Diffusional kurtosis imaging reveals a distinctive pattern of microstructural alternations in idiopathic generalized epilepsy

Authors

  • C.-Y. Lee,

    1. Department of Radiology and Radiological Science, Medical University of South Carolina, Charleston, SC, USA
    2. Center for Biomedical Imaging, Medical University of South Carolina, Charleston, SC, USA
    Search for more papers by this author
  • A. Tabesh,

    1. Department of Radiology and Radiological Science, Medical University of South Carolina, Charleston, SC, USA
    2. Center for Biomedical Imaging, Medical University of South Carolina, Charleston, SC, USA
    Search for more papers by this author
  • M. V. Spampinato,

    1. Department of Radiology and Radiological Science, Medical University of South Carolina, Charleston, SC, USA
    2. Center for Biomedical Imaging, Medical University of South Carolina, Charleston, SC, USA
    Search for more papers by this author
  • J. A. Helpern,

    1. Department of Radiology and Radiological Science, Medical University of South Carolina, Charleston, SC, USA
    2. Center for Biomedical Imaging, Medical University of South Carolina, Charleston, SC, USA
    Search for more papers by this author
  • J. H. Jensen,

    1. Department of Radiology and Radiological Science, Medical University of South Carolina, Charleston, SC, USA
    2. Center for Biomedical Imaging, Medical University of South Carolina, Charleston, SC, USA
    Search for more papers by this author
  • L. Bonilha

    Corresponding author
    1. Center for Biomedical Imaging, Medical University of South Carolina, Charleston, SC, USA
    2. Division of Neurology, Department of Neurosciences, Comprehensive Epilepsy Center, Medical University of South Carolina, Charleston, SC, USA
    • L. Bonilha, Division of Neurology, Department of Neurosciences, Medical University of South Carolina, 96 Jonathan Lucas Street, 3rd floor CSB, Charleston, SC 29425, USA

      Tel.: +1 843 792 3222

      Fax: +1 843 792 8626

      e-mail: bonilha@musc.edu

    Search for more papers by this author

Abstract

Objectives

Idiopathic generalized epilepsy (IGE) arises from paroxysmal dysfunctions of the thalamo-cortical network. One of the hallmarks of IGE is the absence of visible abnormalities on routine magnetic resonance imaging (MRI). However, recent quantitative MRI studies showed cortical–subcortical structural abnormalities in IGE, but the extent of abnormalities has been inconsistent in the literature. The inconsistencies may be associated with complex microstructural abnormalities in IGE that are not completely detectable using conventional diffusion tensor imaging methods. The goal of this study was to investigate white-matter (WM) microstructural abnormalities in patients with IGE using diffusional kurtosis imaging (DKI).

Materials and methods

We obtained DKI and volumetric T1-weighted images from 14 patients with IGE and 25 matched healthy controls. Using tract-based spatial statistics, we performed voxel-wise group comparisons in the parametric maps generated from DKI: mean diffusivity (MD), fractional anisotropy (FA), and mean kurtosis (MK), and in probabilistic maps of WM volume generated by voxel-based morphometry.

Results

We observed that conventional microstructural measures (MD and FA) revealed WM abnormalities in thalamo-cortical projections, whereas MK disclosed a broader pattern of WM abnormalities involving thalamo-cortical and cortical–cortical projections.

Conclusions

Even though IGE is traditionally considered a ‘non-lesional’ form of epilepsy, our results demonstrated pervasive thalamo-cortical WM microstructural abnormalities. Particularly, WM abnormalities shown by MK further extended into cortical–cortical projections. This suggests that the extent of microstructural abnormalities in thalamo-cortical projections in IGE may be better assessed through the diffusion metrics provided by DKI.

Introduction

Idiopathic generalized epilepsy (IGE) is defined by recurrent generalized seizures, such as absence, myoclonic, and generalized tonic–clonic seizures [1]. During the ictal and interictal phases, the electroencephalogram (EEG) of patients with IGE typically demonstrates a generalized epileptiform discharges in symmetrically distributed multiple channels, without consistent localization or lateralization [2]. The mechanisms underlying ictogenesis in IGE remain unclear, but it has been postulated that seizures arise from paroxysmal dysfunction of thalamo-cortical networks [3-5].

By definition, IGE is associated with no epileptogenic structural abnormalities on routine diagnostic magnetic resonance imaging (MRI). Occasionally, clinical MRI scans may demonstrate other structural abnormalities such as subcortical microangiopathy or arachnoid cysts, but these are considered incidental findings without relevance to the pathogenesis of IGE. Thus, some epileptologists do not routinely perform MRI studies on patients with a classical history, seizure semiology, and EEG findings of IGE, given the high likelihood of a normal study.

The concept of ‘MRI-negative’ IGE has been challenged by recent studies using quantitative MRI techniques. Tissue volume studies have demonstrated structural abnormalities in thalamo-cortical networks in patients with IGE [6-11]. However, these results have not been fully consistent. For example, some morphometric studies observed an increase in tissue volume in frontal lobes [6-8]. Other studies showed decreased tissue volume [9, 10] or no changes [11] in thalamus and frontal lobes. Diffusion tensor imaging (DTI) studies also reported multiple regional white-matter (WM) abnormalities in thalamo-cortical networks [12-16].

These inconsistent findings of structural abnormalities in IGE have been attributed to heterogeneous genetic backgrounds [17] and different subsyndromes of IGE [6, 18]. We hypothesized that IGE is associated with a complex pattern of microstructural changes that may not be completely detectable using conventional DTI methods. Given its complexity, slight variations in study design may lead to inconsistent findings. Hence, a biomarker that is sensitive to complex microstructure architecture may better assess the extent of structural abnormalities in IGE.

In this study, we investigate WM microstructural abnormalities in patients with IGE using diffusional kurtosis imaging (DKI) [19]. Compared with conventional DTI (with b = 1000 s/mm2), DKI employs multiple b-values (up to b = 2000 s/mm2) to quantify non-Gaussian water diffusion, which may be associated with membrane permeability [20-22] and heterogeneity in cell compartments [20]. Previous studies have shown that DKI may better characterize epilepsy-related tissue changes [23, 24]. We also study WM volumetric abnormalities in IGE using voxel-based morphometry (VBM). We perform voxel-wise comparisons between patients with IGE and healthy controls. We assess the relationship between microstructural abnormalities in IGE and the patients' clinical variables.

Materials and methods

Subjects

Fourteen consecutive patients [mean age ± standard deviation (SD) = 28.9 ± 9.9 years, nine females] diagnosed with IGE were included in this study. IGE was diagnosed according to the criteria defined by the International League Against Epilepsy [2]. All patients underwent a comprehensive neurological evaluation, epilepsy history, seizure semiology, and interictal EEG recordings, which were compatible with IGE for all patients. The Institutional Review Board of the Medical University of South Carolina approved this study.

From all patients, we collected clinical information related to age of onset of epilepsy, frequency of seizures, duration of epilepsy, and lifetime seizure burden (defined as the frequency of seizures multiplied by epilepsy duration), as summarized in Table 1.

Table 1. Clinical information of 14 patients with idiopathic generalized epilepsy
No.AgeAge of seizure onsetSeizure frequency (monthly)SyndromeSeizure freedom achieved with medication
  1. PGE, primary generalized epilepsy not otherwise classified; JME, juvenile myoclonic epilepsy.

  2. The zero monthly seizure frequency indicates that the patient only has a few isolated seizures during the lifetime.

146160PGEYes
233320PGEYes
31990.1PGEYes
420130PGENo
52370PGEYes
61820.1PGEYes
7331260PGENo
848130PGENo
921200PGEYes
1019153PGEYes
1127170.1PGEYes
1236205JMENo
1331150JMEYes
142520JMEYes

We also studied a control group composed of 25 healthy controls (mean age ± SD = 32.5 ± 7.4 years, 14 females) with no previous neurological or psychiatric history. The patient and control groups were similar in age (t37 = 1.21, P = 0.28) and in gender (Yates' Chi = 0.03, P = 0.86) distributions.

Image acquisition

All subjects underwent MRI scans according to the same protocol. Image acquisition was performed on a Verio 3 Tesla MRI scanner (Siemens Medical, Erlangen, Germany). For each subject, diffusion-weighted images (DWIs) and T1-weighted volumetric images were obtained as follows: (i) DWIs: twice-refocused, single-shot echo planar sequence with diffusion weightings; b-value = 0, 1000, and 2000 s/mm2 applied along 30 non-collinear directions. Other imaging parameters were as follows: repetition time (TR) = 8500 ms, echo time (TE) = 98 ms, field of view (FOV) = 222 × 222 mm2, matrix size = 74 × 74, bandwidth = 1324 Hz/pixel, parallel imaging factor of 2, no partial Fourier encoding, number of excitations (NEX) = 10 for b = 0 s/mm2 and 1 for b = 1000, 2000 s/mm2, slice thickness = 3 mm, and 40 axial slices; (ii) T1-weighted images: magnetization-prepared rapid gradient echo (MPRAGE) sequences with parameters: TR = 2250 ms, TE = 4.18 ms, flip angle = 6°, FOV = 256 × 256 mm2, matrix size = 256 × 256, NEX = 1, slice thickness: 1 mm, and 192 sagittal slices.

Image processing

Diffusion MRI processing

Image post-processing for DKI was performed using in-house software package (http://nitrc.org/projects/dke) [25]. The DWIs were first spatially aligned through a six-parameter rigid-body transformation. Diffusion and diffusional kurtosis tensors were then jointly fitted to the DWIs with b = 0, 1000, and 2000 s/mm2 for each voxel, generating parametric maps of mean, radial, and axial diffusivity (MD, D, D), of fractional anisotropy (FA), and of mean, radial and axial kurtosis (MK, K, K) [25].

The FA map for each subject was spatially normalized to the Montreal Neurological Institute (MNI) standard brain space using the FMRIB Software Library (FSL) [26], and the resulting transformation was applied to the other parametric maps. Fig. 1 shows spatially normalized DKI parametric maps: MD, D, D, FA, MK, K, and K, averaged across all subjects.

Figure 1.

Spatially normalized parametric maps derived from diffusional kurtosis imaging: mean, radial, axial diffusivity (MD, D, D), fractional anisotropy (FA), mean, radial, and axial kurtosis (MK, K, K), and white-matter (WM) volume derived from voxel-based morphometry. These images represent the averaged maps (normalized to standard space) from all subjects.

T1-weighted images preprocessing

All T1-weighted images were spatially normalized to the stereotaxic space using affine and non-linear transformation. The images were iteratively segmented into GM, WM, and cerebrospinal fluid based on the International Consortium for Brain Mapping (ICBM) symmetrical brain template employing the VBM8 toolbox (http://dbm.neuro.uni-jena.de/vbm/) for the software SPM8 (http://www.fil.ion.ucl.ac.uk/spm/software/spm8/). The segmented images were modulated accounting for the deformation during non-linear normalization. The WM volume images were used in the voxel-wise analysis.

Voxel-based analysis of tract-based spatial statistics

Tract-based spatial statistics [27] was used for voxel-wise group comparisons of DKI parametric maps to improve sensitivity and objectivity of the comparison [27, 28]. First, the spatially normalized FA maps of all subjects were averaged to create a mean FA map. This mean FA map was used to generate a mean FA skeleton (Fig. 2) at a threshold: FA > 0.2. Second, each subject's FA map was projected onto the mean FA skeleton with the locally maximum FA values to create a skeletonized FA map. The same projection was applied for other non-FA diffusion metrics to create skeletonized maps of MD, D, D, MK, K, and K. Finally, voxel-wise group comparisons of these skeletonized maps were performed using the permutation-based approach [29]. Five thousand random permutations were chosen with a cluster size threshold of t > 3 and a significance level at P-value < 0.05 corrected for multiple comparisons. The permutation-based approach was also used for voxel-wise group comparisons of WM volume images with the same parameters.

Figure 2.

Tract-based spatial statistics (TBSS) method was employed to evaluate the structural integrity of whole brain white-matter pathways. Voxel-based analyses were performed on the voxels contained in the tract-based skeleton (in red). Regions of interest (ROI) analyses were performed on anterior and posterior thalamic radiation tracts (in purple) overlaid on the skeleton.

Region of interest analysis

We performed a focused analysis of thalamo-cortical pathways by placing regions of interest (ROI) on thalamo-cortical WM tracts. Specifically, the ROIs were placed on the anterior and posterior thalamic radiations (as demonstrated in Fig. 2). The ROIs were selected from an anatomical atlas [John Hopkins University (JHU) WM tractography atlas and JHU ICBM-DTI-81 WM labels atlas] [30].

We measured the average of voxel-wise diffusion parameters within the ROI on spatially normalized skeleton DKI maps and WM volume images. We compared these average values between patients with IGE (n = 14) and healthy controls (n = 25) using the independent-sample t-test (two-tailed, unequal variance assumed). We correlated the average values with clinical variables: age of onset of epilepsy, frequency of seizures, duration of epilepsy, and lifetime seizure burden (defined as the frequency of seizures multiplied by epilepsy duration). The correlation was assessed using the Spearman's rank correlation coefficient to account for possible linear and nonlinear correlations and to reduce bias due to outliers in the data. All correlations were adjusted for patient age. The level of significance was set at P = 0.006 after being adjusted for multiple comparisons using the Bonferroni correction.

Results

Compared with healthy controls, IGE group demonstrated microstructural abnormalities on thalamo-cortical projections (Fig. 3). Increased MD and increased D were observed in the body and splenium of corpus callosum. Reduced FA was observed in anterior and posterior limb of internal capsule and splenium of corpus callosum, corresponding to the regions of reduced D and increased D.

Figure 3.

Voxel-wise group comparisons between healthy controls (n = 25) and patients with idiopathic generalized epilepsy (n = 14) using tract-based spatial statistics (TBSS) method: Significant increases (in red) and significant decreases (in blue) overlaid on the skeleton (in green) created in TBSS method; mean, radial, axial diffusivity (MD, D, D), fractional anisotropy (FA), mean, and radial kurtosis (MK, K). No significant differences were observed in axial kurtosis and white-matter volume.

Interestingly, the abnormalities shown by MK and K were more extensive (Fig. 3). Reduced MK and K were distributed over anterior and superior corona radiation, anterior and posterior limb of internal capsule, entire corpus callosum, superior longitudinal fasciculus, and posterior thalamic radiation. Nonetheless, the abnormalities shown by kurtosis metrics did not fully cover the abnormalities showed by FA and D. No significant group differences were observed in K and WM volume.

Regions of interest analyses demonstrated that WM volume was negatively correlated with epilepsy duration (r = −0.73) (Table 2), indicating that a longer history of epilepsy was associated with reduced WM volume.

Table 2. ROI values of DKI metrics and WM volume – mean (standard deviation) – measured from anterior and posterior thalamic radiations in healthy controls and patients with IGE. Group comparisons were performed with the independent-sample t-test (two-tailed, unequal variance assumed)
DKI metricsControl (n = 25)IGE (n = 14)Group comparison (P-value)Correlations with disease duration (P-value)
  1. MD, mean diffusivity; MK, mean kurtosis; ROI, regions of interest; WM, white-matter; DKI, diffusional kurtosis imaging; IGE, idiopathic generalized epilepsy.

  2. Correlations with disease duration were evaluated using the Spearman's rank correlation coefficient with patient age adjusted; other clinical variables showed no significant correlations.

  3. a

    P < 0.006 (Bonferroni-corrected significance level).

MD (μm2/ms)0.90 (0.02)0.91 (0.03)0.420.82
D (μm2/ms)1.41 (0.03)1.40 (0.03)0.310.67
D (μm2/ms)0.65 (0.03)0.67 (0.04)0.170.65
Fractional anisotropy0.46 (0.02)0.45 (0.03)0.160.20
MK0.99 (0.05)0.92 (0.11)0.030.94
K0.75 (0.03)0.74 (0.04)0.740.37
K1.38 (0.10)1.23 (0.19)0.020.75
WM volume0.86 (0.06)0.83 (0.09)0.230.005a

Discussion

In this study, we evaluated the WM microstructural abnormalities in patients with IGE using quantitative analyses of DKI metrics and WM volume derived from VBM. We observed that, compared with healthy controls, patients with IGE showed no volumetric changes, but demonstrated extensive microstructural abnormalities involving thalamo-cortical and cortical–cortical pathways. The anatomical patterns of abnormalities revealed by each DKI metric were different. In particular, MK and K demonstrated more extensive microstructural abnormalities compared with conventional diffusion metrics such as MD and FA. These abnormalities may constitute the structural basis for the thalamo-cortical dysfunction, which is postulated to represent the main pathological mechanism in IGE.

Tissue volume derived from VBM allows quantifications of macroscopic volumetric changes, whereas DKI metrics such as MD, FA, and MK provide quantitative measures of microstructural changes. Increased MD and decreased FA indicate loss of white-matter integrity, possibly as a result of demyelination, increased axonal membrane permeability (increased D), or axonal damage (decreased D) [31, 32]. On the other hand, the reduced MK and K in IGE indicate decreased diffusional heterogeneity, likely related to impoverishment of cell compartmentalization and an increase in membrane permeability [20-22]. Interestingly, our results showed decreased D in anterior and posterior limb of internal capsule and increased D in other regional abnormalities (Fig. 3). Furthermore, more extensive regions showed reduced MK and K. The unique pattern of microstructural abnormalities shown by each DKI metric suggests that IGE is associated with complex microstructural abnormalities. Different information provided by each DKI metric may better characterize subtle tissue alternations in IGE compared with macroscopic volumetric measures.

Our observed regional abnormalities are consistent with previous DTI findings in IGE [12, 13, 15, 16]. Previous DTI studies in IGE reported regional microstructural abnormalities in anterior limb of the internal capsule [12], anterior and superior corona radiation [15], corpus callosum [13, 15, 16], and superior and inferior longitudinal fasciculus [13]. These abnormalities reported by separate studies are mostly encompassed by our observed regional abnormalities using DKI metrics (Fig. 3). This suggests that inconsistencies in previous DTI findings may be due to accurate albeit incomplete characterization of structural abnormalities. Accordingly, DKI results may provide a more comprehensive overview of the pattern and biology of micro-architecture disruption in IGE.

Our ROI analysis on anterior and posterior thalamic radiations only showed correlations between reduced WM volume and disease duration in IGE (Table 2). No correlations were observed between DKI metrics and clinical variables. This indicates that macroscopic volumetric changes may be more directly associated with seizure activity in IGE, while microstructural abnormalities in IGE may represent subtle neurodevelopmental network rearrangements that precede the onset of epilepsy. Indeed, IGE is traditionally considered to be a genetic form of epilepsy. It is possible that microstructural abnormalities occur during brain development, being present prior to the clinical onset of epilepsy. As such, it remains to be demonstrated that these microstructural abnormalities associated with IGE have significant impacts for the disease assessment.

Although our ROI analysis showed that decreased WM volume correlated with disease duration in IGE, no differences in WM volume were observed between IGE and control groups (Table 2). This may be explained by the relatively small sample size of patients with IGE included in this study (n = 14; 11 with primary generalized epilepsy and 3 with juvenile myoclonic epilepsy), which could lead to increased variability and sampling error due to heterogeneous genetic backgrounds [17] and different subsyndromes of IGE [6, 18].

Diffusional kurtosis imaging data acquisition requires at least three b-values up to 2000–2500 s/mm2 to effectively measure non-Gaussian water diffusion in the brain [19, 20]. Two typical choices of b-values include six b-values (0–2500 in increment of 500 s/mm2) and three b-values (0–2000 in increment of 1000 s/mm2). MD and MK measured by these two choices of b-values were shown to be similar in a healthy volunteer [33]. However, in clinical settings, it remains crucial to shorten the scan time to minimize patient discomfort and movement during MRI scans. Therefore, we employed three b-values (0–2000 in increment of 1000 s/mm2) in our study to demonstrate the effect of DKI obtained from clinically feasible protocols.

In conclusion, we performed WM voxel-wise comparison between patients with IGE and healthy controls. We demonstrated that patients with IGE exhibited extensive microstructural abnormalities in thalamo-cortical and cortical–cortical pathways using diffusion metrics derived from DKI. These results suggest that IGE is associated with extensive thalamo-cortical microstructural abnormalities that may be better assessed with DKI.

Acknowledgment

This work was supported in part by the South Carolina Clinical & Translational Research Institute through NIH grant numbers UL1 RR029882 and UL1 TR000062.

Conflict of interest

The authors report no financial or non-financial conflict of interests associated with this study.

Ancillary