7.1 Introduction

While the study of the gray matter has been the main focus of epilepsy research, observations of white matter (WM) atrophy and gliosis date back to early seminal histopathological assessment of surgically resected tissue in patients with temporal lobe epilepsy.¹ WM may subserve seizure propagation, forming an integral part of the epileptogenic network. Recent advances in MRI techniques and quantitative postprocessing algorithms have provided unprecedented tools to examine the microstructure and architecture of this compartment. While early studies based on manual segmentation have provided volume measurements of this compartment, more recent computational approaches integrating various imaging contrasts, such as T2 relaxometry, diffusion tractography, and myelin imaging, lend biologically meaningful indices of its microstructural properties. In the past two decades, a large body of work has indeed revealed novel information regarding the distribution and nature of WM pathology in epilepsy. Together with gray matter findings, these results consolidate the view that focal epilepsy is a system-level disorder.

7.2 Voxel-Based Techniques

Early MRI studies of patients with temporal lobe epilepsy (TLE) noted WM atrophy alongside hippocampal anomalies during visual inspection.² Such observations motivated quantitative analysis using manual volumetry. Similar to hippocampal volumetry, manual tracing requires expert knowledge of anatomy and provides a single summary statistic of the queried structure. In particular, its application to the fimbria-fornix, the major projection of the hippocampus in TLE, revealed significant atrophy concordant with the side of hippocampal atrophy.³ While the time-consuming nature of the procedure has confined its application to small WM structures, automated methods enabled efficient quantitation of lobar and global volumes. Indeed, several studies have identified marked reduction of the WM lobar volume in the temporal, frontal, and parietal lobes mainly ipsilateral to the side of seizure focus.⁴ The degree of WM loss was associated with disease duration⁴ and the severity of cognitive impairment, including (but not limited to) processing speed and full-scale and verbal IQ.⁵ Voxel-based morphometry studies in TLE have shown WM reduction mostly colocalized with GM abnormalities.⁶ Particularly, our group found WM loss in the temporopolar, entorhinal, and perirhinal areas ipsilateral to the seizure focus, suggesting a disconnection process involving preferentially frontolimbic pathways.⁷

WM pathology has also been studied using T2 relaxometry, a quantitative analysis of T2-weighted signal. Region-of-interest analysis and voxel-based statistical comparisons reported group-level increases in the anterior temporal lobe WM in TLE patients.⁸ We showed that while hippocampal T2 relaxometry provides a correct lateralization of the seizure,⁹ prolongation of WM T2-relaxation time in the temporal stem occurs in a rather bilateral and symmetric fashion.¹⁰ Altogether, these early studies set the basis for large-scale extrahippocampal WM disruption in TLE.

7.3 Diffusion-Weighted MRI

Despite relatively coarse spatial resolution (typically 2 mm), diffusion-weighted MRI and its analytical extensions, particularly diffusion tensor imaging (DTI), exploit the diffusion of water molecules to probe tissue microstructure, which is especially useful in coherently organized tissue, such as the WM.¹¹ While its sensitivity and specificity for the identification of the primary epileptogenic lesion are modest, this technique has revealed WM anomalies in several forms of epilepsy, particularly TLE.¹²

The most often reported diffusion MRI metric is related to diffusion anisotropy (e.g., fractional anisotropy, FA), which tends to be high in regions where densely packed axons have very similar orientations. Reduced diffusion anisotropy at the expense of increased radial diffusivity is interpreted as reduced axonal density and alterations of myelin architecture. Histological examination of the fornix and the temporal pole WM structures derived from TLE surgical specimens have confirmed these anomalies.^{13, 14} Small caliber axons, in particular, are reduced in number in TLE patients.¹⁵ Although perhaps to a lesser degree, other histological features, such as WM gliosis, may also play a role in altered water diffusion metrics. Notably, orientational coherence of axons is reduced by the infiltration of oligodendrocytes, which are more common in regions of WM hyperintensity, and may reflect an attempt to compensate for progressive myelin loss.¹⁶ Independently of gliosis, the number of heterotopic neurons, commonly seen in the superficial WM, is higher in TLE patients than in controls;¹⁷ however, their density and overall dimension appear to be insufficient to modulate water diffusion profiles.

7.3.2 Surface-Based Analysis of the Superficial White Matter

Most DTI studies have focused on assessing the integrity of deep WM bundles. The superficial WM immediately subjacent to the cortical mantle has been largely neglected. Given its anatomical proximity to neocortical networks and role in maintaining long-range connectivity, the analysis of this compartment may provide novel insights in the pathophysiology of various epilepsy syndromes. The complex fiber trajectories of the less densely myelinated superficial WM challenge conventional diffusion tractography and TBSS.³⁹ Conversely, surface-based approaches allow vertex-wise sampling of diffusion parameters at multiple WM depths. In TLE, diffusion alterations (increased diffusivity and decreased anisotropy) of the superficial WM are found primarily in ipsilateral limbic regions⁴⁰ (Figure 7.1). In extratemporal epilepsy related to FCD Type II, subtle diffusion anomalies are found both at the lesional site and in the normal appearing perilesional WM⁴¹ (Figure 7.2).

Figure 7.1. Surface-based mapping of superficial white matter (SWM) diffusion in TLE. (A) After generating the inner (WM-GM, green) and outer (GM-CSF, red) cortical surfaces, we computed a Laplacian potential field between the WM-GM interface and the ventricular walls to guide placement of a surface running 2 mm below the WM-GM boundary (SWM, yellow). Fractional anisotropy (FA) and mean diffusivity (MD) were sampled on this surface. (B) SWM diffusion anomalies in right (RTLE) patients relative to controls showing predominantly ipsilateral temporal and limbic anomalies. Similar anomalies are found in left TLE (not shown). Findings were corrected for multiple comparisons thresholded at P_FWE < .05 using random field theory for nonisotropic images (cluster threshold p < .01).

Figure 7.2. Multisurface profiling of diffusion MRI (dMRI) parameters in focal cortical dysplasia. (A) For lesion profiling, individual patients (normalized with respect to corresponding regions in healthy controls) are plotted as a function of intracortical and subcortical level, separately for Type IIA (red dot) and Type IIB (black dot); zero reference line indicates the mean of controls; mean values and standard deviations in patients are shown as horizontal lines, color coded by patient group. An example case is with arrowheads pointing to the lesion on the T1-weighted MRI. Overall, Type IIB lesions display increased subcortical mean diffusivity (MD), while IIA are associated with decreased fractional anisotropy (FA) at the gray matter (GM)-white matter (WM) interface and subcortical WM. (B) For distance-based profiling, normalized features are averaged across cortical and subcortical surfaces, and plotted relative to the geodesic distance from the primary lesion in steps of 4 mm. Subcortical diffusion alterations extend up to 16 mm, with anomalies being more marked in Type IIB. Conversely, reduced cortical FA, specific to Type IIA, extended up to 6 mm outside the lesion. Asterisks indicate significant differences with controls and between cohort contrasts after correction of multiple comparisons (FDR < .05); small dots indicate uncorrected findings at p < .05.

7.4 Analytical Approaches to the White Matter beyond Tensor Model

The tensor model has provided extremely valuable information regarding WM microstructure in epilepsy, yet DTI has a number of limitations. First and foremost is its incapability of resolving fiber crossing, which may result in erroneous estimations of diffusion anisotropy. Second, the tensor model assumes that diffusion is Gaussian, while in reality the complex intracellular and extracellular environment causes the diffusion of water molecules to deviate considerably from this pattern; this oversimplification prevents precise description of tissue microstructure. Many alternative analytical methods have been proposed in the last decade that extend the capabilities of diffusion MRI and are beginning to be used for the study of epilepsy as acquisition requirements are relaxed to the point that they can be applied in clinical populations. For example, diffusion kurtosis imaging, an extension of tensor model to measure the degree to which diffusion deviates from Gaussian behavior, i.e., mean kurtosis, yields complementary information regarding diffusion heterogeneity.⁵⁴ In TLE, in addition to revealing more marked diffusion profiles along ipsilateral WM fibers,^{55, 56} this technique has shown abnormalities in both gray and white matter extending to regions not detected by conventional DTI measures. Diffusion spectrum imaging, a method querying complex distributions of intravoxel fiber orientation, has revealed that reduced FA in the mesiotemporal lobes in TLE patients likely results from the reduction in intracellular diffusion related to neurite (i.e., dendrites and axonal) loss, whereas reduced FA in extratemporal regions was confounded by fiber orientation changes, potentially reflecting disorganized fiber orientation and packing.⁵⁷ Restricted spectrum imaging is based on a multicompartment model that quantifies the degree to which diffusion heterogeneity within a voxel is driven by intra-axonal water versus extracellular diffusion, while accounting for fiber orientation. This multishell acquisition method has detected more widespread reduction of neurite density than anisotropy.⁵⁸

Fixel-based analysis is a diffusion-weighted MRI reconstruction technique that combines the measurements of fiber cross-sectional area, a measure of morphology, with microstructural information of fiber density,⁵⁹ thereby providing a sensitive marker of intra-axonal volume. Fixel-based morphometry has assisted in identifying WM atrophy secondary to repeated focal seizures and global insults in TLE patients.³² Neurite orientation dispersion and density imaging, commonly referred to as NODDI,⁶⁰ is another advanced diffusion imaging reconstruction technique based on a multishell acquisition protocol that allows an estimation of intra- and extracellular volume fractions in order to measure neurites morphology. This technique has the advantage to model both gray and white matter, a desired ability in the investigation of epilepsy, yet so far not exploited. As more complex models are introduced to probe individual tissue components, histologic-MRI correlational studies may help validate the biological underpinning of the novel metrics.

7.5 White Matter Connectome

Growing evidence of extensive structural and functional abnormalities in epilepsy has emphasized the notion that epilepsy is a disorder affecting distributed neural networks. Recent methodological advances in graph theoretical analysis lend tools to characterize topological aspects of interconnected regions.⁶¹ In this mathematical formalism, the brain connectome is modeled using a parcellation scheme of gray matter regions, i.e., nodes. Importantly, WM pathways reconstructed by in vivo diffusion tractography provide a good approximation of the actual anatomical connectivity, forming the edges between nodes. Notably, however, there is so far no consensus on metrics of connectivity strength, with some studies adopting parameters based on deterministic tractography (including FA, number of streamlines, and fiber lengths), while others using fiber connections density derived from probabilistic tractography.

Most DTI-based network studies have focused on focal epilepsy, particularly TLE. Collectively, findings have argued for a reorganization of macrolevel topological properties. In local circuits, on the other hand, increased integration was found within limbic and default mode networks, even though fiber connectivity among these regions was reduced.⁶² This dichotomy of changes between large-scale topology and individual fiber level illustrates the advantage of network analyses in addressing how regions interrelate as a whole, which is not possible via querying the integrity of a specific connection, and may potentially provide useful diagnostic and prognostic information for the assessment of individual patients.⁶³ Indeed, combined with machine learning algorithms, presurgical connectome data have shown the ability to predict surgery outcome in 70% of TLE patients, an accuracy similar to “expert-based” clinical decision.⁶⁴ Moreover, the presurgical connection strengths between ipsilateral temporal and extratemporal regions have been able to predict seizure-free surgical outcome in TLE patients with performances far beyond that of presurgical clinical data.⁵⁵ Specifically, compared to seizure-free patients, presurgical connectomes in non-seizure-free patients exhibited higher integration within ipsilateral temporal lobe subnetwork.⁶¹

In light of work in healthy populations that utilizes structural connectomes to simulate functional dynamics,⁶⁵ recent advances in network diffusion science offer new models to simulate the communication process between network nodes and have provided preliminary evidence for the spreading pattern of neuronal damage from the seizure focus at a group level in TLE.⁶⁶ Emerging data suggest that computational models of seizure spread and epileptogenicity based on structural connectomes⁶⁷ may also help localizing surgical targets⁶⁸ and possibly improve the prediction of postsurgical seizure outcome.⁶⁹ In patients with IGE, network analysis based on diffusion MRI and resting state functional MRI data has identified altered topology in regions that had previously been hypothesized to play a role in the epileptogenesis of this condition, including the mesial frontal cortex, putamen, and thalamus.^{70, 71}

To date, WM organization in cohorts with validated FCD has been seldom studied. Our recent study employing gray matter structural covariance and functional connectivity analysis suggests marked network rearrangement particularly in FCD I.⁷² In patients with frontal lobe epilepsy and suspected subtle dysplasia, overall reductions of global and nodal efficiency have been observed, as well as network segregation. Given the macro-scale analyses, it is likely that some connectome findings may relate to cognitive difficulties commonly seen in epileptic populations. Indeed, a structural connectivity study reported that patients with severe cognitive impairment may have more marked alterations in network topology compared with healthy individuals as well as patients with only little cognitive impairment.⁷³

7.6 Conclusion

Epilepsy is a network disorder and the WM is the substrate connecting its various components. A proper conceptualization of this condition entails equal attention to both the gray and white matter at the site of the seizure focus and at a distance from it.⁷⁴ Novel in vivo diffusion-weighted MRI acquisition and reconstruction techniques, but also emerging imaging methods assessing myelin content,⁷⁵ provide an increasingly comprehensive view of this pivotal compartment and its relationship to neuronal processes. They have successfully characterized pathology and mapped its regional distribution. Although initial studies have suggested the relevance of single markers of WM integrity to clinical outcomes, a better understanding of this relationship demands combined analysis of descriptors of gray and white matter microstructure, connectome-level features, and lesional markers. In this context, machine learning is the method of choice to extract critical features, patterns, and relationships from such high-dimensional datasets that might otherwise be missed. Moreover, paralleling the move toward big data analyses in neuroscience, epilepsy research would benefit from data-sharing initiatives to validate results and address new hypotheses.