Address correspondence to Leonardo Bonilha, M.D., Ph.D., 96 Jonathan Lucas St, 3rd floor CSB, Division of Neurology, Department of Neurosciences, Medical University of South Carolina, Charleston, SC 29425, U.S.A. E-mail: email@example.com
Purpose: It is unclear whether extrahippocampal brain damage in patients with medial temporal lobe epilepsy (MTLE) is a homogeneous phenomenon, as most data relates to the average volume reduction in groups of patients. This study aimed to evaluate where and how much atrophy is to be expected in an individual patient with MTLE.
Methods: High-resolution T1 magnetic resonance imaging (MRI) was obtained from 23 consecutive patients with unilateral MTLE and from a matched control group. Parametric tests of voxel-based gray matter volume evaluated mean regional atrophy in MTLE compared with controls. Gray matter images were then submitted to a voxel by voxel calculation of the fitted receiver operating characteristic (ROC) curve area, plotting the sensitivity versus 1–specificity for a binary classifier (MTLE vs. controls). The area under the curve (AUC) was calculated for each voxel and a resulting three-dimensional map of gray matter voxel-wise AUCs was obtained.
Results: On average, patients with MTLE showed atrophy in the ipsilateral hippocampus and on a limbic network. Elevated AUC was demonstrated in the ipsilateral hippocampus and medial temporal lobe, the ipsilateral thalamus and occipitotemporal cortex, the ipsilateral cerebellum, the cingulate, the contralateral insula, and the occipitoparietal and dorsolateral prefrontal cortex.
Conclusion: This study suggests that the medial temporal lobe, occipitotemporal areas, the cerebellum, the cingulate cortex, the ipsilateral insula, and thalamus are more likely to be atrophied in randomly selected patients with MTLE. Structures such as the orbitofrontal cortex, the contralateral medial temporal areas and insula, the putamen, and the caudate may be atrophied, but not as consistently.
Medial temporal lobe epilepsy (MTLE) has been traditionally associated with atrophy involving the hippocampus and the medial temporal lobes (Bonilha et al., 2003) (Andermann, 2003). This pattern of atrophy is frequently noticeable on routine magnetic resonance imaging (MRI) scans (Jack et al., 1990), suggesting the presence of hippocampal sclerosis, which is the most common histologic abnormality associated with MTLE (Margerison & Corsellis, 1966). Routine visual inspection of MRI in patients with MTLE does not usually indicate the presence of brain atrophy beyond the medial temporal lobes, but new computational studies using high-resolution MRI have suggested that there is more brain atrophy in MTLE than what meets the eyes. Recent improvements in computer-aided calculations of brain structure using T1 MRI (such as manual morphometry, voxel-based morphometry, and cortical thickness) have enabled the quantification of gray matter volume in stereotaxically defined brain voxels, surface curvature, or regions of interest (Bernasconi et al., 2004; Bonilha et al., 2004, 2005; Keller & Roberts, 2008; McDonald et al., 2008). These studies have consistently demonstrated that MTLE is associated with a large degree of brain atrophy that extends further into the temporal lobes and limbic areas. As a group, patients with MTLE display a network of brain atrophy that involves predominantly the limbic system, with more intense atrophy in regions that are functionally and anatomically connected to the hippocampus (Bonilha et al., 2003, 2004; Mueller et al., 2006).
Interestingly, the mechanisms governing the appearance and progression of atrophy are not well understood, but it has been postulated that the excitotoxic effects of seizure spread or deafferentation from loss of hippocampus efferent fibers may play a role (Sutula et al., 2003). Many studies have also confirmed that extrahippocampal atrophy in MTLE is not an incidental finding. Indeed, the degree of extrahippocampal brain atrophy may be related to the severity of epilepsy (Bonilha et al., 2006; Coan et al., 2009; Labate et al., 2009), and to memory deficits (Focke et al., 2008).
The prevalence of brain atrophy in patients with MTLE, however, is not yet known. Although it is accepted that MTLE is generally related to extrahippocampal atrophy, this conclusion comes from studies that have collectively investigated the pattern of atrophy within MTLE as a group (Keller et al., 2002; Bernasconi et al., 2004; Bonilha et al., 2004). To date, all studies have compared, on average, how the group of patients with MTLE differs from controls. The frequency of these abnormalities “individual by individual” is not yet known. It is uncertain if MTLE is universally tied to extrahippocampal atrophy, or if only a few selected patients have abnormally worse atrophy. Because group studies usually rely on the identification of the average effect, it is possible that a few individuals with more extreme atrophy can drive the group pattern. Although it is now believed that MTLE can be associated with significant limbic atrophy, some clinically relevant questions remain unknown. How frequently does extrahippocampal atrophy happen in MTLE patients? Do all patients show a comparable pattern of atrophy regarding the anatomy of damage? Is the intensity of atrophy similar across different patients?
In this article, we aimed to investigate how common brain atrophy is present in MTLE. We employed a newly defined technique to discern signal to noise from the comparison of gray matter volume between controls and patients with MTLE. With this new technique, we aimed to define where, how much, and how frequently extrahippocampal atrophy occurs in a consecutive group of patients with MTLE.
We studied 23 consecutive patients (mean age 38 ± 11 years, 12 women) who were diagnosed with MTLE according to the parameters defined by the International League Against Epilepsy (ILAE). All patients underwent a comprehensive neurologic evaluation, including video–electroencephalography (EEG) monitoring, and diagnostic MRI (which revealed unilateral or markedly asymmetrical hippocampal atrophy in all cases). Eight patients had right hippocampal atrophy and 15 had left hippocampal atrophy based on clinical visual evaluation. For comparison, we studied a control group of 34 healthy individuals (mean age 33 ± 11 years, 17 women) without any significant past medical history. The control group was similar to the patient group in age (t(55) = −1.9, p = 0.1) and gender distribution (Yates’ chi 0.012, p = 0.91). The Medical University of South Carolina Institutional Review Board committee approved this study. All patients signed an informed consent to participate in this study.
All patients underwent high-resolution MRI in a 3T scanner equipped with an eight-channel head coil, yielding a T1-weighted image with 1 mm isotropic voxels in the sagittal plane with the following parameters: TR 8.1 ms, TE 3.7 ms, flip angle 8°, FOV 256 × 256 mm.
A voxel-based quantification of gray matter volume was performed on the T1 images. Images from the control subjects were used to create a normalized template and tissue priors (of gray and white matter). All images were then submitted to iterative spatial normalization, bias field correction, and tissue segmentation, based on the previously defined tissue priors. These steps were performed employing the software SPM5 (http://www.fil.ion.ucl.ac.uk/spm/software/spm5/) and the toolbox VBM5 (Christian Gaser, http://dbm.neuro.uni-jena.de/vbm/). Finally, normalized, bias corrected, segmented images were submitted to spatial normalization employing a 10-mm isotropic Gaussian kernel. For consistency, the images from patients with right MTLE were flipped across the x-axis, so that all ipsilateral hippocampi to the side of seizure origin would be represented in the left side. This step was chosen, even though there are mild but objective differences between the left and right hemispheres and between left and right MTLE. Conversely, the pattern of extrahippocampal atrophy is fairly consistent across left and right MTLE, with the same structures being atrophied, in general, in the ipsilateral and contralateral hemispheres (Bonilha et al., 2004). For this last reason, we opted to average left and right MTLE to decrease the confidence interval and to obtain a representative sample. Therefore, although these results may be extrapolated to the general MTLE population, mild differences may be observed in patients with left compared with right MTLE.
The preprocessed images were submitted to a group comparison evaluating the voxel-wise differences in gray matter volume between patients and controls. A voxel wise t-test was performed with the software NPM (http://www.sph.sc.edu/comd/rorden/npm/) (Rorden et al., 2007), and the results were corrected for multiple comparisons with a false discovery rate threshold set at p < 0.05. The preprocessed gray matter images were also submitted to a voxel by voxel calculation of the fitted receiver operating characteristic (ROC) curve area. For each voxel comprised within the gray matter map of the brain, an ROC curve was calculated, plotting the sensitivity versus (1 – specificity) for a binary classifier (MTLE vs. controls). An area under the curve (AUC) was calculated for each voxel and a resulting voxel-wise three-dimensional map of gray matter–wise AUCs was obtained. This step was performed with the use of an in-house developed code running under Matlab employing functions of the software SPM5 and its Volumes toolbox (http://sourceforge.net/projects/spmtools).
The t-test map demonstrates where patients, on average, displayed a different volume of gray matter compared with controls. The voxel-wise ROC-AUC map demonstrates the probability that for a randomly selected patient with MTLE, the gray matter volume of that voxel will be reduced in the patient compared to a randomly selected control. Higher values of AUC indicate that, for that voxel, the average sensitivity (for all values of specificity) is high. Sensitivity is the proportion of patients with MTLE correctly classified (out of all patients) by the gray matter volume of a voxel estimated through voxel-based morphometry (VBM). The voxels with the highest AUC will be better (on average) than those voxels with lower AUC at classifying patients with MTLE from controls. The AUC method is related to the nonparametric Wilcoxon two-sample test, a distribution-free test of location shift. The AUC should range between 0.5 (the voxel is no better than chance at classifying patients from controls) and 1.0 (every patient will have reduced gray matter volume for that voxel compared to every control).
Voxel-based morphometry—t-test comparison
As a group, patients with MTLE showed a significant reduction in gray matter volume in the ipsilateral hippocampus and medial temporal lobe. They also showed reduction in gray matter volume in the insula, cerebellum, occipitotemporal cortex, thalamus, cingulate, parietal, and dorsolateral prefrontal cortex. These results are summarized in Fig. 1.
Voxel-wise ROC analysis
Compared to the results from the t-test, some areas atrophied also exhibited a significantly large AUC. In particular, the areas in which the AUC was elevated were the ipsilateral hippocampus and medial temporal lobe, the ipsilateral thalamus and occipitotemporal cortex, the ipsilateral cerebellum, the cingulate, the contralateral insula, and occipitoparietal and dorsolateral prefrontal cortex. These results are summarized in Fig. 2. A high AUC signifies that on average, a patient can be distinguished (from controls) by atrophy in these voxels. It indicates that these voxels exhibit a high true-positive rate (sensitivity) and relative lower false positive rate (1 – specificity) on average, when patients and controls are compared. This concept is illustrated in Fig. 3.
The results involving the medial temporal lobe were also remarkable. Although hippocampal atrophy could be seen bilaterally (more pronounced in the ipsilateral side), the frequency of atrophy (as shown by the AUC maps) was significantly more pronounced in the ipsilateral side, as shown in Fig. 4.
In this study, we aimed to assess the relative individual contribution to the average gray matter atrophy commonly observed in patients with MTLE. This study assessed the reliability of class discrimination between gray matter volume across patients and controls, aiming to determine how commonly and where gray matter atrophy occurs in patients with MTLE. The voxel-wise assessment of gray matter under the form of the AUC of ROC curves showed the areas of gray matter atrophy that most accurately distinguish patients with MTLE. Higher AUC suggests it is more likely that each subject in the MTLE group has a gray matter volume that is lower than the gray matter volume found in controls. This analysis was contrasted with the conventional parametric test comparing group means, which can be influenced by outliers (Rorden et al., 2007). A region with a significant t-test difference does not necessarily imply that that the majority of patients with MTLE has a lower volume, but rather signifies that at least a few patients have significant atrophy.
Previous VBM studies have changed the understanding of MTLE by showing that brain atrophy extends farther than what the visual inspection of MRI suggests (Andermann, 2003; Keller & Roberts, 2008). However, it is crucial for the clinical viewpoint to determine whether this pattern of atrophy is present in the majority of patients and thereby defines the behavior of the disease.
Interestingly, this question cannot be addressed by the comparison of individuals with MTLE (one by one) versus controls with current high-resolution MRI, for a few reasons. First, structural resolution of even the most advanced scanners is still not able to yield information comparable to histology. Second, the forms of structural analysis can be influenced by structural variations in three-dimensional space and small gray matter structures, making it difficult to discern signal from noise. Therefore, when whole-brain analyses are performed with one or a few individuals in one group the confidence interval is large and the results are prone to type II error (false negatives), given the small sample size. Therefore, to overcome this issue, we opted to assess the signal to noise ratio of the differences between groups, focusing on the probability that for a randomly selected MTLE patient, the gray matter volume will be reduced. Our group analysis (MTLE vs. controls) shows a similar pattern on extrahippocampal atrophy as reported by the literature (Keller & Roberts, 2008), but we also demonstrated which areas are more likely to be more frequently atrophied in an MTLE population.
Interestingly, the results from this study show that there is a significant overlap between the regions with group differences (results from the t-test) and the regions with high probability of atrophy for a random patient with MTLE (compared to control) (results from the AUC analysis). This is illustrated in Fig. 5. This finding suggests that most patients with MTLE exhibit atrophy in extrahippocampal and extratemporal structures. Although previous group analyses have suggested that brain atrophy is present in patients with MTLE on average, this study suggests that brain atrophy is a frequent phenomenon in patients with MTLE, further illustrating that the neuronal loss in MTLE may encompass broader regions in temporal and extratemporal areas.
In conclusion, this study shows that atrophy involving the medial temporal lobe, occipitotemporal areas, cerebellum, cingulate, insula, and thalamus are consistently atrophied in consecutive patients with MTLE. Structures such as orbitofrontal cortex, contralateral medial temporal areas and insula, putamen, and caudate can be atrophied, but are not as consistently abnormal. These findings corroborate that MTLE is related to limbic network atrophy, suggesting that a core of atrophy is common to the majority of patients, whereas atrophy in structures adjacent to the limbic network can also occur in some patients.
The authors report no financial or nonfinancial conflicts of interest associated with this study.
We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.