Introduction

Despite the methodological challenges encountered in implementing the technique, it has been shown to provide a new form of localizing information and even shed new light on the physiology of interictal and ictal epileptic phenomena.

This chapter is organized as follows: we begin by reviewing the results of the application of EEG-fMRI in patients with both focal and generalized epilepsy. Central to the technique is the modeling of the BOLD signal based on the observed EEG, which forms a large part of the discussion. This has important implications for not only the technique’s sensitivity, and therefore its potential clinical usefulness, but also our understanding of the relationship between EEG activity and brain hemodynamics. The main technical and methodological issues of EEG-fMRI are reviewed at the end of the chapter.

Interictal Activity in Focal Epilepsy

The primary aim of developing EEG-fMRI in the first instance was to infer the location of irritative and seizure onset zones with the hope of providing useful clinical information, particularly in patients undergoing presurgical evaluation.

Using an interictal epileptic discharge (IED) triggered technique, whereby images were only acquired following the observation of an EEG event of interest and compared voxel by voxel to images acquired following periods of background EEG, initial studies in selected patients with focal epilepsy revealed significant BOLD increases in a large proportion of cases that were reproducible,¹ although the statistical tests applied varied widely²^–⁸ (see Table 6-1). It should be noted that the early studies focused on the identification of BOLD signal increases, neglecting the possibility of BOLD decreases, which were subsequently shown to be of considerable interest (see discussion later in this chapter).

TABLE 6–1 Significant Case Series and Milestones in EEG-fMRI (Interictal Studies)

The continuous EEG-fMRI acquisition approach was made possible by the use of EEG processing methods to correct scanning-related artifact ⁹ (see the section on technical aspects at the end of this chapter for further discussion). The analysis of this type of data requires the creation of a model of the BOLD signal over the entire experiment by identifying images that coincide with EEG events of interest, such as IED.¹⁰ This model is then used to find voxels with BOLD signal time courses using correlation. A map of the IED-correlated BOLD signal change is then created across the brain (for examples, see Figures 6-1 and 6-2). The identification of the events of interest is usually made by human observers and suffers from the well-documented limitations of this approach.¹¹ In addition to the limits of an observer-derived EEG model per se, not all patients will have IEDs in the scanner, thus presenting a further difficulty. In the two largest case series of continuously recorded EEG-fMRI in focal epilepsy,^12,¹³ around 50% of patients had IEDs during scanning. Approximately 60% of these had a BOLD signal change recorded in the vicinity of the electroclinical localization of seizure onset (known as concordant BOLD signal change; see Figure 6-1 for illustration). This figure represents the “yield” of EEG-fMRI in a given study. It is improved by modeling runs of IEDs rather than single events in addition to improvement in the EEG model by adequate spike detection and classification by the observer.^11,^13,¹⁴ The patterns of IED-related BOLD signal changes can be complex, with clusters commonly observed in close proximity to or overlapping the presumed seizure onset zone and remote from the epileptogenic zone and irritative zone regions of BOLD decrease most often observed at remote sites. The interpretation of these remote changes remains one of the active areas in EEG-fMRI research.

Figure 6–1 Positive BOLD signal change correlated with frequent right midtemporal sharp waves in a patient with right temporal lobe epilepsy (maximum T8), overlaid on normalized T1-weighted structural MRI in MNI space (coronal plane). Intracranial recording confirmed an extensive irritative zone in the right midtemporal lobe extending posteriorly (SPM-T p < 0.05 familywise error correction for multiple comparisons, t = 6.42 at voxel of maximum intensity).

Figure 6–2 SPM{F} from a patient with frequent GSW discharges. Top left: BOLD response correlated with GSW in glass brain display (the red arrow marks the global maxima). Top right: design matrix: first column shows the effect of interest (GSW) convolved with canonical HRF; subsequent columns represent motion and cardiac regressors. At the bottom: BOLD response overlaid onto normalized T1-weighted MRI in MNI space (p < 0.05 corrected). Note bilateral thalamic and right cingulate gyrus activation, right precuneus (global maxima showed by cross hair), bilateral middle frontal gyrus, right occipital lobe (lingual gyrus) deactivation (localization confirmed using Talairach coordinates in Talairach Daemon, http://ric.uthscsa.edu/project/talairachdaemon.html).

Image courtesy of Dr. A. E. Vaudano.

As a general rule, the results of EEG-fMRI studies in focal epilepsy suggest that the time course of the BOLD increases (known as the hemodynamic response function, or HRF12,15) related to focal spikes matches the so called canonical shape (i.e., peaks at 5 to 6 seconds after the event, returning to baseline roughly 15 seconds later), of responses observed in relation to events (e.g. stimuli) in healthy brains.

Deviations from the normal time course, however, have been observed in epilepsy. Although the BOLD response, both positive and negative (to external stimuli), has been studied extensively and is well documented in reference to neuronal activity in healthy subjects, some have suggested that this relationship may be altered in the case of interictal epileptiform discharges.¹⁶^–¹⁸ A formal study investigating this possibility found that the inclusion of noncanonical time courses does not lead to an important increase in yield.¹² A more recent study suggests noncanonical HRFs associated with IEDs are often remote from the presumed seizure onset zone,¹⁹ but whether this represents artifact, propagation of epileptiform activity (time-locked activity) or another phenomenon is not clear. Others have, however, emphasized the importance of intersubject variability in the HRF and routinely use individualized HRFs to analyze IED-correlated fMRI,^16,²⁰ and the issue is still under debate. Early BOLD signal change (time locked to IEDs) has been reported in focal epilepsy,²¹ and BOLD activation has also been reported in the thalamus prior to the onset of generalized spike and wave discharges.²² It is conceivable that these reflect neuronal activity time-locked but preceding an event observed on the scalp EEG and for example may represent a phenomenon similar to observed preictal BOLD signal change in humans²⁵ and preceding the occurrence of epileptic spikes in an animal model.²⁶ Another possibility is that this is a reflection of abnormal neurovascular coupling.

However, the specificity of this type of BOLD time course to epilepsy remains to be properly assessed.

Furthermore the relationship between blood perfusion and BOLD changes linked to epileptiform discharges in focal (one case) and generalized epilepsies are consistent with preservation of neurovascular coupling in epilepsy.^23,²⁴

LOCALIZATION OF BOLD CHANGES IN FOCAL EPILEPSY

As has been discussed earlier, IED-correlated BOLD signal change may be colocalized with the irritative or epileptogenic zones.²⁷ We will now discuss the efforts to validate EEG-fMRI localization by comparing other noninvasive localization methods and the gold standard of intracranial recording.

Some reports, predominantly within larger series, of “concordance” of IED-correlated BOLD response with seizure onset were identified by intracranial recording,^3,^4,^12,¹³ but the only more systematic study was limited to a five-case series. Here it was found that where EEG-fMRI activations were identified, at least one active electrode was on intracranial recording in the same location. The same series made a comparison with source analysis from standard scalp EEG.²⁸

Comparison of noninvasive EEG source localization and EEG-fMRI activations have been made using both spike-triggered fMRI and continuous acquisition. An initial study observed that dipolar sources were often remote (2 to 6 cm) from EEG-fMRI localization.²⁹ This was corroborated by a separate study that suggested distributed source analysis might be a more effective way of comparing EEG source and BOLD activation.³⁰ A recent systematic study using calculated measurements over the cortex revealed IED-associated BOLD clusters that were highly concordant with distributed sources in most patients’ recordings, but also that other EEG-fMRI sources were present that were not concordant with the distributed sources,³¹ particularly negative BOLD responses. An initial study observed that multi-dipolar sources were generally in anatomical (lobar level) agreement with EEG-fMRI localization based on BOLD increases, and lesser agreement with BOLD decreases.³²

Despite comparative studies made between modalities as described earlier, the role of EEG-fMRI in presurgical evaluation remains undefined. The only attempt to assess its practical benefit suggests a limited role in evaluating the group of patients where surgery was not able to be offered.³³ Eight patients from this group had significant IED-correlated BOLD signal increase concordant with presumed seizure focus when studied with EEG-fMRI. Of these, four had a unifocal activation, which narrowed the location of the seizure onset zone and was concordant with intracranial recordings in two. In four patients, multifocal BOLD activation concordant with electroclinical evaluation was observed. Larger studies comparing multimodal invasive and noninvasive investigation in addition to postsurgical outcome are required to complete the picture.

BOLD CHANGES ASSOCIATED WITH SPECIFIC PATHOLOGIES IN FOCAL EPILEPSY

It is known in lesional epilepsy that the irritative zone and epileptogenic zone may extend beyond the anatomical boundary of pathological abnormality, and in addition, both in vitro and animal models suggest abnormal subpopulations of neurons within dysplastic areas.³⁴ EEG-fMRI has therefore been used as a tool to evaluate the hemodynamic response in areas of cortical malformation and other pathologies.³⁵^–³⁹

In general, EEG-fMRI studies of malformations of cortical development have shown variability in BOLD response within pathologically abnormal regions with broadly concordant activations reported in both gray matter heterotopia ³⁵ and focal cortical dysplasia.³⁸ In both these studies, BOLD decreases were observed remote to the area of pathological abnormality. Other studies of malformations of cortical development (MCD) have supported these findings.³⁷ The frequent IED-related BOLD decreases, particularly in MCD, have been attributed to loss of neuronal inhibition (in the presence of normal neurovascular coupling) in the regions surrounding the abnormality or abnormalities in neurovascular coupling itself. The significance of these deactivations will be discussed in more detail later.

A recent study of IED-related BOLD signal change in cavernomas ³⁶ suggested BOLD signal change both within the area of anatomical abnormality and within areas remote to it. Caution is required in interpreting BOLD signal change in these very vascular lesions as T2* sequences used in EEG-fMRI are very sensitive to hemosiderin within the lesions.

EEG-fMRI results in tuberous sclerosis reflect a similar pattern. In a recent study exclusively using pediatric patients, it was found that BOLD activation was variable between tubers and extended beyond the border of tubers as identified on structural images once again supporting the view that EEG-fMRI may be a useful technique in assessing the epileptic network beyond areas of structural abnormality.⁴⁰