SEARCH

SEARCH BY CITATION

Keywords:

  • Brain;
  • Electroencephalography;
  • Epilepsy;
  • Focal cortical dysplasia;
  • Magnetic resonance imaging

Abstract

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES

Summary:  Purpose: Focal cortical dysplasia (FCD) is a frequent cause of partial epilepsy. Its diagnosis by visual evaluation of magnetic resonance images (MRIs) remains difficult. The purpose of this study was to apply a novel automated and observer-independent voxel-based technique for the analysis of 3-dimensional (3-D) MRI to detect and localize FCD.

Methods: The technique was based on algorithms of the SPM99 software and included the spatial normalization of 3-D MRI data sets to a common stereotaxic space and the segmentation of cortical grey matter. The resulting data sets represented grey-matter density maps where each voxel encoded the grey-matter concentration at the corresponding position in the original MRI. A normal database was set up by calculating and averaging the grey-matter density maps of 30 healthy volunteers. The MRI data sets of seven epilepsy patients with FCD were evaluated retrospectively for dysplastic lesions by voxelwise subtraction of the mean grey-matter density map of the normal database and searching automatically for local and global maxima in the resulting data set.

Results: In all patients, the results of voxel-based 3-D MRI analysis corresponded both to the location of the dysplastic lesions in conventional MRI and to seizure semiology and EEG findings. In one case, surgery was performed, and the diagnosis FCD was supported by histology.

Conclusions: The technique of voxel-based 3-D MRI analysis and comparison with a normal database seems to provide a valuable additional screening tool for the detection of FCD.

Focal cortical dysplasia (FCD; i.e., neuronal derangement due to developmental malformation) was described as a pathologic entity first in 1971 by Taylor et al. (1). The histologic features of FCD range from mild disruption of the cortical organization to more severe forms with marked cortical dyslamination, voluminous balloon cells littered throughout the cortex, and astrocytosis (2). Because of improved capabilities of brain imaging, FCD is more and more frequently recognized as an underlying cause of formerly cryptogenic focal epilepsy. Outcome data from epilepsy surgery suggest that in medically resistant epilepsy patients, the dysplastic cortex must be completely removed to obtain freedom from seizures. Planning of the surgical intervention is, of course, decisively influenced by the detection and delineation of the lesion by imaging (3).

High-resolution magnetic resonance imaging (MRI) is considered the imaging technique of choice and is responsible for the increased detection of FCD in the past years. Other imaging modalities such as positron emission tomography (PET) or single-photon emission computed tomography (SPECT) may add valuable information on metabolic (4) and perfusion changes (5) associated with the lesion, but their inferior spatial resolution per se limits their role in FCD assessment.

MRI features of FCD include cortical thickening, abnormal gyral and sulcal contours (i.e., broad gyri, shallow sulci), and blurring of the grey matter–white matter interface (6,7). To obtain high-quality MRI data that allow the identification of even minor structural abnormalities, three-dimensional (3-D) MRI data sets with thin partition size are increasingly used. However, diagnosis by visual inspection of MRIs can be difficult, and subtle dysplastic lesions (i.e., small areas of discrete cortical thickening) often remain unrecognized (8).

Moreover, visual analysis is subjective and highly depends on the attention and expertise of the observer. Thus there is need for more advanced and objective tools for analysis of MRI data in FCD. The aim of this study was to detect and localize areas of increased grey matter indicative of FCD. For that purpose, a novel automated and observer-independent technique for the analysis of 3-D MRI was applied. Following the principles of voxel-based morphometry (9), this technique primarily included a grey-matter segmentation by using SPM99 (10) and a comparison with a normal database.

PATIENTS AND METHODS

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES

Patients and Normal Database

The 3-D MRI data of seven (four male, three female subjects; mean age, 31; range, 14–51 years) with epilepsy were investigated. Patients were chosen from the Epilepsy Center when FCD was diagnosed based on imaging criteria. The visual analysis of clinical standard high-resolution MRI was done by two experienced neuroradiologists. Clinical data of the patients (age, types of seizures) and electroencephalography (EEG) findings (based on video-EEG telemetry) are summarized in Table 1. The neurologic status of all patients was normal. Anticonvulsant medication (AEDS) were carbamazepine (CBZ; patients 2, 6, 7), valproic acid/lamotrigine (VPA, LTG; patients 3, 5), carbamazepine/topiramate (CBZ/TPM; patient 1), and phenytoin/lamotrigine/phenobarbital (PHT/LTG/PB; patient 4), respectively. Patient 4 underwent surgery. The resection of the dysplastic lesion was based on the concordant results of conventional visual MRI analysis and the method presented here. The other patients did not receive a neurosurgical intervention: in some cases, the patients could not decide to undergo the intervention, and in others, FCDs were located adjacent to eloquent cortical areas; in patient 5, EEG telemetry could not reveal an epileptic focus.

Table 1.  Clinical data of patients
Patient no.Age/sexSymptomatic zone (semiology)EEG findings: irritative area and seizure-onset zoneLocalization of cortical lesion in clinical MRILocalization results of voxel-based 3-D MRI analysis [MNI coordinates x y z]Difference from normal data base (standard deviations)
  1. MRI, magnetic resonance imaging; SPECT, single-photon emission computed tomography; ECD, electron capture detector.

123 yr/MLeft frontal (complex partial seizures with tonic extension of the right extremities, hypermotor elements, rapid reorientation)Left frontal irritative zone (sharp waves F3); left frontocentral seizure-onset zone (repetitive spikes F3-C3)Left frontal lobeLeft frontal lobe [27 21 30]12.99
251 yr/FFrontal (complex partial seizures with bilateral tonic arm extension, bicycling, rapid reorientation)Left frontotemporoparietal irritative zone (sharp waves F7-T3-P5); seizure-onset zone not defined with surface recordingsLeft parietotemporal areaLeft parietal lobe [46 –23 30]8.96
339 yr/FRight frontal (simple partial motor seizures with left face cloni)No epileptiform potentials on surface EEG; (ictal ECD-SPECT showed right frontocentral hyperperfusion)Right frontal lobe (precentral)Right frontal lobe (precentral) [−38 –6 36]8.68
414 yr/MLeft frontal and insular (epigastric aura; complex partial seizures with tonic extension of the right arm, hypermotor elements, responsiveness impaired to a variable degreeLeft frontal irritative and seizure onset zone (continuous and repetitive spiking F7, FT7)Left frontal lobeLeft frontal lobe [32 29 21]6.78
537 yr/MWithout clear localization (cephalic aura, complex partial seizures with oral and manual automatisms)No epileptiform potentials on surface EEGLeft occipital lobeLeft occipital lobe [24 –72 2]6.40
636 yr/MFrontal (mostly complex partial hypermotor seizures)Right frontotemporal irritative zone (sharp waves F8/T2); nonlateralized frontocentral seizure-onset zone (rhythmic delta or beta activity)Right frontopolar areaRight frontal lobe [−29 49 1]3.27
718 yr/FRight frontal (complex partial seizures with head version to the left, tonic extension of the left arm)Right frontocentroparietal irritative zone (sharp waves maximum C4-P4); right frontocentral seizure-onset zone (rhythmic delta activity)Right frontal lobeTwo local maxima: right frontal lobe [−18 13 57] [−31 10 48]2.54 2.70

An age-matched normal database was set up by calculating grey-matter density maps (cf. Methods) of 30 healthy volunteers (17 male, 13 female). The mean age of the volunteers was 31 years (range, 15–52 years). All volunteers were neurologically and neuropsychologically normal; none had any history of neurologic or psychiatric disease.

METHODS

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES

MRI was performed on a 1.5-T scanner (Vision; Siemens, Erlangen, Germany) with a transmit/receive full-head standard head coil. The 3-D MRI data set consisted of 160–180 T1-weighted sagittal slices. The sequence was magnetization-prepared rapid-acquisition gradient echo (MP-RAGE) with time to repeat of 9.7 ms, time to echo of 4 ms, flip angle of 12 degrees, matrix of 256 × 256 mm2, and voxel size of 1 × 1 × 1 mm3.

MRI data were analyzed by using statistical parametric mapping software (SPM99; Wellcome Department of Cognitive Neurology, London, U.K.). For FCD detection, a screening 3-D MRI analysis technique was used based on the principles of voxel-based morphometry, as described by Ashburner and Friston (9). The 3-D MRI data sets were spatially normalized to a common stereotaxic space (voxel size 1 × 1 × 1 mm3). For this purpose, the standard brain by the Montreal Neurological Institute (MNI) was used. The first step involved matching the MRI by estimating the optimal 12-parameter affine transformation. In a second step, global nonlinear shape differences were modeled by linear combination of 7 × 8 × 7 smooth spatial basis functions. It should be noted that this method of spatial normalization does not attempt to match every cortical feature exactly, but corrects only for global brain shape differences (11). Then the cortical grey matter was automatically segmented by using the SPM segmentation algorithm (9). The grey-matter segment was smoothed by using a fixed gaussian kernel of 6-mm full width at half maximum (FWHM; i.e., about the size of the lesions to be detected). The resulting data sets represented grey-matter density maps in which each voxel encoded the average concentration of grey matter from around the voxel (defined by the form of the smoothing kernel) at the corresponding position in the original structural MR image.

Data acquisition and processing for the normal database were the same as described earlier for the patient data sets. A mean grey-matter density map was calculated by averaging the single density maps.

The grey-matter density maps of the seven epilepsy patients were evaluated for dysplastic lesions by subtracting the mean grey-matter density map of the normal database voxel by voxel and searching automatically for local and global maxima in the resulting difference images. These maxima were compared with the standard deviation of the values at the corresponding locations in the single density maps of the normal database. For quantification of the differences, the multiplication factor by which the grey-matter density value at the maximum of the difference image exceeded the standard deviation in the density maps of the normal database was determined.

A last step aimed at an improved visualization of the results. To emphasize those locations where the grey-matter density map of the patient differed clearly from the mean density map of the normal database, all values of the difference map lower than one standard deviation were set zero, and the remaining values were squared.

For control purposes, the 3-D MRI data of the 30 subjects constituting the normal database were analyzed in the same way as the patients, and each resulting difference image was checked for the global maximum.

RESULTS

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES

In six of seven patients, the global maximum, which represented the highest difference in grey-matter density between the patient's data and the normal data base, corresponded to the location of the dysplastic lesion as diagnosed by conventional visual analysis of the MRI data. The difference of grey-matter density at these locations exceeded the concentration as expected from the normal database by a mean of 7.77 standard deviations (range, 3.27–12.99 standard deviations). The global maxima were localized in the frontal cortex in four cases, in the parietal cortex and in the occipital cortex in one case each; four were left, and two were right hemispheric. For a synopsis of the localizations of individual global maxima and values of grey-matter density in comparison with the normal database together with clinical data, EEG, and clinical MRI results cf. Table 1. Patient 4 underwent surgery with resection of the frontal lesion identified as FCD by results of imaging including voxel-based 3-D MRI analysis and of EEG analysis. Histologic examination of the specimen showed brain tissue with atypically formatted shallow gyri and abnormal cytoarchitecture, and FCD/microdysgenesis was diagnosed. Thus in this patient, the diagnosis of the cortical lesion as identified by 3-D MRI analysis was histologically proven. Noticeably, this patient had received two previous MRI investigations in other hospitals without detection of a focal lesion. Furthermore, in five of six patients, the global maxima in 3-D MRI analysis were in agreement with the epileptic foci as identified by EEG analysis (cf. Table 1). In patient 5, no clear localizing hints for the seizure-onset zone could be drawn from semiology and EEG analysis.

In patient 7, the dysplastic lesion as diagnosed by conventional visual analysis corresponded to two local maxima of the difference map, both in the right frontal lobe and correlating with EEG results, whereas at the site of the global maximum, no dysplasia could be ascertained (Table 1).

Representative images of the results after different steps of the 3-D MRI analysis are shown in Fig. 1A–D for four different patients. In each figure part, MRI slices of the original 3-D data sets, normalized to the common stereotaxic space, are shown in the first row, grey-matter density maps after automatic segmentation in the second row, the difference images after voxel-wise subtraction (smoothed data sets of the patients—mean grey-matter density map of the normal database—in the third row, and the difference images after squaring of the values differing from the mean density map of the normal data base by ≥1 standard deviation. The crosshairs mark the global maxima of the “difference image” and the corresponding positions in the other maps. In the control analysis of the 3-D MRI data sets of the normal database, the difference of grey-matter density at these maxima exceeded the density as expected from the normal database by 1.79 to 4.14 standard deviations, with a mean of 2.98.

imageimageimageimage

Figure 1. Results of voxel-based 3-D magnetic resonance imaging (MRI) analysis in four patients: column 1 for AF (patient 3 in Table 1), column 2 for MG (patient 1), column 3 for RB (patient 2), column 4 for SB (patient 4). All images are shown as image slices in coronal, sagittal, and axial views. The cross-hairs mark the global maxima of the “difference image” and the corresponding positions in the other maps. Different steps of the data processing are presented from row 1 to 4. Row 1 shows the normalized 3-D MRI data sets (MP-RAGE). Row 2 shows the grey-matter density images after automatic segmentation. Row 3 displays the difference images resulting from the comparison with the normal database by subtraction of the mean grey-matter density image of the healthy volunteers. At this step, an automatic search for maxima is conducted. Row 4 shows the differences in grey-matter concentration accentuated by squaring all values of the difference image that exceed one standard deviation of the normal database at the corresponding location. For clinical data of the patients, details of maxima locations, and scores, cf. Table 1.

DISCUSSION

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES

In this study, we applied an automated and observer-independent technique using voxel-based 3-D MRI analysis for the detection of FCD in epilepsy patients. The method compared the 3-D images of the grey-matter segments of individual patients (where each voxel represents the concentration of grey-matter at the corresponding 3-D MRI location) voxel by voxel with a mean grey-matter density image of a normal database within a standardized space. The maxima of the resulting difference images were regarded as locations that might represent focal cortical abnormalities. In six of seven patients investigated so far, the global maxima identified by this technique coincided with the regions regarded as FCD according to visual inspection of MRI. A comparison of the results of 3-D MRI analysis and the histologic analysis of the neurosurgically removed lesion was obtained in one patient and confirmed the diagnosis.

This novel analysis is assumed to be a screening technique for the detection and localization of FCD in 3-D MRI. Other analysis methods have been recently described [e.g., voxel-based morphometry (12) or curvilinear reformatting by Bastos et al. (8)]. The voxel-by-voxel approach is a statistically rigorous and completely unbiased method and, unlike region-of-interest analysis techniques, not based on a priori knowledge. Of course, voxel-by-voxel comparisons are based on a comparison with a normal database and therefore require previous normalization of the 3-D MRI data sets to reduce the overall variability in size and shape between brains (13). In general, the spatial resolution of the method is limited by the voxel size of the original 3-D data sets (1 × 1 × 1 mm3 in this study) and the smoothing factor in the further data processing.

The technique is based on the calculation of the grey-matter density at each voxel, as expected from the normal database. In the comparison of the patients' data with the normal database [which must be as well age-matched as possible to avoid influences of aging processes (14)], one global maximum and further local maxima are found in each resulting difference image. As a score for the relevance of these maxima, the grey-matter density in the difference image is compared with the standard deviation of the grey-matter density in the normal database at the corresponding location.

In healthy subjects, of course, there also will be areas where the grey-matter density differs from the mean of the normal database. For control purposes, the 3-D MRI data of the 30 healthy subjects constituting the normal database were analyzed in the same way as the patients, and the scores for the individual maxima were calculated as described earlier. Because of the variability of the brains of healthy humans, these scores reached values up to 4.14 standard deviations. In single patients, the scores at the global maxima were below this upper limit as derived from the normal database. Thus in single cases, the global maximum identified by voxel-based 3-D MRI analysis may not necessarily represent pathologic findings. However, the mean value for the global maximum of the within-normal-base analysis was 2.98 standard deviations. Scores of >6 standard deviations (as calculated in five of seven patients) therefore have a high probability of representing pathologic cortical lesions. Nevertheless, only descriptive statistical methods were used for this screening tool to keep the analysis simple, easily applicable, and less time consuming. Statements on sensitivity and specificity have to await future investigations with higher numbers of patients included.

It is important to note that this technique not only tries to detect subtle cases in which the visual examination fails to find a lesion, although in patient 4, the focal lesion had not been detected in two previous MRI investigations performed in other hospitals. The method is also intended to serve as a screening tool to help save time by providing initial information on the location of possible dysplasia.

Obviously, the diagnosis of FCD as the epileptogenic lesion and consequently the decision about epilepsy surgery can never rely on one diagnostic tool alone. However, with respect only to brain imaging, MRI seems to be most important. Different applications of MRI might be used (e.g., diffusion tensor imaging obtains additional information about diffusion changes in tissue around the FCD) (13, 15). Irrespective of the chosen analysis technique, a 3-D approach seems to be advantageous and has been reported to be superior to conventional interpretation of planar MRI in FCD detection (16). In addition, an automated and objective voxel-based 3-D MRI analysis technique, including a comparison with a normal database, appears to be a promising additional screening tool for the detection of FCD.

REFERENCES

  1. Top of page
  2. Abstract
  3. PATIENTS AND METHODS
  4. METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES