in the more commonly observed NAA, and the pattern of these abnormalities differs according to whether the subjects have hippocampal sclerosis or are MRI-negative (Woermann et al., 1999). The application of short echo time MRSI to study the temporal lobes presents a number of technical challenges. The proximity of the temporal bones and sinuses makes it difficult to achieve good homogeneity. Rapid flow of blood and CSF through the plane, as well as the poor homogeneity, makes water suppression difficult. Lipid signal from the scalp can be a problem when examining lateral temporal cortex, necessitating

501

1053-8119/01 $35.00 Copyright © 2001 by Academic Press All rights of reproduction in any form reserved.

502

MCLEAN ET AL.

FIG. 1. Acquisition strategies investigated. (A) Single PRESS excitation box covering both temporal lobes (bilobe approach), with 24 ⫻ 24 phase encodes over 32 cm FOV. (B) PRESS excitation boxes separate for each temporal lobe (unilobe approach), with 20 ⫻ 20 phase encodes over a 26 cm FOV.

careful placement of the region of interest (ROI). One group overcame the problems of lipid contamination by preceding the spectral acquisition with an inversion pulse designed to null the lipids (Hetherington et al., 1995). This allowed them to obtain data from much closer to the scalp than the more commonly used methods using volume selection alone to exclude contaminating signal. However, such inversion pulses complicate quantification, due to their different effect on metabolites with differing T 1. We have investigated two strategies for acquiring short echo time temporal lobe MRSI data using PRESS localization (Fig. 1). The first, most commonly used method (Ende et al., 1997; Cendes et al., 1997; Constantinidis et al., 1996) excites a single large ROI covering both temporal lobes: we refer to this as the bi-lobe approach. The second method (Ng et al., 1994; Xue et al., 1997; Wang et al., 1999) acquires data from only one side at a time: we refer to this as the unilobe approach. As well as the methodological difficulties in acquiring short-TE MRSI data from the temporal lobes, there are serious methodological considerations in the quantitative analysis of the data. For any spectra at short echo times, sophisticated peak-fitting routines are necessary for analysis, to account for the signal from many small coupled metabolites and for the non-linear baseline. One approach is to develop spectral simulations: this method has been applied by Soher et al. (1998) to estimate NAA, Cr and Cho from temporal lobe MRSI data at an echo time of 135 ms and to quantitate

short-TE MRSI data in other parts of the brain. The other option is to obtain the necessary prior knowledge of overlapping metabolite signals by building a library of spectra from metabolite solutions. This is our approach: we apply the user-independent frequency-domain fitting routine LCModel (Provencher, 1993) to quantify metabolite concentrations from short-echo time spectra from MRSI of the human temporal lobes. One previous study (Wang et al., 1999) used LCModel to analyze spectra from temporal lobe MRSI with a 30 ms echo time as well as 135 ms, but reported only ratios of NAA, Cr, and Cho. A further difficulty in quantification of MRSI data, though not limited to short echo times, is the contribution of partial volume effects. Temporal lobe voxels can contain a large amount of cerebrospinal fluid (CSF), which contributes a negligible amount of metabolite signal. Additionally, voxels of interest may lie close to the edge of the PRESS-excited volume of interest (VOI), and thus lose signal due to partial excitation. We include in our quantitative analysis corrections for these effects (McLean et al., 2000). A final methodological consideration is the likely variation in metabolite content between and even within anatomic structures in the temporal lobe. Differences in metabolite ratios have been reported between the mesial and lateral temporal lobe (Wang et al., 1999), between the left and right temporal lobes (Bernard et al., 1996), and along the length of the hippocampus (Vermathan et al., 2000). This makes it imperative to have a processing method which allows

503

TEMPORAL LOBE MRSI

us to compare metabolite concentrations from anatomically similar locations on the left and right sides, and among subjects. This is not possible by analyzing the voxels where they happen to lie: the MRSI grid is always centred at the isocentre of the magnet, which is rarely at the same anatomic point in the brain among subjects. The correspondence between MRSI grid and anatomic landmarks is dependent on head size and position. Furthermore, the hippocampus, a tissue of great interest in epilepsy, curves in the plane of the MRSI slab, making it desirable to analyze voxels which are aligned with its curvature. In any MRSI dataset, it is possible to shift the location of the grid by phase multiplication during postprocessing (Brown et al., 1982). We have developed a novel routine in Sage/IDL (General Electric, Milwaukee, WI) to allow analysis of voxels shifted to conform with a user-defined region of interest, which can be tailored to the in-plane curvature of the hippocampus (McLean et al., 1999). This should maximize the hippocampal content of voxels; although with a slice thickness of 12 mm, and considering the point-spread function in spectroscopic imaging data, significant contributions from extrahippocampal structures also are present. Additionally, we have incorporated automated tissue segmentation into the analysis of metabolites, to investigate the signal dependence on the tissue composition of the voxels. The aims of this study were: (1) To evaluate acquisition strategies for quantitative short echo time MRSI in the human temporal lobes, (2) To develop a processing method which would allow us to compare metabolite concentrations from anatomically similar locations on the left and right sides, and among subjects, and (3) To investigate putative metabolic variations between left and right sides, mesial and lateral temporal lobe, and along the length of the hippocampus. METHODS MRSI Acquisition A 1.5T GE Signa Horizon Echospeed scanner was used, running 5⫻ software and using the standard quadrature head coil (General Electric). Controls were scanned with the long axis of the hippocampus aligned orthogonally to the long axis of the magnet by positioning subjects with their necks slightly extended, as in our earlier single-voxel studies (Woermann et al., 1999). Two scanning protocols were compared for efficacy (Fig. 1). In the first, referred to as bilobe (n ⫽ 6 subjects, median age 25, range 19 –30) a single PRESSexcited volume covered both right and left hippocampi and lateral temporal cortex. The acquisition matrix was 24 ⫻ 24 over a 32-cm FOV, with a 12-mm slice thickness, yielding a nominal voxel size of 2.13 cc and an acquisition time for 1 excitation (TE/TR ⫽ 30/3000) of 29 min. This gave a total examination time including

routine imaging, localization, and shimming of about an hour. In the second acquisition technique, referred to as unilobe (n ⫽ 10 subjects, median age 31, range 23 to 36), the two temporal lobes were studied separately, similar to the technique of Ng et al. (1994). The acquisition matrix was 20 ⫻ 20 over a 26-cm FOV, with a 12-mm slice thickness, giving a nominal voxel size of 2.03 cc. The acquisition time was 20 min for each temporal lobe, for a total examination time of approximately 75 min. The TR used (3 s) is very long for MRSI experiments—more commonly 2 s or even shorter TR is used; but this was not thought to be acceptable for the purposes of attempting quantification, since the variation in metabolite T 1 among the tissues of the temporal lobe has not been definitively determined. MRSI Processing We analyzed the data using LCModel Interactive (McLean et al., 1999), a locally developed program in Sage/IDL, which allows the user to define regions of interest (ROIs) by drawing with a cursor on a reference image. The program then extracts voxels shifted to be aligned maximally with the ROI for analysis using LCModel. This facilitates comparison of anatomically matched voxels from left and right sides and among subjects. It was used to examine three voxels from each mesial temporal lobe, encompassing the hippocampus, labeled A (anterior), M (middle), and P (posterior) (Fig. 2). The program was also used to define an ROI in the lateral cortex with voxels at roughly the same anteroposterior positions as the mesial temporal voxels. Partial Volume Effect Corrections The voxel content of grey matter, white matter, and CSF was determined by segmenting T1-weighted 3-D gradient echo datasets using SPM 96, as described previously (McLean et al., 2000). Inversion-prepared spoiled gradient echo (IRPSPGR) images were acquired (124 ⫻ 1.5-mm-thick), with the FOV the same as the MRSI FOV (26 or 32 cm), and a matrix of 256 ⫻ 128. TE/TI/TR ⫽ 4.2/450/16 ms, giving an acquisition time of 5:47. Segmentation was performed without prior normalization, but with a tilt of approximately 15° applied to more closely approximate the axial orientation of the template. Following segmentation, the dataset was reoriented to the original angle, and the nine slices covering the thickness of the MRSI slab were extracted from the segmented datasets and averaged to create images of grey matter, white matter, and CSF distribution. The metabolite concentrations (Cmet) were corrected for the partial volume effect due to CSF by finding the average fractional content of CSF in each voxel (F CSF), and applying the correction: C met ⫽ Cmet*共1/共1 ⫺ FCSF兲兲.

(1)

504

MCLEAN ET AL.

The metabolite concentrations were also corrected for estimated losses due to imperfections in the PRESS excitation slice profiles, as described elsewhere (McLean et al., 2000). Two images of the MRSI slab were acquired using PRESS localization: one image with the voxel placed as for the ROI and one with the voxel the same size as the field of view (FOV). The ROI image was divided by the FOV image to produce a map of the excitation profile in two dimensions. The average pixel intensity of the excitation profile image over the region corresponding to each MRSI voxel, as a fraction of the theoretical maximum (F PRESS) was used to correct for signal loss: C met ⫽ Cmet/FPRESS.

(2)

B 0 Homogeneity Since an important limitation of studying the temporal lobe is known to be the difficulty in achieving acceptable B 0 homogeneity over the volume of interest, we acquired field maps to assess the two acquisition strategies. The method we used was similar to one previously applied to derive frequency corrections for MRSI datasets (Soher et al., 1996). Gradient echo images were acquired with TE ⫽ 10 ms, TR ⫽ 250 ms, 45° flip. They were acquired immediately following the MRSI acquisition, using the same shim settings. The raw real and imaginary images were processed using viewit (NCSA Biomedical Imaging Group, UIUC) to produce field maps with sufficient dynamic range for comparison:

ing phasing and frequency assignments. The exception was that frequency shifts across the FOV due to inhomogeneity sometimes caused a complete failure in assignment of peaks; in these cases the data could be salvaged by changing the default reference frequency in the LCModel control file (setting the control variable PPMSHF ⫽ ⫺0.2). LCModel results were scrutinized for nonrandom residuals and for sharp variations in the baseline, but it was not necessary to reject any data for these reasons, since exclusion on the basis of FWHM appeared to be sufficient. Examples of spectra extracted from single voxels of MRSI datasets from the mesial temporal lobes are shown in Fig. 2. The signal to noise ratio is adequate for peak fitting of the major metabolites: the lowest SNR value reported was 5. The spectral resolution was noticeably poorer in the anterior voxels, but the creatine and choline peaks were still well enough resolved to estimate them individually. The peaks in the glutamate– glutamine region to a large extent merged, such that an estimate of the individual concentration of either was assumed to be unreliable. However, for glutamate ⫹ glutamine (Glx), LCModel reported on average SD ⱕ 10% at every voxel position, whereas metabolites with SD up to 15% are generally considered reliably determined. The true confidence in quantification of Glx is rather worse than this would suggest, largely due to the underlying macromolecule signal, which is not excluded; but Glx results have been reported here as an interesting indication of the amount of signal present in this region of the spectrum. Partial Volume Effect Corrections

⫺1

Field map ⫽ 1000*tan 共imaginary/real兲.

(3)

Statistics Statistical analysis was performed with SPSS 9.0. The asymmetry index (AI) for each mesial temporal voxel position was calculated as follows: AI ⫽ 关共right ⫺ left兲/0.5*共right ⫹ left兲兴*100%.

(4)

The AI was tested for significant difference from zero using a one-sample t test. Metabolite concentrations and ratios were compared across the anterior, middle, and posterior positions using ANOVA, with and without the use of grey matter fraction as a covariate. Paired t test comparisons were also performed between the posterior mesial and lateral voxels. RESULTS Spectral Quantification Even in the absence of any pre-processing, the results from LCModel were generally acceptable regard-

Phantom validations of our partial volume corrections based on segmentation and on PRESS excitation profiles are presented elsewhere (McLean et al., 2000). In this study we merely confirmed that application of the corrections for the PRESS excitation profile and the CSF content of the voxels decreased the coefficient of variation of the metabolite concentrations. The mean coefficient of variation (CV) over all 5 metabolites and all three mesial temporal voxel locations was reduced from 16 to 11% when the correction was applied (Table 1). B 0 Homogeneity Examples of the field maps acquired following bilobe and unilobe temporal MRSI acquisitions are shown in Fig. 3. The bilobe acquisition (Fig. 3A) shows hot spots over both temporal bones, with a number of contours over both hippocampi. This was one of the best datasets from this series: most had even more contours over the central region. The unilobe acquisition (Fig. 3B) by contrast shows uniform intensity over the hippocampus being studied, but contours over the contralateral temporal lobe, particularly severe over the anterior

505

TEMPORAL LOBE MRSI

FIG. 2. Application of ROI-drawing routine to study metabolite concentrations in the anterior (1), middle (2), and posterior (3) of the mesial temporal lobe, and in the posterior lateral temporal lobe (4). (A) Scout image with voxel positions indicated. (B) Extracted spectra from each voxel. Acquisition was performed with TE/TR ⫽ 30/3000 ms. For display, 1.25 Hz apodization is applied.

cortex. Phase variation remains within the lateral anterior section of the region-of-interest even in the unilobe approach, which might explain the observation of reduced metabolite signal in this region. In theory a correction could be derived from the field maps for signal loss due to B 0 inhomogeneity. In practice, the signal loss in this area is severe enough (and the likely error inherent in deriving this correction great enough) to render the data unsalvageable from a quantitative point of view. However, peak ratios may still be valid in the anterior extra-hippocampal cortex, as different metabolites should be affected similarly by the B 0 imho-

mogeneity. Over the mesial temporal lobes, and in the posterior lateral temporal lobe, we believe that a quantitative approach is valid, in the case of the unilobe acquisition strategy only. Mesial Temporal Metabolite Concentrations The metabolite concentrations were found to be symmetrically distributed on the right and left. Therefore, the means of data from left and right sides were used for subsequent comparisons. The metabolite concentrations from the mesial and lateral temporal lobes are

TABLE 1 Metabolite Concentrations [mM, mean ⫾ SD (CV)] and Ratios in the Anterior (AMT), Middle (MMT), and Posterior (PMT) Mesial Temporal Voxels, and from the Posterior Lateral Temporal Lobe (PLT), Using the Unilobe Approach (n ⫽ 10)

NAc Glx Cr Ins Cho NAc/(Cr ⫹ Cho) Grey/(grey ⫹ white)

AMT

MMT

PMT

PLT

ANOVA

ANCOVA

7.8 ⫾ 0.5 (6) 12.7 ⫾ 2.6 (20) 6.3 ⫾ 0.7 (11) 6.8 ⫾ 0.8 (12) 1.8 ⫾ 0.2 (11) 0.96 ⫾ 0.07 (7) 0.90 ⫾ 0.07 (8)

7.8 ⫾ 0.2 (3) 10.2 ⫾ 1.2 (12) 5.8 ⫾ 0.5 (9) 5.1 ⫾ 0.7 (14) 1.5 ⫾ 0.2 (13) 1.08 ⫾ 0.10 (9) 0.72 ⫾ 0.10 (14)

8.3 ⫾ 1.1 (13) 11.9 ⫾ 1.4 (12) 6.4 ⫾ 0.8 (13) 5.4 ⫾ 0.6 (11) 1.4 ⫾ 0.1 (7) 1.07 ⫾ 0.09 (8) 0.76 ⫾ 0.10 (13)

8.0 ⫾ 0.7 (9) *9.0 ⫾ 1.2 (13) *4.4 ⫾ 0.7 (16) 4.8 ⫾ 0.6 (12) 1.3 ⫾ 0.2 (15) *1.40 ⫾ 0.14 (10) *0.26 ⫾ 0.13 (50)

NS P ⬍ 0.05 NS P ⬍ 0.001 P ⬍ 0.001 P ⬍ 0.05 P ⬍ 0.005

NS NS NS P ⬍ 0.05 P ⬍ 0.005 NS NA

Note. The final column gives the significance of analysis of variance tests of metabolite concentrations from anterior, middle, and posterior mesial temporal voxels, with (ANCOVA) and without (ANOVA) use of grey matter fraction as a covariate. NS, not significant; NA, not applicable; NAc, N-acetyl aspartate ⫹ N-acetyl aspartyl glutamate; Glx, glutamate ⫹ glutamine; Cr, creatine ⫹ phosphocreatine; Ins, myo-inositol; Cho, choline-containing compounds; grey(grey ⫹ white) refers to the fractional brain content of grey matter estimated using SPM 96. * Indicates P ⬍ 0.001 in paired t tests of posterior mesial and lateral temporal lobes.

506

MCLEAN ET AL.

FIG. 3. Field maps acquired over the 12-mm-thick slice used in the two acquisition strategies. (A) In the bilobe approach, hot spots appear over both temporal bones, with contours extending across both hippocampi as well. (B) In the unilobe acquisition, a lesser hot spot appears over the temporal bone on the side being studied, with fewer contours across the ipsilateral hippocampus.

summarized in Table 1. The coefficient of variation in the measurements ranged from 3 to 20%. As expected, the range of measurements was greatest for Glx, which is difficult to determine reliably. The relative tightness of the control range makes it possible to look for variations in metabolite content along the length of the hippocampus, although of course extrahippocampal tissues also contribute significant signal to these mesial temporal voxels. Grey matter content was higher anteriorly (P ⬍ 0.005); therefore, the effect of using grey matter content as a covariate in the analysis was also investigated. We observed a downward trend in NAc/(Cr ⫹ Cho) toward the hippocampal anterior (P ⬍ 0.05), but the significance was lost when grey matter was included as a covariate. The concentrations of NAc and Cr did not vary significantly, but there was a significant increase in Cho toward the anterior (P ⬍ 0.001), which may explain the observed variation in NAA/(Cr⫹ Cho). Myo-inositol (P ⬍ 0.001) and Glx (P ⬍ 0.05) also were observed to increase toward the anterior hippocampus. With grey matter as a covariate, the anterior elevation remained significant for Cho (P ⬍ 0.005) and Ins (P ⬍ 0.05), but not for Glx.

Lateral Temporal Cortex Metabolite Concentrations As discussed under B 0 homogeneity, it was thought to be appropriate to investigate metabolite concentrations only in the posterior voxels of the lateral temporal cortex (Table 1). The tabulated values again are the mean of left and right sides, where both are of acceptable quality; if one side is rejected the contralateral side alone is used. The lateral temporal cortex voxels analysed contained predominantly white matter (mean grey matter fraction 0.26); therefore, many of the differences observed between mesial and lateral voxels may be attributed to this difference. Paired t tests showed significantly higher NAc/(Cr ⫹ Cho) in the lateral than mesial temporal cortex, which appeared to be due to low lateral Cr. Similarly, lower Glx and Ins were found in the lateral temporal lobe. DISCUSSION Methodological Considerations We have found the bilobe acquisition technique most commonly used for long echo time MRSI to be inappli-

TEMPORAL LOBE MRSI

cable for quantitative studies at short echo times on our system. Like Ng et al. (1994), we have found it necessary to make the sacrifice of extending the scanning time in order to obtain better homogeneity over the temporal lobes by scanning them individually. Perhaps this would not be necessary with the addition of higher-order shims: on our system only adjustment of X, Y, and Z shims is available. An additional advantage of the unilobe approach is that if there is a slight asymmetry in head position, it is possible to center the acquisition of the two lobes on different axial slices. This improves the consistency of voxel placement, and avoids the necessity of placing the ROI on one side too close to the temporal bone in order to include all of the contralateral hippocampus in the same slice. Our qualitative assessment of the field maps produced from the slice of interest confirms that the major cause of inhomogeneity is the proximity of the temporal bone below the slice and not the proximity of the sinus anterior to the PRESS voxel. Invariably hot spots were visualized above the temporal bone, with contours radiating through the hippocampi in the case of bilobe acquisition. No T 1 and T 2 corrections were applied, because they would be relatively small given the short TE and long TR used, and the corrections would necessitate very lengthy measurements. As an example, for an in vivo T 2 of 190 ms and T 1 of 1550 ms for creatine (Choi and Frahm, 1999), the signal lost at TE/TR ⫽ 136/2000 ms would be 64%, whereas at TE/TR ⫽ 30/3000 ms only 26% of the signal would be lost. Additionally, the true corrections for absolute quantification would be even smaller, as they depend only upon the difference between the in vivo relaxation times and the in vitro relaxation times of the LCModel basis set. One of the main MRI signs of hippocampal sclerosis is elevation in the T 2 of water, so it is not unreasonable to suspect that there might be additional abnormalites in the T 2 and T 1 relaxation times of some or all of the metabolites studied in patients. The use of short echo times and long repetition times minimizes this effect upon metabolite estimation, although it does not remove it completely. Our application of corrections from partial volume effects arising both from CSF inclusion and from proximity to the edge of the ROI were found to decrease the coefficient of variation of the control mesial temporal data, and thus to be necessary for quantitative analysis. Indeed, voxels near the edge of the excitation profiles would be expected to have incorrect metabolite ratios as well as concentrations, due to the effects of chemical shift artifact. In studies of other parts of the brain, this problem can sometimes be minimized by exciting a PRESS volume much bigger than the real region of interest, and discarding the outer rows and columns. In the temporal lobe, the proximity of sinus and scalp makes such an approach undesirable.

507

Apart from the correction for the amount of CSF present in the voxels, an additional source of variability remains due to variation in tissue content in the voxels. The overall nominal volume of the three mesial temporal voxels analyzed was over 6 cc, which is roughly twice the volume of the hippocampus. Due to the shape of the hippocampus, it is likely that less than 50% of the posterior voxel was hippocampus, and more than 50% of the anterior voxel. Other tissues present included the parahippocampal gyrus, temporal lobe white matter, amygdala, and thalamus. The concentration of NAA has been suggested to be relatively high in the thalamus (Pouwels and Frahm, 1998); the variation in metabolite content among the other tissues has not been studied thoroughly, due to even worse magnetic susceptibility effects in these areas than in hippocampus. From studies of cerebral cortex, it has been shown that metabolite differences between grey and white matter appear to exceed regional differences in grey matter (Noworolski et al., 1999; McLean et al., 2000). We investigated whether a similar effect might be present in mesial temporal structures by performing ANCOVA with grey matter fraction as a covariate, which appeared to explain much of the variation seen. However, the small number of data points (and the limited accuracy of segmentation, given that no correction for the point spread function was used) precluded the possibility of determining with certainty between the models of grey/white matter variation and a full model of all the different tissues included. We believe that this is the best that can be achieved with the current limitations in spatial resolution of MRSI. Our method of analysis of the MRSI datasets is novel: by shifting the data to be aligned with anatomic landmarks, we eliminate a major source of variability in our control group. Evidence of our consistency is found in the small variation in grey matter content in our anatomically defined voxels: if voxel placement were arbitrary, depending entirely on head position in the coil, it would be impossible to attain this level of consistency. Mesial Temporal Lobe Metabolite Concentrations It has been suggested that NAA/Cr and NAA/Cho are higher in the left temporal lobe, reaching significance only in right-handed subjects (Bernard et al., 1996). We found no such tendency: the asymmetry indices of all metabolites were insignificantly different from zero. This is in agreement with single voxel studies of the hippocampus from our own group and others (Choi and Frahm, 1999; Woermann et al., 1999). Previous spectroscopic imaging studies have alluded to the possibility of anterior-posterior trends in hippocampal metabolite concentrations. It should be noted that no study to date, including this one, has obtained data from voxels exclusively within the hippocampus

508

MCLEAN ET AL.

along its entire length: its small size precludes the attainment of sufficient signal for quantitative analysis, even at high field strengths. Therefore inferences must be drawn from voxels containing typically 50% or less hippocampus. For example, the finding of lower NAA/(Cr ⫹ Cho) toward the anterior of the hippocampus (Vermathen et al., 2000) might be explained by the use of a relatively thick slice (15 mm), which increased the contribution of extrahippocampal tissue, particularly posteriorly where the hippocampus is thinnest. In a later report from the same group (Schuff et al., 2000) the authors segmented the hippocampus manually, and correlated metabolite concentration with hippocampal content of voxels. They found NAA to be inversely correlated with hippocampal content, and concluded that NAA concentration in hippocampus is lower than in surrounding tissues. Similar conclusions were reached in a recent study at 4.1T (Chu et al., 2000): although they used a thinner slice (10 mm), it was still thicker than the posterior hippocampus, and the authors speculated that the increased posterior NAA/Cr was due to increased contribution from extrahippocampal tissues. We found a trend of lower NAA/ (Cr ⫹ Cho) toward the anterior of the hippocampus, but it appeared in our study to be due more to increased Cho than to changes in NAA or Cr (Table 1). Glx and Cr have been found to correlate positively with parietal grey matter content (Pan et al., 1996; Noworolski et al., 1999; McLean et al., 2000); in the present study they were highest in the anterior mesial temporal lobe, which had high grey matter content. When grey matter was included as a covariate, Cr and Glx did not vary significantly along the length of the hippocampus. On the other hand, anteroposterior variation in Cho and Ins remained significant when grey matter was used as a covariate. Mesial temporal grey matter appears therefore to have a higher Cho content than grey matter in other brain regions, in agreement with our single voxel studies, which show higher Cho in hippocampal single voxels (1.5 mM ⫾ 0.2; n ⫽ 15) than in single voxels from frontal cortex (1.15 mM ⫾ 0.1; n ⫽ 5) or occipital cortex (0.9 mM ⫾ 0.1; n ⫽ 10; data not shown). The higher Glx toward the anterior hippocampus is interesting in light of the region’s greater tendency toward epileptogenicity. In our single-voxel study (Woermann et al., 1999) we have shown increases in ipsilateral Glx in patients who have temporal lobe epilepsy without any signs of hippocampal sclerosis. However, we are unable using the current techniques to discriminate between Glx and macromolecule resonances: it is anticipated that macromolecules contribute both to the increased concentrations and to the increased variability of Glx toward the anterior. Current work is in progress to develop editing techniques to resolve glutamate and glutamine from each other

and from the macromolecule signal that contributes to the baseline at short echo times. Lateral Temporal Lobe Metabolite Concentrations In parietal cortex (McLean et al., 2000), we have shown that the concentrations of Cr, Glx, and Ins correlated strongly with grey matter content; perhaps this is also the case with mesial temporal grey matter and may explain why these metabolites were found to be more concentrated in mesial than in lateral temporal lobe. This appears to be consistent with the findings of Wang et al. (1999), who found higher NAA/Cr in lateral than in mesial temporal lobe: we suggest that this was likely due to lower Cr content of temporal white matter than grey. These results emphasize the importance of understanding tissue content of spectroscopic voxels in analysis of temporal lobe abnormalities. Not only concentrations but also ratios of metabolites were markedly different in mesial grey matter and lateral white matter; therefore, spectra from voxels which contain an unknown mixture of the two tissue types would be difficult to interpret. CONCLUSION Short echo time MRSI has been applied to study metabolite concentrations in the human temporal lobes. The reliability of the technique was found to increase substantially when the two temporal lobes were studied separately, due to the improvement in achievable B 0 homogeneity. The scatter in measurements was also reduced by application of corrections for partial volume effects. A novel processing method has been developed to ensure the consistent analysis of matching anatomic regions. Significant anteroposterior trends in mesial temporal metabolite concentrations have been observed, which are consistent with our findings of correlations between metabolite concentrations and grey matter content in other areas of the brain. No differences were found between left and right temporal lobes. It was not possible to quantify metabolite concentrations reliably in the anterior lateral temporal lobe due to residual problems with homogeneity. However, analysis of metabolites in the posterior lateral temporal lobe showed lower concentrations of Cr, Glx, and Ins, and higher Nac/(Cr ⫹ Cho) than in mesial temporal lobe, which seems consistent with the predominantly white matter content of the lateral voxels. This study will form an important basis for the analysis of abnormalities in patients with epilepsy. ACKNOWLEDGMENTS The authors thank the National Society for Epilepsy, Medical Research Council, SmithKline Beecham, and the Multiple Sclerosis Society of Great Britain and Northern Ireland for their support.

TEMPORAL LOBE MRSI

REFERENCES Bernard, D., Walker, P. M., Baudouin-Poisson, N., Giroud, M., Fayolle, H., Dumas, R., Martin, D., Binnert, D., and Brunotte, F. 1996. Asymmetric metabolic profile in mesial temporal lobes: Localized H-1 MR spectroscopy in healthy right-handed and non-righthanded subjects. Radiology 199: 381–389. Brand, A., Richter-Landsberg, C., and Liebfritz, D. 1993. Multinuclear NMR studies on the energy metabolism of glial and neuronal cells. Dev. Neurosci. 15: 289 –298. Brown, T. R., Kincaid, B. M., and Ugurbil, K. 1982. NMR chemical shift imaging in three dimensions. Proc. Natl. Acad. Sci. USA 79(11): 3523–3526. Cendes, F., Caramanos, Z., Andermann, F., Dubeau, F., and Arnold, D. L. 1997. Proton magnetic resonance spectroscopic imaging and magnetic resonance imaging volumetry in the lateralization of temporal lobe epilepsy: A series of 100 patients. Ann. Neurol. 42: 737–746. Choi, C. G., and Frahm, J. 1999. Localized proton MRS of the human hippocampus: Metabolite concentrations and relaxation times. Magn. Reson. Med. 41: 204 –207. Chu, W. J., Kuzniecky, R. I., Hugg, J. W., Abou-Khalil, B., Gilliam, F., Faught, E., and Hetherington, H. P. 2000. Statistically driven identification of focal metabolic abnormalities in temporal lobe epilepsy with corrections for tissue heterogeneity using 1H spectroscopic imaging. Magn. Reson. Med. 43(3): 359 –367. Constantinidis, I., Malko, J. A., Peterman, S. B., Long, R. C., Epstein, C. M., Boor, D., Hoffman, J. C., Shutter, L., and Weissman, J. D. 1996. Evaluation of 1H magnetic resonance spectroscopic imaging as a diagnostic tool for the lateralization of epileptogenic seizure foci. Br. J. Radiol. 69: 15–24. Ende, G. R., Laxer, K. D., Knowlton, R. C., Matson, G. B., Schuff, N., Fein, G., and Weiner, M. W. 1997. Temporal Lobe Epilepsy: Bilateral hippocampal changes revealed at proton MR spectroscopic imaging. Radiology 202: 809 – 817. Hetherington, H., Kuzniecky, R., Pan, J., Mason, G., Morawetz, R., Harris, C., Faught, E., Vaughan, T., and Pohost, G. 1995. Proton nuclear magnetic resonance spectroscopic imaging of human temporal lobe epilepsy at 4.1T. Ann. Neurol. 38: 396 – 404. McLean, M. A., Woermann, F. G., Barker, G. J., and Duncan, J. S. 2000. Quantitative analysis of short echo time 1H-MRSI of cerebral grey and white matter. Magn. Reson. Med. 44(3): 401– 411. McLean, M. A., Barker, G. J., Tofts, P., and Duncan, J. S. 1999. Interactive ROI analysis in 1H-CSI. Proc. 7 th Int. Soc. Magn. Reson. Med., p. 1562.

509

Ng, T., Comair, Y. G., Xue, M., So, N., Majors, A., Kolem, H., Luders, H., and Modic, M. 1994. Temporal lobe epilepsy presurgical localization with proton chemical shift imaging. Radiology 193: 465– 472. Noworolski, S. M., Nelson, S. J., Henry, R. G., Day, M. R., Wald, L. L., Star-Lack, J., and Vigneron, D. B. 1999. High spatial resolution 1H-MRSI and segmented MRI of cortical gray matter and subcortical white matter in three regions of the human brain. Magn. Reson. Med. 41(1): 21–29. Pan, J. W., Mason, G. F., Pohost, G. M., and Hetherington, H. P. 1996. Spectroscopic imaging of human brain glutamate by watersuppressed J-refocused coherence transfer at 4.1T. Magn. Reson. Med. 36: 7–12. Pouwels, P. J. W., and Frahm, J. 1998. Regional metabolite concentrations in human brain as determined by quantitative localized proton MRS. Magn. Reson. Med 39(1): 53– 60. Provencher, S. W. 1993. Estimation of metabolite concentrations from localized in vivo proton NMR spectra. Magn. Reson. Med. 30: 672– 679. Schuff, N., Ezekial, F., Du, A. T., Amend, D., and Weiner, M. W. 2000. Automated voxel selection and assessment of partial volume effects in 1H MRSI studies of hippocampus. Proc. 8 th Int. Soc. Magn. Reson. Med., p. 425. Soher, B. J., van Zijl, P. C. M., Duyn, J. H., and Barker, P. B. 1996. Quantitative proton MR spectroscopic imaging of the human brain. Magn. Reson. Med. 35(3): 356 –363. Soher, B. J., Young, K., Govindaraju, V., and Maudsley, A. A. 1998. Automated spectral analysis III: Application to in vivo proton MR spectroscopy and spectroscopic imaging. Magn. Reson. Med. 40: 822– 831. Vermathen, P., Laxer, K. D., Matson, G. B., and Weiner, M. W. 2000. Hippocampal structures: Anteroposterior N-acetylaspartate differences in patients with epilepsy and control subjects as shown with proton MR spectroscopic imaging. Radiology 214(2): 403– 410. Wang, Y., Yang, B., and Ng, T. C. 1999. Differences of metabolites in mesial and lateral temporal lobes. Proc. 7 th Int. Soc. Magn. Reson. Med., p. 1429. Woermann, F. G., McLean, M. A., Bartlett, P. A., Parker, G. J. M., Barker, G. J., and Duncan, J. S. 1999. Short echo time single-voxel 1 H magnetic resonance spectroscopy in magnetic resonance imaging-negative temporal lobe epilepsy: Different biochemical profile compared with hippocampal sclerosis. Ann. Neurol. 45: 369 –376. Xue, M., Ng, T. C., Majors, A., Modic, M., and Kolem, H. 1997. Oblique proton chemical shift imaging for the presurgical localization of mesial temporal epilepsy. Magn. Reson. Med. 37: 619 – 623.