Identification of the Epileptogenic Lobe in Neocortical Epilepsy with Proton MR Spectroscopic Imaging


Address correspondence and reprint requests to Dr. K.D. Laxer at Pacific Epilepsy Program, California Pacific Medical Center, 2100 Webster Street, Suite 115, San Francisco, CA 94115, U.S.A. E-mail:


Summary: Purpose: The aim of this study was to evaluate the usefulness of multislice magnetic resonance spectroscopic imaging (MRSI) in combination with tissue segmentation for the identification of the epileptogenic focus in neocortical epilepsy (NE).

Methods: Twenty patients with NE (10 with MRI-visible malformations, 10 with normal MRI) and 19 controls were studied. In controls, N-acetylaspartate NAA/Cr and NAA/Cho of all voxels of a given lobe were expressed as a function of white matter, and thresholds were determined by calculating the 95% prediction intervals (PIs) for NAA/Cr and NAA/Cho. Voxels with NAA/Cr or NAA/Cho values less than the 95% PI were defined as “pathological.” Z-scores were calculated. Depending on the magnitude of those z-scores, we used two different methods (score-localization or forced-localization) to identify in a given subject the lobe with the highest percentage of pathological voxels, which was supposed to represent the epileptogenic lobe.

Results: MRSI correctly identified the lobe containing the epileptogenic focus as defined by EEG in 65% of the NE patients. MRSI localization of the focus was correct in 70% of the patients with an MRI-visible malformation and in 60% of the patients with normal MRI. Of the patients, 15% had metabolically abnormal brain regions outside the epileptogenic lobe, and 35% of the patients had evidence for secondary hippocampal damage.

Conclusions: MRSI may be helpful for the identification of the epileptogenic focus in NE patients, even in NE with normal MRI.

In 20 to 40% of all patients with partial epilepsy, seizures arise from neocortical structures (1). In >30% of patients with neocortical epilepsy (NE), the seizures are refractory to medical treatment, and these patients may benefit from epilepsy surgery. However, compared with the good results of surgery in patients with medial-temporal lobe epilepsy [mTLE; outcome Engel class I–II in 80–90% (2,3)], surgery in NE is often less successful [(outcome Engel class I–II in 45–60% (4–6)]. This may be because the accurate identification of the epileptogenic focus is more challenging in NE than in mTLE, particularly in NE without a magnetic resonance imaging (MRI)-visible structural abnormality. In addition to initial ictal signs, ictal and interictal electroencephalography (EEG) recordings, often with intracranial electrodes, are traditionally used to localize the epileptogenic focus. However, initial clinical signs may be very subtle and may be overshadowed by more impressive manifestations of seizure spread. Scalp EEG is often not well localized (7,8), which makes the decision where to place intracranial electrodes difficult. Furthermore, results from intracranial recordings may be misleading because of limited spatial sampling. Therefore noninvasive techniques that can assist in localizing the seizure focus take on a greater role in NE. High-resolution MRI, for example, identifies the epileptogenic lobe in NE in ∼55% (9,10), interictal single-photon emission computed tomography (SPECT) in ∼30–50% (9–11), ictal SPECT in ∼56–80% (9,11–13), and positron emission tomography (PET) in ∼33% of the patients with normal MRI but in ≤100% of the patients with acquired structural lesions or tumors (4,9,11,14–17). Compared with these other techniques, few studies used magnetic resonance spectroscopy (MRS) in NE patients (18–21). Furthermore, in all these previous studies, the measurements were restricted to the lobe suspected to contain the epileptogenic focus, as determined by other techniques, and none of them actually tested the ability of MRS to identify the lobe with the epileptogenic focus by measurements in all lobes. In this study, we used a multislice spectroscopic imaging technique (MRSI) covering ∼50–70% of the whole brain in combination with tissue segmentation in NE patients with and without MRI-visible malformations with the following aims: (a) to evaluate the usefulness of MRS for the correct identification of the lobe or lobes containing the epileptogenic focus as determined by EEG; (b) to determine the frequency of metabolically abnormal brain regions outside the epileptogenic lobe; and (c) to seek evidence for secondary damage to the hippocampus in NE. As most patients did not undergo surgery, ictal surface EEG recordings were used as the “gold standard” for comparison with focus localization by MRSI.


Study population

The committee of human research at the University of California, San Francisco (UCSF) approved the study, and written informed consent was obtained from each subject according to the Declaration of Helsinki. From 1996 to 2002, 20 consecutive patients (11 women and nine men) between ages 16 and 42 years (mean age, 25.5 ± 7.5 years) with drug-resistant NE were recruited from the Northern California Comprehensive Epilepsy Center, UCSF, where they underwent evaluation for epilepsy surgery. The identification of the epileptogenic focus was based on seizure semiology and prolonged ictal and interictal video/EEG/telemetry (VET) (10–20 system). Twelve patients had also 18-fluorodeoxyglucose [18FDG]-PET. Two patients (17 and 20) had SISCOM studies, which were inconclusive. All patients had MRI examinations, which showed evidence for a cortical malformation in 10 patients and were normal or showed other abnormalities (e.g., hippocampal sclerosis) in the other 10. Table 1 displays the clinical characteristics of the patients. All patients had been seizure free for ≥24 h before the MRSI examination. The control population consisted of 19 age- and gender-matched healthy volunteers (11 women and eight men), between ages 16 and 46 years (mean age, 29.1 ± 8.7 years).

Table 1. Characteristics of Patients

Patient No
Gender/Age at
Focus in
ictal EEG


  1. M, male; F, female; R, right; L, left; F, frontal; T, temporal; P, parietal; O, occipital; C, central; hypo, hypometabolism; no, normal; nd, not done; SPS, simple partial seizures; HA, hippocampal atrophy.

“Old” protocol
 1M/19L FPL POT subependymal heterotopia, polymicorgyria with schizencephalyL P hypoSPS only/Gliosis
 2F/32L FL TO cortical dysplasiaL TP hypond
 3M/35R TPR TP subependymal heterotopia and polymicrogyria with schizencephalyndnd
 4M/24R+L Tnonond
 5F/25R FnoR T hypond
 6F/28R FCnondnd
“New” protocol
 7F/21R TPnoR T hypoSeizure Free/Cortical Dysplasia
 8M/17R+ L FR F& L P& RP transmantle cortex dysplasiaL T hypond
 9F/31R TOR TO subependymal heterotopia, polymicrogyria with schizencephalyR T hypoSeizure Free/Cortical Dysplasia
 10M/23R FCR motor strip cortical dysplasianoNo improvement/Cortical Dysplasia
 11F/28L FCTnondnd
 12F/27L+R FnoR T hypond
 13F/16L TOL TOP polymicrogyriandnd
 14M/18R OR HAR T + hemisphere hypond
 15F/18L PL P polymicorgyriandnd
 16F/23R OR O cortical dysplasiandnd
 17F/42L+R FCnoL T hypond
 18M/15L FCnondnd
 19M/37L POL TPO subependymal heterotopia, polymicrogyria with schizencepahly. Dandy-Walker-syndromendnd
 20M/31L FnoL TPO hypond

Structural MRI acquisition

All subjects were scanned on a 1.5-T VISION MR system (Siemens Inc., Iselin, NJ, U.S.A.). The following sequence of images was acquired: (a) a double spin-echo sequence (DSE; TR/TE1/TE2, 2,500/20/80 ms timing; 1.0 × 1.4 mm2 in-plane resolution; slice thickness, 3 mm); (b) a volumetric magnetization-prepared rapid gradient echo (MPRAGE; TR/TE/TI, 13.5/7/300 ms timing; 15 degrees flip angle; 1.0 × 1.0 mm2 in-plane resolution; slice thickness, 1.4 mm). The segmentation procedure used for this study is described in detail elsewhere (22,23). In addition to the standard operator-assisted postprocessing of the segmented images done in our laboratory (24), the interhemispheric fissure, the frontal, temporal, parietal, and occipital lobes, insula, brainstem, and cerebellum were manually delineated by using anatomic landmarks (25), thereby further categorizing cortical gray matter and white matter into left and right frontal cortical gray and white matter, temporal cortical gray and white matter, parietal cortical gray and white matter, occipital gray and white matter, and insular gray matter.

[1H]-MRSI acquisition and spectral processing

Multislice [1H]-MRSI data (TR/TE, 1,800/135 ms; 45 min total acquisition) was acquired from three 15-mm-thick slices by using slice-selective inversion recovery (TI, 170 ms) to null the lipid signal and chemical shift–selected water suppression (CHESS). The k-space sampling was accomplished with 36 × 36 circularly bounded phase encoding steps across a 280 × 280 mm2 field of view, yielding a nominal voxel size of ∼0.9 ml. The subjects were studied with two different protocols. From 1996 through March 1997, the “bottom” slice was angulated along the long axis of the hippocampus. The other two slices were aligned along the optic nerve: the “middle” slice was positioned to include the corpus callosum, and the “top” slice, slightly above the corpus callosum (“old protocol”: seven controls, six patients). After March 1997, all three slices were angulated along the anteroposterior axis of the corpus callosum –10 degrees. The “bottom” slice was placed covering the tail of the hippocampus, the “middle” slice right below the inferior aspect of the corpus callosum, and the “top” slice slightly above (“new protocol”: 12 controls, 14 patients). With the “old protocol,” an average of 1,513.4 ± 112.2 voxels/subject and with the “new protocol,” an average of 1,553.1 ± 115.9 voxels/subject were acquired. The “old protocol” provided more voxels from temporal lobes than did the new protocol (125.3 ± 42.4 vs. 93.6 ± 40.5 voxels) but fewer voxels from the occipital lobes (27.0 ± 10.2 vs. 87.0 ± 30.4) and from the parietal lobes (150.3 ± 32.0 vs. 205.7 ± 42.4), and no voxels from the insular region at all (0 vs. 14.8 ± 6.2), cf. Fig. 1.

Figure 1.

T1-weighted images demonstrating the position of the three magnetic resonance spectroscopic imaging (MRSI) slices in (A) the “old protocol” and (B) in the “new protocol.” bm, bottom or hippocampal slice; md, “middle” or ventricular slice; tp, top or supraventricular slice.

The [1H]-MRSI data were zero-padded to 64 × 64 points in the spatial domain and 1,024 points in the spectral domain. Before Fourier reconstruction, the time-domain data were 4 Hz gaussian filtered. Reduction of spurious resonances from extracranial lipids was accomplished by selective k-space extrapolation (26). A fully automated spectral-fitting software package developed in this laboratory (27,28) was used to fit the peak areas of N-acetylaspartate (NAA), creatine/phosphocreatine (Cr), and choline compounds (Cho). Quality control was ensured by rejecting voxels with NAA peaks that had a <4 Hz or >9 Hz line width at half peak height and/or fits with residual sum squares that were outside the upper 95th percentile distribution of residuals from all fits. Typically no more than 10 to 20% of all voxels of the “middle” and “top” slice were rejected by these criteria. However, ∼70% of all voxels in the frontal lobe region (“new protocol”) and 40% of all voxels in temporal lobe region (“old and new protocol”) in the “bottom” slice were rejected. The spectral quality in these regions was adversely affected by susceptibility artifacts. Altogether, in the “new protocol," 82.8% of all voxels fulfilled the quality criteria and were used for further analysis, whereas in the “old protocol,” only 75.9% could be analyzed (p = 0.004, two-tailed Student's t test). NAA/Cr and NAA/Cho of each voxel were calculated to eliminate the effects of cerebrospinal fluid (CSF) inclusion and B0 inhomogeneities. The ratios of NAA/Cr and NAA/Cho were chosen because a previous study in this laboratory in cortical malformations had found these ratios to be abnormal in a high percentage of the voxels.

Identification of metabolically abnormal voxels

The segmented MRIs were aligned with the MRSI slices, by using slice position and orientation information. The tissue composition for each MRSI voxel was then computed by convolving each tissue map of the segmented MRIs with the discrete transform of the MRSI spatial response function and MRSI slice profile, including corrections for chemical shift displacement (24). Because the brain regions covered by the MRSI slices were different between the two protocols, the following calculations were done for each protocol separately. To determine threshold values for voxels with abnormal ratios in a lobe, the 95% prediction interval was calculated by using all voxels in a given lobe in the control group. To this purpose, all voxels from a respective control group containing ≥50% of a lobe (e.g., right frontal lobe or left parietal lobe) were selected. For the insula, voxels with ≥30% insular cortex were selected. Of the selected voxels, those containing >15% cerebellum were excluded to account for the higher concentration of Cr and Cho (29) in the cerebellum. Additionally, all voxels from the frontal lobes in the “bottom” slice of the “new protocol” were excluded because only ∼30% of them satisfied the criteria for good spectral quality. This number was considered to be too small to represent the orbitofrontal region appropriately. NAA/Cr and NAA/Cho of all so-defined voxels of a lobe of interest were then analyzed as a function of white matter percentage (WM%) by using linear regression analysis to calculate the mean metabolite intensity or ratio expected for given WM% in a voxel (Mmean; cf. Fig. 2):


where b0 represents the intercept, bwm the slope of the regression line, and %WM the percentage of white matter in a given voxel. This regression analysis was used to calculate the 95% prediction interval for an additional observation by using the following formula:


where M(p=α) is the threshold value corresponding to a p value ≤α for ratio when α= 0.05; tα (n-2) is the t value for p =α for n−2 degrees of freedom; n is the number of voxels; SEres is the residual standard error; %WM is the percentage of white matter in the observed voxel; %WMmean is the mean white matter percentage of the control population; and SD%WM is the standard deviation of the white matter percentage in the control population.

Figure 2.

Linear regression plots of the ratio of N-acetylaspartate to creatine/phosphocreatine (NAA/Cr) and the ratio of N-acetylaspartate, to creatine/choline compounds (NAA/cr) of the right frontal lobe from the middle slice of the control group as a function of white matter percentage for the two different protocols. Dashed line, 95% confidence interval; b0, intercept; bwm, slope of the regression function.

To account for a rostrocaudal gradient resulting in lower ratios in the lower slices, the thresholds were determined for every slice separately. Voxels with NAA/Cr or NAA/Cho outside the corresponding 95% prediction interval were considered to be “pathological.”

To correct for differences between subjects in the representation of a lobe due to slightly different positioning of the slices and exclusion of voxels of bad spectral quality, the sum of all voxels with either NAA/Cr or NAA/Cho <95% prediction interval (NAAratio↓) in a lobe in a slice was expressed as a percentage of all voxels in this lobe in the slice fulfilling the criteria for good spectral quality. The localizations of NAAratio↓ were indicated on the corresponding MRI slices for anatomic reference.

To account for lobes insufficiently represented in a slice because of bad spectral quality, only lobes in which ≥50% of all voxels present in the slice were of good quality were included in the analysis. To identify, in an individual subject, representations of lobes in a slice with an abnormally high percentage of “pathological voxels,” the percentages were ln-transformed to account for right-skewed data, and z-scores were calculated.

Definition of epileptogenic focus as determined by MRSI

We assumed the following to characterize the lobe containing the epileptogenic focus: (a) in analogy to findings in TLE and cortical malformations, the epileptogenic focus is characterized by decreased NAAratio (NAAratio↓); and (b) metabolic abnormalities in the lobe containing the epileptogenic focus are more widespread than are those in remote regions. Based on these assumptions, the lobe most likely to contain the primary epileptogenic region was identified by going through the following steps:

  • 1A brain region with NAAratio↓ and a z-score ≥ 2 was considered to represent the epileptogenic focus if this was the only brain region in a patient fulfilling those criteria.
  • 2If more than one region in a patient fulfilled those criteria, and two or more of them were contiguous on the reference images, those regions were identified as the epileptogenic focus. This was done to account for the fact that an epileptogenic focus has not necessarily to be restricted to one lobe but may involve two or even more lobes, in which case it would be divided on two or more smaller z-scores and thus eventually not be identified as focus by criterion 1. If they were noncontiguous on the reference images, the region with the highest z-score was identified as the epileptogenic focus. Identification of the epileptogenic focus by 1 or 2 is referred to as “score-localization.”
  • 3If no brain region fulfilled the criteria under 1, the brain region with NAAratio↓ and the highest z-score was identified as the epileptogenic focus. The reference images were used to detect brain regions with NAAratio↓ contiguous with that region. Identification of the epileptogenic focus by 3 is referred to as “forced-localization.”

The localization of the epileptogenic focus by MRSI was defined as “concordant” with the EEG localization if both indicated the same lobe or in cases in which the EEG findings involved more than one lobe, if MRSI identified at least one of the involved lobes. Structurally normal-appearing brain regions with NAAratio↓ and a z-score ≥ 2 not concordant with the epileptogenic lobe, as identified by EEG, were defined as “remote metabolically abnormal brain regions.”


Focus identification by MRSI

None of the controls had brain regions with NAAratio↓ and z-score ≥2, but all except two controls had at least one brain region with a z-score <2. Altogether, the mean z-scores in the control group were significantly smaller (p = 0.04; two-tailed Student's t test) than in the patient group (controls, –0.07 ± 1.0; patients, 0.27 ± 1.3). This indicates that the percentage of pathological voxels in NE was higher than that in controls. Table 2 displays the identification of the epileptogenic focus by MRSI. In five patients, the lobe containing the epileptogenic focus was identified by “score-localization” (“old” protocol, one; “new” protocol, four), and in the other 15, by “forced-localization” (“old “protocol, five; “new” protocol, 10). The identification by “score-localization” was correct in four (80%) patients (“old” protocol, one; “new “ protocol, three), and the identification by “forced-localization” in nine (60%) patients (“old” protocol, one; “new” protocol, eight). Altogether, the identification of the epileptogenic focus as defined by MRSI was concordant with the EEG localization in 65% of NE patients. The percentage of correctly localized epileptogenic foci was higher in patients studied with the “new” protocol than in those studied with the “old” protocol (78.6% vs. 33.3%). The superiority of the “new” protocol for focus localization is probably a consequence of its more balanced brain coverage and the higher number of voxels fulfilling the criteria for good spectral quality (cf. Fig. 3).

Table 2. Identification of Seizure Focus and other metabolically abnormal Brain Regions with MSSI
Localization by decreased
NAAratio and z-score
% of pathological voxels in brain
brain region identified as focus
Concordance with
EEG localization
  1. L, left; R, right; F, frontal; O, occipital; P, parietal; T, temporal; IN, insula; C, concordant with EEG localization; nc, non concordant with EEG localization;

“Old” protocol
 1L P 33.30C
 2 L TP44.40nc
 3 R P 6.60C
 4 L P 7.50nc
 5 LF 8.80nc
“New” protocol
 7R T+IN 38.00C
 8 L F 1.00C
 9 R T+IN 5.50C
 10 R T15.00nc
 11 L F 1.60C
 12 L>R F 8.00C
 13 L T+IN11.10C
 14R>L F 34.70nc
 15 L P 5.40C
 16 L F 6.30nc
 17R>L F+In 25.70C
 18R>L F 24.40C
 19 L P 2.40C
 20 L F 1.80C
 20 L F 1.80C
Figure 3.

A: Patient 7 with NAAratio↓ (see text for definition of NAAratio↓) and a z-score ≥ 2 right temporal, insula, and frontal. Because the reference image shows the first two to be connected, the right temporal–insular region was identified as focus in concordance with the EEG, and the right frontal lobe was identified as “additional metabolically abnormal zone.” Those voxels that are consistent with the electrocorticographic (ECoG) identification of the epileptogenic focus are depicted in black. This patient also shows a cluster of voxels with NAAratio↓ in the right hippocampus and both medial frontal lobes, indicating a secondary involvement of these structures by seizure spread. B: Patient 12 with a left dominant bifrontal focus as identified by “forced localization” in concordance with the EEG. C: Patient 14 with a NAAratio↓ and z-score ≥ 2 in the right frontal lobe, which was not concordant with the focus in the right occipital lobe, as identified by EEG.

In patients with MRI-visible cortical malformations, the identification of the lobe containing the epileptogenic focus was correct in 70%. In patient 2, the left temporooccipital malformation was identified as epileptogenic focus by MRSI with “forced localization.” However, the EEG showed the seizures to originate from left posterior lateral-frontal region. In patient 10, the seizures originated from a cortical dysplasia in the right precentral gyrus, a region not covered by the MRSI slices. MRSI identified a region with NAAratio↓ in the right hippocampus as probable epileptogenic focus by “forced localization.” Fifty-six percent of the voxels in the right parieto-occipital cortical dysplasia identified by EEG as epileptogenic focus in patient 16, had NAA/Cr or NAA/Cho below the 95% prediction interval. However, because these voxels were divided between the parietal and occipital lobes, the two lobar z-scores were lower than those of an additional region with NAAratio↓ in the frontal lobe, which was thus identified as the primary epileptogenic region by “forced localization.”

In patients with unspecific or normal MRI findings, MRSI correctly identified the epileptogenic region in 60%. Patient 4, with bilateral, independent EEG foci in the both temporal lobes, had regions with NAAratio↓ in both temporal lobes, but an additional region in the left parietal region was identified as epileptogenic focus by MRSI because of its higher z-score. MRSI identified the left temporal lobe in patient 5 as epileptogenic focus by “forced localization.” EEG showed a right frontal focus in this patient, who had also a smaller area with NAAratio↓ in this lobe. Patient 6, with a right frontocentral focus in EEG, had noncontiguous regions with NAAratio↓ in both frontal lobes. The z-score was higher on the left side, which was therefore identified as the epileptogenic focus by “forced localization.” Finally, patient 14 had widespread bifrontal contiguous regions with NAAratio↓ with a higher z-score on the right side, but no metabolic abnormality in the right occipital lobe, where the EEG showed the seizures to originate.

Patients with invasive EEG explorations

Four patients (patients 1, 7, 9, 10) underwent further, invasive exploration with electrocorticography (ECoG). In patient 1, a grid was laid over the MRI-visible malformation in the left parietooccipitotemporal area. The exploration demonstrated an epileptogenic zone in a circumscribed region of the malformation in the lower part of the inferior parietal lobule. Functional exploration showed this region to be involved in language production. Therefore this area was not resected, but a biopsy was taken from a neighboring, nonfunctional brain region. In this patient, z-score localization identified the area of MRI-visible malformation as represented in the MRSI slice as the epileptogenic focus. The area of the malformation identified as am epileptogenic zone by ECoG was not covered by this MRSI slice. In patient 7, ECoG identified a small epileptogenic zone of ∼3 cm length in the posterior superior temporal gyrus, which was completely resected. The resected area corresponded well with one of the regions with NAAratio↓. Additional areas were found in the hippocampus and both medial frontal lobes (cf. Fig. 3A). In patient 9, with a large right temporooccipital malformation, the anterior 7 cm of the temporal lobe were resected together with the ipsilateral hippocampal formation and amygdala. MRSI revealed only a small area with NAAratio↓ in the anterior part of the malformation and the insula. However, a separate analysis of the malformation (to be published elsewhere) found large areas with increased NAA/Cho and NAA/Cr within the malformation. In patient 10, as mentioned earlier, seizures originated from a malformation in the right precentral gyrus not covered by the MRSI slices. Because functional exploration showed the malformation to be involved in leg movements, the surgical intervention was restricted to a subpial transection and a small biopsy from a nonfunctional region. Interestingly, this patient, who had frequent simple partial seizures in the left leg, also showed a small area with NAAratio↓ in the rostral cingulate area. This nonprimary motor area has been shown to be activated during voluntary movements in PET studies (30).

Remote metabolically abnormal brain regions

Three patients (7, 14, 17, all studied with the “new” protocol; i.e., 15%) had remote metabolically abnormal regions as defined in Methods (cf. Table 3). These “remote metabolically abnormal brain regions” showed no side preference and could be found in the ipsilateral hemisphere as well as in the contralateral hemisphere. Except for a hippocampal sclerosis in patient 14, all these patients had normal MRI findings.

Table 3. Patients with “Additional metabolically Abnormal Brain Regions”
by EEG
Remote Metabolically
Abnormal Region

  1. L, left; R, right; F, frontal; O, occipital; P, parietal; T, temporal. bold, brain regions with NAAratio↓

7R TPR Fnew
14R OL, P, R Fnew
17FCL O, R Pnew

Visual inspection of the reference images showed seven patients (“old” protocol, two; “new” protocol, five; i.e., 35%), with regions with NAAratio↓ in the hippocampus (cf. Table 4), two of them with bilateral hippocampal NAAratio↓. Four patients had an ipsilateral temporal neocortical EEG focus; the other three had ipsilateral extratemporal EEG foci. Of those four with a temporal EEG focus, two patients had cortical malformations in the same temporal lobe; which involved a part of the hippocampus in one (patient 13). One of the patients (patient 14) with extratemporal EEG foci had evidence for a hippocampal sclerosis in the MRI; the other three had normal MRIs.

Table 4. Patients with Hippocampal Abnormalities
by EEG
with NAAratio↓

  1. L, left; R, right; F, frontal; O, occipital; P, parietal; T, temporal, HA, hippocampal atrophy; no, normal.

“Old” protocol
 3R TPR>LL POT subependymal
 4R+L TL>Rno
“New” protocol
 7R TPRno
 10R FCRR F cortical dysplasia
 13L TOLL TOP polymicorgyria
 17F CRno


The three major findings in this study are as follows: (a) The largest region with NAAratio↓ identified the lobe containing the epileptogenic focus correctly (i.e., concordant with the EEG localization) in 65% of all patients with NE. In patients with an MRI-visible malformation, MRSI localization was correct in 70%, and in patients with nonspecific or normal MRI, MRSI localization was correct in 60%; (b) Metabolically abnormal brain regions (i.e., regions with NAAratio↓ in lobes not containing the epileptogenic focus) were found in 15% of the patients with NE; and (c) 35% of the patients showed a hippocampal area with NAAratio↓ suggesting a secondary involvement of the hippocampus in NE.

The first major finding of this study was that, based on the assumptions that the epileptogenic focus is characterized by NAAratio↓ and that these metabolic abnormalities are more widespread in the focus than in remote brain regions, MRSI identified the lobe containing the epileptogenic focus correctly (i.e., in concordance with the localization by ictal EEG) in 65% of the NE patients. This is considerably higher than the ability of this method to identify correctly the lobe containing the epileptogenic focus by chance, which would be 10%. The analysis of the seven patients with incorrect identification of the epileptogenic focus by MRSI showed that only two had no spectroscopic abnormality in the lobe with the epileptogenic focus. Therefore our first assumption that NAAratio↓ characterizes the epileptogenic focus in NE seems to be correct. The other five patients with false focus localization by MRSI had additional regions with NAAratio↓ concordant with the EEG localization of the epileptogenic focus. These regions had been rejected as representing the epileptogenic focus because other regions with NAAratio↓ had a higher percentage of pathological voxels. Therefore despite leading to the correct identification of the epileptogenic lobe in 13 patients, our second assumption of more widespread metabolic abnormalities in the epileptogenic focus was not always correct. In this context, it also is of interest to note that only 25% of the patients had brain regions with NAAratio↓ large enough to qualify for “score-localization” (i.e., in most cases, focus identification was based on “forced localization”). By definition, “forced localization” will identify a brain region as an epileptogenic focus in a subject as long as just one voxel has a “pathological” NAAratio↓ in the whole data set. Thus “forced localization” should theoretically lead to false localization in a considerably higher percentage of patients than does “score-lateralization.” However, identification of the epileptogenic lobe by “score-localization” and “forced localization” had about the same accuracy in this study. This unexpected finding might be explained by two shortcomings of “score localization.” First, even if the three-slice MRSI technique used in this study covered a relatively large part of the brain, no lobe was fully covered. Additionally, some brain regions (e.g., temporal lobes) were more affected by B0 inhomogeneities, resulting in loss of information from these regions. An epileptogenic focus either partly covered by an MRSI slice or in a region with bad spectral quality may eventually have a lower z-score than a fully covered remote area in a region with good spectral quality and thus not be recognized as a focus by MRSI. A second reason for the epileptogenic focus not to be characterized by the highest z-score is that the focus is not necessarily restricted to one lobe but may include two or even more lobes. Because z-scores are calculated for lobes, such a region would be expressed by two or more smaller z-scores and thus eventually not be identified as epileptogenic focus by MRSI, as demonstrated in patient 17. Therefore although z-scores were useful for focus identification in NE, their limitations must be kept in mind, particularly when “forced localization” is used for focus identification. Additional factors, like positioning of the MRSI slices, spectral quality, and distribution of the pathological voxels on the reference images must be taken into account for their interpretation.

Of the NE patients, 15% had metabolic abnormalities in remote regions found ipsi- and contralateral to the primary epileptogenic region. These results are consistent with PET studies, which found metabolic abnormalities in remote brain areas in ≤80% of the NE patients (4,16).

Several possible mechanisms may explain remote abnormalities. First, some of the remote areas with NAAratio↓ might represent brain regions secondarily involved in seizure spread (23). This hypothesis is supported by the findings in patient 7, in whom ECoG localized the epileptogenic focus in the posterior superior temporal gyrus (cf., Fig. 3A). Because of the anatomic connections of this region, a secondary involvement of the hippocampal formation and the medial frontal lobes is very likely. Second, they might indicate cortical malformations too subtle to be detected by MRI but still associated with neuronal dysfunction. Finally, remote abnormalities also may represent projection areas with functional disturbances due to loss of neuronal input from the epileptogenic focus. Any of these three mechanisms could be associated with intrinsic epileptogenicity and may predict continuing seizures after surgical removal of the focus.

Of the NE patients, 35% (57% with normal MRI) had areas with NAAratio↓ in the ipsilateral hippocampus, indicating a secondary involvement of the hippocampus in NE. This finding contrasts with a previous study from this laboratory that found no evidence for secondary hippocampal damage in NE patients with normal MRIs (31). Because the previous study used hippocampal MRSI, which is more suited to detect hippocampal damage than is the multislice MRSI used in this study, this divergent finding must be explained by differences between the two patient populations.

Secondary hippocampal damage in NE might be of clinical importance. In ∼30% of the patients with medically refractory epilepsy, evidence exists for “dual pathology” (i.e., an extrahippocampal lesion, usually a cortical malformation, associated with radiologic evidence for hippocampal sclerosis) (32). The percentage of NE patients with “dual pathology” might be even higher, because in some patients with extrahippocampal lesions, spectroscopic evidence for hippocampal damage also has been detected in absence of MRI signs for hippocampal sclerosis (33). The fact that surgical outcome in patients with “dual pathology” is best when hippocampus and extrahippocampal lesion can be removed has been interpreted as evidence for both being involved in the generation of epileptogenic seizures (34,35). In this study, two patients had purely extrahippocampal cortical malformations (i.e., fulfilled the criteria for “dual pathology”). The other three patients had normal MRIs. However, this does not exclude the presence of subtle or microscopic malformations. Histopathologic examination revealed a cortical dysplasia in one of them (patient 7), who underwent successful temporal lobe resection without resection of the hippocampus. However, further studies in a larger patient group are needed to determine whether hippocampal abnormalities in NE patients with normal MRIs indicate a hidden malformation or otherwise independent hippocampal epileptogenic activity.

This study has limitations: (a) The focus localization was based on ictal scalp EEG recordings The gold standard for focus identification in NE is a good outcome after epilepsy surgery, but only four patients had surgery. In contrast to ictal scalp EEG, which allows focus identification on a lobar scale only, MRSI allows a precise localization of brain regions with NAAratio↓. However, if for NE, the same is true as for TLE [i.e., that brain regions with NAA reductions correlate well with brain regions with ictal and interictal SEEG abnormalities (36)], then MRSI abnormalities might be helpful for the planning of intracranial EEG exploration in NE. (b) Although using multislice MRSI allows coverage of a large part of the brain, epileptogenic foci only partially covered or outside the MRSI slices may be missed. In such patients, focus localization by MRSI will be wrong. Furthermore, the ability of MRSI to detect foci in the orbitofrontal and temporal regions is limited because of the B0 inhomogeneities in these regions. Using a 3D MRSI sequence could solve the problem of the limited brain coverage; however, the problem of data loss in the basal regions of the brain due to B0 inhomogeneities would persist.

In conclusion, multislice MRSI appears to be a useful tool for the focus localization in NE (i.e., a group of patients in whom the identification of the epileptogenic zone is challenging). In ∼15% of the patients, spectroscopic abnormalities also could be found in remote areas. These remote, metabolically abnormal brain regions might be of clinical importance, because they could potentially represent additional epileptogenic foci.


Acknowledgment:  We thank Drs. D.G. Vossler, R.C. Knowlton from the Swedish Epilepsy Center of the University of Washington, Seattle, and Dr. M.C. Salinsky from the Oregon Health Sciences University Epilepsy Center, Portland, Oregon, for the referral of some of their patients for this study. This work was supported by NIH grant ROI-NS31966 (K.D.L)

S.G.M was supported by a grant from the Swiss National Science Foundation.