MRI segmentation analysis in temporal lobe and idiopathic generalized epilepsy

This was a retrospective study conducted at the Buffalo Neuroimaging Analysis Center (BNAC) and the Comprehensive Epilepsy Program at the Jacobs Neurological Institute, Department of Neurology, State University of New York at Buffalo, with approval of the study protocol by the institutional review board (IRB). The study consisted of comprehensive review of medical records. Brain MRI segmentation analysis was performed on the previously performed MRI. A waiver of informed consent was obtained from the IRB.

The study included three population groups: TLE patients, IGE patients and healthy controls. The first two patient population groups were retrieved through a patient epilepsy monitoring unit (EMU) database following IRB approval. All patient demographics were de-identified. The inclusion criteria for TLE patients consisted of: age >18 years at time of MRI, diagnosis of TLE supported by history, documented seizures on EMU long term monitoring (LTM) video electroencephalogram (EEG), and having underwent a 3 T MRI using a standard epilepsy protocol at a single site within 12 months of the LTM. The TLE patients’ were further subdivided into lesional (L-TLE) and non-lesional (NL-TLE) based on the presence or absence of temporal pathology on MRI as identified by the report of a certified neuro-radiologist. The inclusion criteria for IGE patients consisted of: age > 18 years, and supportive ictal findings on LTM. The IGE patients’ MRI were classified as normal or with a low number of non-specific white matter changes not related to the subcortical deep grey matter structures. The exclusion criteria included any MRI-detected structural abnormalities beyond abnormalities seen in TLE that would preclude the segmentation procedure. We enrolled only patients with TLE and IGE that were 18 years and older, as we only had age-matched MRI controls for this age group.

Clinical data of all TLE and IGE patients were obtained from medical history and LTM reports, and included location of epileptic focus (for TLE patients), International League Against Epilepsy seizure classification, frequency of seizures, age at epilepsy onset and duration of disease.

All subjects underwent MRI testing at a single 3 T GE Signa Excite HD 12.0 Twin Speed 8-channel scanner (General Electric, Milwaukee, WI). Volumetric analysis was based on an axial T₁ Inversion Recovery Fast Spoiled Gradient Echo (IR-FSPGR) sequence with flip angle = 20°, repetition time = 9.46 ms, echo time = 3.87 ms, matrix size of 256×256 pixels, and voxels of 1 × 1 × 1.5 mm. The lesions were assessed on 2D scans (proton density [PD]/T2, Fluid attenuated inversion recovery [FLAIR] and spin echo [SE] T1), with 48 slices collected, with a thickness of 3 mm, and no gap between slices.

To all images we applied an automatic inhomogeneity correction [27] to overcome distortions of intensity non-uniformity created by the scanner. The volumetric analyses were performed with the use of FMRIB tools (Oxford Centre for Functional MRI of the Brain, version 4.1) [25, 26]. The volumetric analysis was performed in a blinded manner in regard to the qualitative MRI results provided by the neuro-radiologist.

The first stage of analysis used the structural image evaluation using normalization of atrophy, cross-sectional (SIENAX) [28, 29] to estimate the normalized brain volume (NBV), normalized grey matter volume (NGMV), and normalized white matter volume (NWMV). SIENAX starts by extracting brain and skull images from single whole-head input data [30]. The brain image is then affine-registered to MNI152 space [31, 32], using the skull image to determine the registration scaling factor (to be used as a normalization for head size). Next, tissue-type segmentation with partial volume estimation is carried out in order to calculate the total volume of brain tissue (including separate estimates of grey and white matter volumes) [33]. The second stage of analysis used FIRST (FMRIB’s Integrated Registration and Segmentation Tool) [34‐36] to estimate the volumes of the following subcortical deep grey matter structures in both hemispheres: hippocampus, amygdala, thalamus, putamen, pallidum and caudate. FIRST is a model-based automated segmentation/registration tool. The shape models used in FIRST are constructed from manually segmented images provided by the Center for Morphometric Analysis, Massachusetts General Hospital, Boston. The manual labels are parameterized as surface meshes and modeled as a point distribution model. Deformable surfaces are used to automatically parameterize the volumetric labels in terms of meshes. The deformable surfaces are constrained to preserve vertex correspondence across the training data. Furthermore, normalized intensities along the surface normals are sampled and modeled. The shape and appearance model is based on multivariate Gaussian assumptions. Shape is then expressed as a mean with modes of variation (principal components). Based on learned models, FIRST searches through linear combinations of shape modes of variation for the most probable shape instance given the observed intensities in the T1 image. An example of a segmented brain is presented in Figure 1.

Statistical analysis was performed with R version 3.1.0 (http://​www.​R-project.​org/​). A GLM-based analysis of covariance (ANCOVA) model was used to evaluate group differences in volume measures between controls, IGE, NL-TLE, and L-TLE while controlling for variation in age and gender. Where group was a significant factor, post-hoc pair-wise comparisons were performed to identify specific differences. In the primary set of analyses, total tissue volumes and bilateral structure volumes were compared. In a secondary set of analyses, laterality was evaluated by comparing left/right asymmetry between groups. Asymmetry was calculated as the absolute difference between left and right structures divided by the total volume (left + right). Finally, ipsilateral and contralateral structures (as related to epileptic focus localization) in L-TLE were compared to HC to evaluate whether there was evidence of contralateral atrophy. For this final analyses, individual rather than left/right averaged structure volumes were used, so a mixed-effect model was employed using laterality, age, and gender as fixed effects and subject as a random effect. We used a conservative type 1 error threshold of p < 0.01 to correct for multiple testing.

Demographic and clinical information of patients and controls is presented in Table 1. The epilepsy populations were initially composed of 44 patients diagnosed with TLE and 30 patients with IGE. Eighteen TLE and 15 IGE subjects were consequently dropped due to exclusion criteria of having no recorded ictal events during LTM, having multifocal seizure onset (only for TLE), or other MRI-detected abnormalities (brain tumor, multiple sclerosis, sub-optimal MRI study, etc.) that would affect the segmentation procedure. The final study groups consisted of 26 patients with unilateral TLE (15 NL-TLE and 11 L-TLE), 15 patients with IGE and 26 healthy controls. There were no significant differences between the groups’ demographic distributions, other than a female predisposition for IGE patients as compared to TLE and controls. IGE patients were also of notably younger ages.

In the TLE group, 16 had the seizure focus on the left hemisphere and 10 in the right hemisphere (for L-TLE alone, 4 right, 7 left). Fifteen patients in the TLE group had complex partial seizures with secondary generalization where 11 had complex partial seizures without secondary generalization. In the IGE group, 14/15 subjects had generalized tonic-clonic seizures, 11/15 had absence seizure and myoclonic seizures. MRI abnormalities included hippocampal atrophy in 5 patients and other findings in the 6 patients (cortical dysplasia 1, venous anomaly 1, atrophy 2, and non-specifc juxtacortical lesions 2).

Figure 2 compares the tissue-specific volumetric measures between groups. After correcting for age and gender, NBV (F = 13.72, p < 0.001) and NWMV (F = 16.32, p < 0.001) were significantly different between groups. Post-hoc analysis showed that HC had greater NBV and NWMV as compared to all epilepsy groups (p < 0.001). This indicates that the whole brain volume changes in epilepsy are predominantly the result of WM volume loss. Within epilepsy groups, there were no significant tissue-wide differences, although there was a general trend for L-TLE to have the lowest volumes.Figure 3 compares structure-specific volumetric measures between groups. There were significant group effects in the hippocampus (F = 7.18, p = 0.001), amygdala (F = 14.77, p < 0.001), and caudate (F = 4.56, p = 0.006), and a trend in the thalamus (F = 3.95, p = 0.012). Post-hoc analysis between groups in the significant structures revealed lower hippocampal volume in L-TLE compared to both HC (p = 0.001) and IGE (p < 0.001), lower amygdala volume in all epilepsy groups compared to HC (p < = 0.004), and a trend for lower caudate volume in L-TLE compared to HC (p = 0.012) and IGE (p = 0.042).Figure 4 compares asymmetry between structures and groups. There were no statistically significant differences between groups for any structures, although there was a weak trend for L-TLE to have more hippocampal asymmetry than HC (p = 0.071).Figure 5 shows the results of laterality analysis between the L-TLE and HC groups. For the hippocampus, the ipsilateral side was significantly smaller than HC (p < 0.001), with a trend for the contralateral side as well (p = 0.03). Both ipsilateral and contralateral amygdalae were significantly smaller than in HC (p < 0.001). Putamen differences were not significant, but showed trends for both ipsilateral (p = 0.044) and contralateral (p = 0.0101). There were similar bilateral non-significant trends in ipsilateral (p = 0.02) and contralateral (p = 0.02) pallidum. No significant differences were observed in the thalamus or caudate. Within the L-TLE group, only the hippocampus showed a trend toward lower volume in the ipsilateral vs. contralateral side (p = 0.011).

Prior to the initiation of the study, approval was obtained from the Institutional Review Board of the State University of New York at Buffalo.