Introduction

Over sixty-five million people worldwide and three million people in the United States are diagnosed with epilepsy, 15 to 20 % of which remain medically refractory in spite of antiepileptic medical therapy (Kohrman 2007; England et al. 2012). Mesial temporal lobe epilepsy (mTLE) is the most common form of surgically remediable focal epilepsy, accounting for 60–75 % of patients undergoing surgery for medically refractory epilepsy (Engel 1996). Intracranial electroencephalography (icEEG) optimizes localization of focal epileptogenicity, although it incurs great expense (Kuzniecky et al. 1997; Bulacio et al. 2012), and carries risks of infection, intracranial hemorrhage and elevated intracranial pressure (Arya et al. 2013). This has inspired further work with noninvasive neuroimaging methods to provide better definition of focal epileptogenicity and obviate the need for invasive study in some patients and perhaps altogether (Aghakhani et al. 2014; Zhang et al. 2014).

Magnetoencephalography (MEG) is a noninvasive method of recording the magnetic fields that principally arise from intracellular electric currents flowing in active neurons (Cohen and Hosaka 1976; Hamalainen et al. 1993). MEG is used clinically to localize interictal and, sometimes, ictal activity (Barkley and Baumgartner 2003; Ebersole and Hawes-Ebersole 2007; Knowlton 2008; Sutherling et al. 2008; Englot et al. 2015a, b). Abnormal transients and oscillations can be modeled to identify underlying sources. Synchronization of neuronal activity, a characteristic of epileptogenicity, can be quantified by coherence, a measure of the strength of functional interrelation between pairs of cerebrocortical regions. Use of EEG mean phase coherence has revealed that regions of highly coherent nodes in the cerebral cortex are adjacent to seizure onset zones (Schevon et al. 2007). During the interictal period, local increases in coherence between EEG electrodes have been reported (Towle et al. 1999) as well as increased levels of synchronization in the involved hemisphere (Kraskov 2004). MEG coherence analysis was first used in 2011 to ascertain the laterality of epileptic networks in epilepsy patients where highly coherent activity was found in epileptic neural networks while control subjects lacked similar manifestations (Elisevich et al. 2011).

Diffusion tensor imaging (DTI) is a noninvasive MRI technique which quantifies diffusion properties of water molecules and the degree and direction of anisotropy in biological tissues. Cerebral tissue has highly heterogeneous diffusion properties due to regional differences in nerve fiber density, concentrations of macromolecules and intracellular organelles and myelination density. Tractography can identify virtual pathways of major nerve fiber tracts and quantify abnormalities in these tracts that underlie disruption of the microstructural environment with subsequent reduction of diffusion anisotropy (Rugg-Gunn 2007, Yogarajah and Duncan 2007).

MEG and DTI data have been independently considered for evaluation of epilepsy surgery candidates (Stefan et al. 2007). MEG data has been used in combination with DTI to evaluate subcortical white matter adjacent to cortical tubers (Widjaja et al. 2010). Reduced FA and increased radial diffusion in the subcortical white matter of epileptogenic zones contrasted with adjacent nonepileptogenic zones determined by coregistered MEG data. However, to the best of our knowledge, no study has used MEG data in nonlesional cases to identify white matter fiber tracts intimate with such epileptogenic zones and to evaluate the diffusion properties of these tracts. This paper presents a method by which both MEG and DTI can be used in such a fashion to investigate a putative epileptogenic network. We hypothesize that MEG coherence laterality can be related to the nodal degree laterality extracted by analysis of DTI connectivity, and both indices contribute to noninvasive lateralization of mTLE patients.

Methods and Materials

Subjects

This retrospective study was approved by the Institutional Review Board (IRB) of the Henry Ford Health System. Seventeen consecutive adult patients with refractory TLE (mean age: 33.5 ± 15.4 years; nine females) who had undergone both preoperative MEG evaluation and DTI study with 25 gradient directions, and achieved a post-surgical outcome of Engel class I for at least 1 year follow-up were selected for this study. A retrospective review of each patient’s information was obtained through Henry Ford Health System’s electronic medical record system (CarePlus Henry Ford Health System, Detroit, MI, USA.). Each had undergone inpatient video-electroencephalography, MRI, single photon emission computed tomography (SPECT), neuropsychological evaluation and intracarotid sodium amobarbital injection for evaluation of verbal memory capacity. Those lacking sufficient lateralization by this stage underwent further study by intracranially implanted electrodes for extraoperative electroencephalography. Patients were excluded if their MRI indicated cortical dysplasia, tumor, dilated ventricles or previous resection. Four patients had pathologically-proven hippocampal sclerosis. Table 1 summarizes the patient demographic data. Seventeen age- and gender- matched healthy control subjects without neurologic disorders (mean age: 34.2 ± 15.3 years; nine females) underwent both MEG and DTI study with the same parameters.

Table 1 Patient demographic and clinical data with MEG and DTI features
Full size table

MEG Imaging and Analysis

For each epilepsy patient, 10 min of spontaneous resting state MEG data sampled at 508 Hz with band pass filters set from 0.1 to 100 Hz was acquired while the subject was asked to lie still and minimize movement. Post-acquisition data processing was performed using MEG Tools, an open-source Matlab (The Mathworks Inc., Natick, MA, USA) -based software module for cortical source imaging including single current dipole and multiresolution focal underdetermined system solution (MR-FOCUSS) methods (Moran et al. 2004, 2005, 2006) (http://www.megimaging.com). The data were forward and backward filtered using a 3–50 Hz bandpass filter to remove movement and synchronous breathing and heart artifacts, as well as high frequency electronic noise and 60 Hz powerline. In addition, Independent Component Analysis (ICA) was used to remove cardiac artifact from the MEG data and singular value decomposition (SVD) method was used to eliminate high amplitude artifacts associated with head, eye and mouth movement (Duda and Hart 1973; Tufts et al. 1982). All data were visually inspected for epileptic spikes by board-certified neurophysiologists. During their 10-min recording of resting state MEG, no patient showed ictal events. The American Clinical Magnetoencephalography Society (ACMEGS) clinical practice guideline 1 indicates that there must be more than five epileptic spikes seen before the findings are classified as interictal (Bagic et al. 2011). Hence, five patients were declared to have no interictal events (i.e., less than 5 spikes), while 12 patients had occasional epileptic spikes or sharp waves (over 100 for four cases) scattered across the cortex. We did not see any slow waves as they were filtered out by our bandpass filter.

MEG coherence source imaging (MEG-CSI), capable of source space analysis, was calculated using the 10 min of resting state MEG data. Coherence cross-spectra between all active cortical sites was performed separately for each 7.5 s of data of relatively uniform brain behavior (Moran et al. 2004, 2005). For these calculations, 20 fast Fourier transform (FFT) spectra with 2 Hz bin resolution between 3 and 50 Hz, were generated for 0.5 s long data segments that were Hanning windowed and overlapped by 25 %. For each of these data segments, signals from neuronal sources were isolated using an ICA spatiotemporal decomposition technique designed to extract signals from distinct compact sources that exhibit burst behavior and minimal temporal overlap with other active sources. These ICA signal components have MEG spatial magnetic field patterns corresponding to one or a few spatially distinct compact sources which can be imaged accurately in source space using MR-FOCUSS (Moran et al. 2005). Finally, for the entire study, coherence spectra, both real and imaginary components, were average across time and frequency to obtain a final measure of connectivity.

Gray Matter Model

To localize cortical source activation of epileptogenic activity, a model of gray matter was constructed for each individual’s T1-weighted high-resolution volumetric MR image (see “Appendix”). We used a probabilistic brain atlas composed of 56 structures from manually delineated MRI data constructed by Shattuck et al. (2008) as a standard volumetric head model with each location specified in MNI305 coordinates. This atlas contains all cerebral lobes and, specifically, the right and left hippocampi, limbic gyri, insular cortices, caudate, putamen, cerebellum and brainstem. Excluding the cerebellum and brainstem reduced the number of anatomical regions to 27 in each cerebral hemisphere (Table 2; Fig. 1) (Shattuck et al. 2008). The realistic head model consisted of X-, Y- and Z- oriented dipoles at approximately 4000 locations such that every location represented the same amount of gray matter identified in the individual’s MR image. These MR images were coregistered with the individual’s digitized head shape recorded at the time of MEG data collection.

Table 2 Anatomical sites in the left and right hemisphere where the MEG coherence and nodal degree laterality measures were calculated
Full size table
Fig. 1
figure 1

Left Cortical modeling of the brain in the MNI coordinates; the average cortical model and its surface in red. Right The delineated anatomical structures (Shattuck et al. 2008). Regions of this slice are color-coded according to their anatomical identification using MEG Tools. Note that the average cortical model is blurred and its surface is smooth as it is created from a large number of volumetric MRIs of normal individuals (40 cases) (Color figure online)

Full size image

Calculating MEG Coherence Laterality

The imaging results and the signal cross-spectrum were used to calculate the coherence between all pairings of each of the 54 cortical locations within each of the 24 frequency bins. Finally, for each active source, the average coherence with all other sources was calculated for each frequency and then averaged across the bandwidth of 3 to 50 Hz, ranging from 0 (no coherence) to 1 (highly coherent) (Moran et al. 2006; Elisevich et al. 2011). In these coherence imaging results, the localization of imaged brain activity is strongly dependent on the frequency bands with greatest power. When these coherence results are averaged across the full 10 min of data, only cortical sources that are consistently engaged in synchronous activity contribute to the final results. The MEG-CSI results were coregistered to individual volumetric MRI scans and areas of significant coherence were identified for each subject (“Diffusion Tensor Imaging and Analysis” section).

The coherence laterality Coh_Lat was computed to determine which hemisphere exhibited higher coherence over the entire time interval of spontaneous acquisition. It was calculated for each cortical site as:

$$Coh\_Lat_{ } \left( i \right) = Coh\left( {i + 27} \right) - Coh\left( i \right) , i = 1:27$$
(1)

where \(Coh\left( i \right)\) and Coh(i + 27) represent the coherence for the site i in the left and right hemispheres, respectively. A positive value indicates that a greater fraction of the right hemisphere was engaged in coherent activity of a cortical site compared to the left hemisphere on average for the patient across the 10 min of MEG data.

Diffusion Tensor Imaging and Analysis

All subjects underwent preoperative imaging in a 3.0T MRI system (GE Medical Systems, Milwaukee, USA.) using a standardized protocol for image acquisition. Coronal T1-weighted images were acquired using the spoiled gradient echo (SPGR) protocol with TR/TI/TE = 10400/4500/300 ms, flip angle = 15°, near cubic voxel size = 0.9375 × 0.9375 × 1.00 mm3, \({\text{imaging matrix }}256 \times 256 ,\) field-of-view (FOV) of 240 × 240 mm2 that includes the entire skin surface of the head for construction of head and cortical model for MEG analysis. DTI images (b-value of 1000 \(\frac{\text{s}}{{{\text{mm}}^{2} }}\)) along with a set of null images (b-value of 0 \(\frac{\text{s}}{{{\text{mm}}^{2} }}\)) were acquired using echo planar imaging (EPI) (Stieltjes et al. 2001; Mori and van Zijl 2002) with TR/TI/TE = 7500/0/76 ms, flip angle = 90°, voxel size = 1.96 × 1.96 × 2.6 mm3, \({\text{imaging matrix }}128 \times 128\), FOV of 240 × 240 mm2 and 25 diffusion gradient directions.

DTI Preprocessing

Before tractography and connectivity analysis, the DTI data were prepared by interpolation to a cubic voxel size of 1.96 mm and tensor, FA and MD calculation (Pierpaoli et al. 2001, Nazem-Zadeh et al. 2012). For the purpose of tractography, the principal diffusion direction (PDD), the eigenvector corresponding to the largest eigenvalue of the tensor, was also calculated from the tensor. Using an affine transformation (FSL, FLIRT), T1-weighted images and subsequently the gray matter anatomical regions extracted by MEG Tools were coregistered to the DTI data to establish the input ROI for tractography (Jenkinson et al. 2002).

DTI Tractography and Connectivity

Tractography offers a method of identifying diffusion parameters associated with white matter tracts that may facilitate the propagation of epileptic activity. The tractography was performed automatically using FACT Streamline (Mori and van Zijl 2002) implemented in a home-made tractography application between all 54 anatomical regions. Streamline fiber tracking parameters comprised an FA threshold = 0.10, minimum fiber length = 0.10 mm, and maximum allowed angle bending between two fiber segments = 45°. The FA threshold eliminated the inclusion of gray matter to allow comparison of diffusion properties of only the resultant white matter tracts. The minimum fiber length threshold eliminated the irrelevant minor fibers that are essentially constructed from DTI noise. The maximum allowed angle threshold guaranteed a smooth fiber trajectory expected in practice. With the reconstructed fibers, a connectivity matrix was constructed for each subject by calculating the mean FA value of the voxels of all fibers connecting each pair of regions:

$$ConnectivityMat\left( {i,j} \right) = \mathop {\text{mean}}\limits_{k} FA(Fiber\left( {i,j,k)} \right)$$
(2)

where k represents the kth fiber connecting the brain sites i, j.

Nodal Degree Laterality

The nodal degree (ND) of a brain site i is defined as (DeSalvo et al. 2014):

$$ND\left( i \right) = \mathop {\text{mean}}\limits_{j\varepsilon S} \left[ {ConnectivityMat\left( {i,j} \right) - \mathop {\text{mean}}\limits_{k\varepsilon C} ConnectivityMat\left( {i,j,k} \right)} \right]$$
(3)

where C and S denote the cohorts of controls and all subjects, respectively. The nodal degree laterality ND_Lat was computed to determine which hemisphere exhibited the higher nodal degree:

$$ND\_Lat_{ } \left( i \right) = ND_{ } \left( {i + 27} \right) - ND_{ } \left( i \right) , i = 1:27$$
(4)

where \(ND\left( i \right)\) and ND(i + 27) represent the nodal degree for the site i in the left and right hemispheres, respectively.

Statistical Analysis

Two-way repeated measures analysis of variance (RMANOVA) was used to examine the relationships of the MEG coherence laterality measurements and the DTI nodal degree laterality measurements with the brain regions (i.e., a repeated factor) and the mTLE laterality type (i.e., a fixed factor) (Nazem-Zadeh et al. 2015). Of particular interest were tests for interaction between a region and a laterality type, since a significant interaction would imply that separate one-way ANOVAs are required to assess mTLE laterality type.

For each region, one-way ANOVA on the mTLE laterality type was performed and multiple comparisons were addressed by Bonferroni adjustments for three pairwise comparisons between laterality types (p < 0.05/3 = 0.0167). However, the one-way ANOVAs were considered statistically significant only if the overall ANOVA F-test for all mTLE laterality types was also significant after a multiple comparisons adjustment (p < 0.05/27 = 0.0019).

One-way RMANOVA was performed for 27 anatomical regions followed by paired t-tests between the corresponding ipsilateral and contralateral subregions. It was considered statistically significant only if the overall RMANOVA F-test was also significant for the corresponding sites after Bonferroni adjustments for 27 pairwise comparisons between ipsilateral and contralateral sides (p < 0.05/27 = 0.0019).

Individual Analysis of Laterality

Although the anatomical sites with the highest laterality may be ranked, as in the case of coherence, using MEG Tools (Moran et al. 2005), the threshold for establishing the number of sites with the highest laterality may vary from one subject to another. Alternatively, the observed laterality variation for MEG coherence and DTI nodal degree in anatomical sites in individual patients may be compared to the laterality variation uncertainty (LVU) level estimated from controls. If it falls beyond a certain level, the site can be considered a high coherence site. The LVU is estimated by laterality analysis of a control cohort who have undergone imaging with the same scanner, under the same imaging conditions. Any laterality variation in control cohort should be attributable to natural physiological occurrences, thus, no significant laterality is expected to be observed in control subjects (Nazem-Zadeh et al. 2014c).

Lateralization Response-Driven Models

For both coherence and nodal degree, statistically significant laterality measures between the left and right mTLE cohorts were considered as multivariate independent variables and incorporated into the development of response-driven models of laterality using logistic function regression (Hosmer et al. 2013; Nazem-Zadeh et al. 2014a, b).

Correlation Between MEG Coherence and DTI Nodal Degree Laterality

Among the anatomical sites where significant differences in laterality of both coherence and nodal degree between the right and left mTLE patients were seen, the correlation between the laterality measures were evaluated using the coefficient of determination denoted by R 2 extracted by linear regression. A statistically significant high value of R 2 implies a high level of correlation between the MEG and DTI laterality measures.

Results

The ages of the male and female subjects across any of the right and left mTLE and control cohorts were statistically comparable. Table 1 shows the significant anatomical sites with the laterality variation of MEG coherence and DTI nodal degree beyond the LVU level for individual patients.

MEG Coherence Laterality and mTLE Lateralization

MEG-CSI identified epileptic network sites for each of the 17 mTLE patients (Fig. 2).

Fig. 2
figure 2

MEG -CSI map for patient #16 in Table 1 with right TLE. High coherence was detected in the right mesial temporal region (Color figure online)

Full size image

Two-way RMANOVA demonstrated significant interaction between regional and mTLE laterality type with coherence laterality measures (p < 0.001). Statistical analysis in mTLE patients using RMANOVA followed by t-tests between pairs of ipsilateral side (i.e., the resected side) versus contralateral side showed significant interhemispheric variation in MEG mean coherence (p < 0.0019), particularly the insular cortex (ipsi: contra, 0.027 ± 0.004:0.012 ± 0.001), the superior temporal (ipsi:contra, 0.111 ± 0.009: 0.064 ± 0.006), lateral orbitofrontal (ipsi:contra, 0.102 ± 0.011: 0.054 ± 0.008) and middle temporal (ipsi:contra, 0.120 ± 0.012: 0.084 ± 0.009) gyri and putamen (ipsi:contra, 0.077 ± 0.003:0.007 ± 0.001) (p < 0.05; Fig. 3). For controls, no single anatomical site showed a significant difference in coherence between the right and left sides, although the overall RMANOVA F-test was significant across all sites (p < 0.0019). This finding is in concordance with the evidence that control subjects did not exhibit areas of high coherence during spontaneous MEG study (Elisevich et al., 2011).

Fig. 3
figure 3

MEG coherence at anatomical sites with significant differences between the ipsilateral and contralateral structures of the mTLE patients and the average coherence on the left and right sides of the controls. The error bars represent the standard error of the mean coherence

Full size image

For the insular cortex and the lateral orbitofrontal and superior temporal gyri, the overall ANOVA F-test in all mTLE laterality types was significant after Bonferroni adjustments (p < 0.0019). In t-tests between pairs of laterality types, the coherence laterality showed significant differences in insular cortex for the right versus left mTLE cases, right mTLE versus control cases and left mTLE versus control cases (p < 0.0167; Fig. 4). However, for the lateral orbitofrontal and superior temporal gyri, significant differences were found in coherence laterality only for the right versus left mTLE cases and the right mTLE versus control cases (p < 0.0167). Figure 5 shows the MEG mean coherence in the insular cortex and the lateral orbitofrontal and superior temporal gyri overlaid upon the MNI registered brain (Xia et al. 2013) for the right and left mTLE patient cohorts. Figure 6 shows the MEG coherence laterality in these regions for individual patients.

Fig. 4
figure 4

MEG coherence laterality in the brain sites with significant differences between the right and left mTLE patients. The error bars represent the standard error of the mean coherence laterality

Full size image
Fig. 5
figure 5

The MEG mean coherence in insular cortex and the lateral orbitofrontal and superior temporal gyri overlaid upon the MNI registered brain (Xia et al. 2013) in the right and left mTLE patients where significant differences in coherence laterality between the right and left mTLE patients are demonstrated. The spheres and lines show the significant cortical sites and their corresponding connections, respectively. The right and left cortical sites are shown in yellow and red, respectively. The mean coherence values are represented by the size of the spheres (Color figure online)

Full size image
Fig. 6
figure 6

MEG coherence laterality in individual patients for insular cortex (a) and the lateral orbitofrontal (b) and superior temporal (c) gyri where significant differences exist between the right and left mTLE patients. Patient numbers correspond to the numerical allocation in Table 1

Full size image

Figure 7 shows the laterality models by logistic regression of the MEG coherence laterality data in these anatomical sites with significant differences identified between the left and right mTLE patients. The laterality results of these models agreed with the side of epileptogenicity for the coherence laterality in the insular cortex and the lateral orbitofrontal and superior temporal gyri for 82, 88 and 88 % of patients, respectively. Combining the laterality measures in these three anatomical sites improved lateralization results to 94 % of the patients (Fig. 7d).

Fig. 7
figure 7

The laterality model of the right mTLE (i.e., probability of being right mTLE) by logistic regression of the MEG coherence laterality data in the insular cortex (a) and the lateral orbitofrontal (b) and the superior temporal (c) gyri, and the integrated model (d). The result of laterality achieved by these models agreed with the side of epileptogenicity for 82, 88, 88 and 94 % of patients, respectively

Full size image

DTI Nodal Degree Laterality and mTLE Lateralization

Two-way RMANOVA demonstrated significant interaction in DTI nodal degree laterality between both regional and mTLE laterality types (p < 0.001). In mTLE patients, statistical analysis using RMANOVA followed by t-tests between case pairs of ipsilateral side (i.e., resected side) versus contralateral side showed no significant interhemispheric variation in the DTI nodal degree. The same finding was evident in the side-to-side comparison in the control cases. However, the nodal degree laterality showed significant differences in the gyrus rectus, insular cortex, precuneus and the superior temporal gyrus in both overall ANOVA F-test for mTLE laterality types and t-tests between pairs of right and left mTLE, after Bonferroni adjustments (p < 0.0019 and p < 0.0167, resp; Fig. 8). Figure 9 shows the mean nodal degree in these same anatomical sites in both right and left mTLE patient cohorts and Fig. 10 shows the nodal degree laterality in these sites for each patient.

Fig. 8
figure 8

DTI nodal degree laterality at anatomical sites with significant differences between right and left mTLE patients. The error bars represent the standard error of the mean nodal degree laterality

Full size image
Fig. 9
figure 9

The DTI nodal degree in gyrus rectus, insular cortex, precuneus and superior temporal gyrus overlaid upon the MNI registered brain (Xia et al. 2013), in the right and left mTLE patients, where significant differences in the nodal degree laterality were identified between the right and left mTLE patients. The spheres and lines show the significant cortical sites and their corresponding connections, respectively. The right and left cortical sites are shown in yellow and red, respectively. The nodal degree values are represented by the size of the spheres (Color figure online)

Full size image
Fig. 10
figure 10

DTI nodal degree laterality in gyrus rectus (a), insular cortex (b), precuneus (c) and superior temporal gyrus (d), where significant differences were identified between the right and left mTLE patients. Patient numbers correspond to the numerical allocation in Table 1

Full size image

Figure 11 shows the averaged density map of tracts originating from the gyrus rectus, insular cortex, precuneus and superior temporal gyrus, for the cohorts of left and right mTLE patients. As can be seen, fewer tracts were reconstructed originating from the ipsilateral insular cortex (posteriorly) and superior temporal gyrus, compared to the side contralateral to the epileptogenicity. A substantial interhemispheric variation of the density map was also observed for the gyrus rectus only in the left mTLE cohort.

Fig. 11
figure 11figure 11

Averaged density map of tracts overlaid upon the MNI registered brain (Xia et al. 2013), between the gyrus rectus (a), insular cortex (b), precuneus (c) and the superior temporal gyrus (d) as the input ROI and all other anatomical ROIs. The density maps are averaged on the left mTLE patients (left images) as well as right mTLE patients (right images). Note that in (b), the tracts originating from the insular cortex are deep intracranial tracts and cannot be rendered on the surface of the MNI registered brain (Color figure online)

Full size image

Laterality models determined by logistic regression on the DTI nodal degree laterality data agreed with the side of epileptogenicity as it pertained to the gyrus rectus, insular cortex, precuneus and superior temporal gyrus for 88, 71, 82 and 94 % of patients, respectively (Fig. 12). Combining the laterality measures in these four anatomical sites improved the results further with correct lateralization of 100 % for all patients (Fig. 12e).

Fig. 12
figure 12

The laterality model (i.e., probability of right mTLE) by logistic regression on the DTI nodal degree laterality data in the gyrus rectus (a), insular cortex (b), precuneus (c) and superior temporal gyrus (d) and the integrated model (e)

Full size image

Correlation Between MEG Coherence and DTI Nodal Degree Laterality

A high correlation was observed between the MEG coherence and the DTI nodal degree laterality in the insular cortex, where significant differences were observed between the right and left mTLE patients with R 2 = 0.46 (p = 0.003; Fig. 13).

Fig. 13
figure 13

A high level of correlation between MEG coherence and DTI nodal degree laterality measures in the insular cortex with R2 = 0.46 (p = 0.003)

Full size image

Discussion

This study described a method of combining MEG and DTI to investigate white matter fibers associated with epileptiform activity in the mTLE patients. The MEG coherence laterality in the insular cortex was correlated with the DTI nodal degree laterality obtained from analysis of DTI connectivity. Both insular cortex and superior temporal gyrus showed significant differences in both MEG coherence and DTI nodal degree laterality measures for the right and left mTLE. These same regions in both modalities could be used to model laterality. The MEG coherence laterality applied to the lateral orbitofrontal area and the DTI nodal degree laterality to both the gyrus rectus and precuneus also showed significant differences between the right and left mTLE patients with subsequent successful laterality modeling. These findings imply that there are both common and distinctive hubs for functional activity and diffusivity detectable by the MEG coherence and the DTI connectivity measures. It also shows that the hemisphere containing the epileptic focus in the mTLE patients may be determined using a dual MEG-DTI laterality model involving these brain areas.

The MEG coherence laterality measure of an anatomical region was calculated by averaging the MEG activities between that and all other anatomical regions. Similarly, the nodal degree laterality measure was calculated by averaging the fractional anisotropy of all tracts originating from that and connecting all other regions. An anatomical node such as hippocampus may have some laterality-sensitive connections when considered individually, as well as some laterality-insensitive connections. Therefore, it is not surprising that the hippocampus stays insensitive to the laterality by lumping all the propertied of its functional or structural connections. Nevertheless, the other anatomical regions were found sensitive enough for the purpose of lateralization with sufficient reliability when the functional and structural properties are averaged between the connections. Laterality, in the end, is determined more by the local regional network activity rather than a single anatomical site.

In mTLE, epileptic network behavior is associated with alteration of both of the gray and white matter (Bernasconi et al. 2004). Loss of anisotropy has been demonstrated in the arcuate fasciculus (Powell et al. 2007; Govindan et al. 2008), uncinate fasciculus (Rodrigo et al. 2007), external capsule (Gross et al. 2006), inferior longitudinal fasciculus (Govindan et al. 2008), corpus callosum (Chahboune et al. 2009, Nazem-Zadeh et al. 2015), the afferent and efferent tracts of the parahippocampal gyrus (Yogarajah et al. 2008), fornix and cingulum (Concha et al. 2005, 2009; Nazem-Zadeh et al. 2014c, 2015), thalamic fibers (Bonilha et al. 2012) and widely spread tracts in the temporal lobe ipsilateral to epileptic focus (Concha et al. 2005, Focke et al. 2008). The apparent diffusion coefficient (ADC) is calculated from the trace of the diagonalized diffusion tensor to give the mean bulk mobility of water without directional information. Increases of ADC in epileptogenic zones have been identified in mTLE suggesting an increased extracellular space possibly attributable to changes in cellular volume or geometry congruent with interictal spiking (Thivard et al. 2006). Taken together, these studies suggest that abnormal water diffusion, identified as changes in FA or ADC, may be found in identified white matter tracts of epileptic networks in both temporal and extratemporal structures.

Investigation of structural connectivity in mTLE using DTI and functional activity from fMRI (Sporns 2011) has shown similar promise in lateralizing epileptogenicity (Lemkaddem et al. 2014). In this context, the brain is modeled as a network(s) of connected nodes. The nodes are selected based upon structural and functional parcellation of the brain. The connectivity matrix is then established either from DTI or fMRI connectivity analysis. The entire connectivity matrix or specific connections may be compared between groups of individuals. Major differences may be found between groups, although at the expense of higher sampling volume and more complex statistical analysis (Zalesky et al. 2010). An alteration in structural connectivity of the temporal pole and the inferolateral and perisylvian cortices has been identified in unilateral mTLE (Besson et al., 2014). A decrease in structural connectivity with DTI, both within-module and between-module, throughout the default mode network (DMN) has also been observed in mTLE compared to nonepileptic subjects (Vaessen et al. 2011; Chiang and Haneef 2014; DeSalvo et al. 2014). This appears to underlie a loss of functional connectivity determined by fMRI (Skudlarski et al. 2008; Pittau et al. 2012). A widespread increase in global network efficiency is seen within the DMN in mTLE (Vaessen et al. 2011; Chiang and Haneef 2014; DeSalvo et al. 2014), implying a facilitation of propagation of epileptogenicity throughout the region. Some reorganization of the limbic system in mTLE has also been identified (Bonilha et al. 2012). Functional MRI connectivity analysis has shown increased DMN connectivity with other brain regions in left TLE (Centeno and Carmichael 2014), but decreased connectivity in right TLE (Haneef et al. 2014a). Several investigations using seed-based analysis of specific cerebral connections in TLE have focused upon the thalamus (Barron et al. 2014; Keller et al. 2014; Barron et al. 2015; He et al. 2015) and hippocampus (Haneef et al. 2014b; Dinkelacker et al. 2015). While most studies address either DTI or fMRI connectivity, a few have investigated both connectivity measures (Dinkelacker et al. 2015). Moreover, some have used connectivity analysis for lateralization of epileptogenicity (Morgan et al. 2012, Barron et al. 2015). A comprehensive review of fMRI connectivity studies in epilepsy can be found elsewhere (Centeno and Carmichael 2014).

EEG coherence in sensor space (i.e., at the electrode site) has been widely used in studying epileptiform activity to determine ictal onset zones. Brazier (1972) was the first to use coherence to detect the influence of one brain region over another during an ictus (Brazier 1972). Later Gotman (1981) made the method more reliable by including more frequencies and validating the use of this method to detect inter hemispheric interactions (Gotman 1981). A recent study (Song et al. 2013) has shown that EEG coherence can be used to characterize a pattern centered upon temporal lobe structures. A problem with this traditional way of mapping coherence in sensor space concerns the spread of the electric current over the surface of the head. Electromagnetic field spread creates a problem when trying to interpret coherence at the sensor level. Since recordings from all EEG electrodes or MEG coils measure the sum of the activity from several brain regions, it is difficult to resolve coherence to a specific location (Schoffelen and Gross 2009). In the past 10 years, developments in the computational analysis of source localization for MEG have advanced the ability for connectivity to be imaged directly within specific regions (i.e., source space), providing a better anatomical localization as well as greater ease for co- registering to the DTI data. The high temporal resolution of MEG allows for investigations of function and effective connectivity with millisecond precision. It is possible now to study the mechanisms by which information is exchanged across brain regions, including oscillatory and synchronized neuronal activity. Only a handful of MEG coherence studies using different inverse methods (dipoles, minimum norm or beamformers) to localize source space coherence have been performed in the past decade. Elisevich et al. (2011) showed that MEG-CSI can provide targets for successful surgical resection in patients with epilepsy with a detection rate of 77 % of the patients who had high coherence in the area of the later resection (Elisevich et al. 2011). MEG-CSI uses a current distribution technique, MR-FOCUSS, to image underlying sources based on a 10 min resting state scan. More recently, Englot et al. (2015a, b) used a beamforming technique to image underlying sources and studied regional and global functional connectivity of MEG coherence in patients with epilepsy, based on a 1-min resting state scan (Englot 2015a, b). An Engel class I outcome was seen in 87.5 % of patients where increased connectivity was found in the region of the later resection. The current paper has shown how coupling MEG coherence to DTI nodal imaging can increase this predictability.

Ictal activity in epilepsy patients who are sent for a MEG scan can influence coherence between brain structures across the entire frequency band with significant individual variation in the degree of disturbance of the normal brain network activity. However, in our application, averaging the coherence over the full pass band over the entire 10 min decreases the variance of the measured coherence across patients and the underlying focal network ongoing in the background of the spike activity can be detected.

Moreover, we have seen that ictal activity often has a peak coherence in the 20–40 Hz range. Therefore splitting up this range into smaller bands (beta: 15–30 Hz and gamma: 30–50 Hz) may divide the region of interest inadvertently. There might still be some activity in the network of interest outside but near this frequency range (20–40 Hz) that should be preserved.

Coherence was used to quantify the contribution of individual brain structures to network interactions. MEG-CSI utilizes coherence from both the real and imaginary sides, with the magnitude of a site’s functional connectivity within the brain network calculated as an average of the (full) coherence of the site with all the other active cortical sources. This avoids the need to account for the effect of image blur on the individual site-to-site coherence calculations between active brain sites, and provides a better measure of direct and indirect connectivity, with better statistical performance than imaginary coherence alone. This method has been validated with EEG coherence in a small cohort of patients that had a MEG then were implanted with icEEG (Moran et al. 2006).

Comparison of MEG results across subjects becomes possible when images of individual subjects are coregistered with a common brain coordinate system such as with Talairach or MNI coordinates. Likewise, multimodal comparisons of MEG with other imaging studies becomes possible when coordinates of individual imaging data are coregistered with a common brain coordinate system. Establishing correspondence of subject MRI voxel locations to MNI locations requires the mapping of the subject’s brain image to the corresponding MNI brain structure using AC-PC coordinates. A variety of linear and nonlinear transform techniques have been developed to accomplish this task. These techniques attempt to match either the volume or surface structure with the corresponding structure in the MNI brain atlas. The three popular registration techniques used are AIR (Woods et al. 1998), FLIRT (Smith et al. 2004) and SPM (Ashburner and Friston 2005). The MRI/MNI coregistration techniques are not part of a unified MEG/MRI imaging application and, therefore, are not readily applicable for MEG. Moreover, MEG involves the allocation of cortical activity, often in diseased or surgically altered brain, onto a standard MRI introducing potential error in localization. The nonlinear MRI-to-MNI transforms developed here are part of a complete MRI import and processing utility, available in the MEG imaging software, MEG TOOLS, (www.megimaging.com ), designed to accommodate a wide range of MRI distortions, brain malformations and surgical resections. This transform included both MEG to MRI and MRI to MNI coregistration as well as cortical volume extraction by a partially automated algorithm. MRI to MNI coregistration utilizes a combination of linear and second-order polynomial transforms to produce volumetric warping of brain structure such that the patient brain surface matches the brain surface template of the MNI305 average brain. The anatomical identity of all cortical locations is determined by consulting an anatomical atlas corresponding to the MNI-305 brain.

Conclusions

An epileptic network is comprised of multiple neuronal sites behaving synchronously connected by white matter tracts that propagate this activity. In this paper we investigated the relationship between cerebral functional abnormalities quantified by MEG coherence and structural abnormalities quantified by DTI in mesial temporal lobe epilepsy (mTLE). Two recent studies have shown that MEG-CSI provides clinicians with valuable information regarding surgical candidacy. We have combined a MEG-CSI method with the DTI nodal degree (i.e., number of links connected to a node) to further analyze the detection of the side of epileptogenicity. With increasingly sophisticated signal processing methods and the use of multimodal neuroimaging and neurophysiological biomarkers, noninvasive investigational techniques may ultimately supplant invasive monitoring as a means of localizing focal epileptogenicity and establishing surgical candidacy.