Source localization of interictal epileptiform discharges (IEDs) recorded with electroencephalography has traditionally been performed via visual analysis. As Rose and Ebersole have written, visual analysis relies on three assumptions: (1) the electrode(s) demonstrating the highest amplitude abnormal potential directly overlie the generator, (2) a cortical generator always produces a focal potential, and (3) a widespread potential indicates a diffuse source or multiple sources (Rose and Ebersole, 2009). All three assumptions are frequently incorrect. To go beyond visual analysis, a variety of mathematical techniques of source localization have been proposed and utilized including equivalent current dipoles (ECD; Henderson et al., 1975), minimum norm estimates (MNE; Hauk, 2004; Silva et al., 2004), standardized low-resolution brain electromagnetic tomography (sLORETA; Pascual-Marqui, 2002), and dynamic statistical parametric mapping (dSPM; Tanaka et al., 2009). Each of these techniques have advantages and disadvantages, but particularly in combination with higher density EEG arrays and magnetoencephalography, the single ECD model has proven value in helping to identify the ictal onset zone in neocortical epilepsies (Huiskamp et al., 2010; Shiraishi, 2011). As an example, source estimates performed using a single ECD model for IEDs which appear predominantly unifocal in their generation such estimates have proven to be highly accurate, falling within millimeters of those demonstrated by electrocorticography (Ishibashi et al., 2002). Despite this success, the single ECD solution is limited, best describing non-moving sources. Sources that move over time have been described by moving dipoles or multiple dipole models (Ochi et al., 2000), or iterative application of the ECD model (Papanicolaou et al., 2006), but mathematical solutions that incorporate more details of the complexity of the generating cortical tissue have multiple solutions, become computationally intractable, or both (Yetik et al., 2005). Alternative methods for IED source localization such as minimum norm estimation have been proposed, and may be theoretically better than ECD given the potential complexity of the generators involved, but the vast majority of clinical validation has been performed with the ECD model (Wheless et al., 1999; Pataraia et al., 2004). Given the fairly large area of cortex involved in spike generation (at least those identified on scalp EEG which serve as the basis for MEG ECD localization), the limitation of the ECD method in reducing cortical sources to dimensionless points renders it less than ideal. While appropriate pre-processing of the signal data and adjustments to the head model can produce significant improvements in source localization results, all of these methods generally evaluate a single point in time, so that a spike, for example, is reduced to the moment of peak negativity (unless you use the ECD iteratively). An analysis method that evaluates signal change over time might contribute useful information to the existing models.

In the field of optics, the goal of optical flow estimation is to compute an approximation to the motion field from time-varying image intensity (Fleet and Weiss, 2005). Optical flow analysis of distributed source models of MEG or EEG data has recently been proposed as a means to estimate the current motion field of cortical activity, or “cortical flow.” This technique can be used to estimate local kinetic energy of cortical surface currents, and has been used to characterize correspondence between the speed and direction of the surface current flow within the visual cortex and the dynamical properties of the visual stimulus itself (Lefevre and Baillet, 2008, 2009). The motion field so defined can be used to identify dynamic features of interest such as patterns of directional flow, current sources, and sinks.

The Helmholtz–Hodge Decomposition (HHD) is a technique frequently applied in fluid dynamics to separate a flow pattern into three components: (1) a non-rotational scalar potential U describing sinks and sources, (2) a non-diverging scalar potential A accounting for vortices, and (3) an harmonic vector field H (Chorin and Marsden, 1993; Tong et al., 2003). A recently published abstract demonstrated the use of the HHD for mapping and characterizing current flow over primary somatosensory cortex during sensory-stimulation triggered evoked potentials (Khan et al., 2009). As IEDs seem likely to represent periods of highly correlated directional flow of cortical currents, the HHD U potential lends itself to the characterization of spikes in terms of current sources and sinks. The methodology is reviewed in greater detail in the addendum.

In this study, its relative efficacy compared to that of the standard ECD model, was assessed with a series of six candidates for epilepsy surgery.

The sequential results for the first patient (2092) are presented in detail. The original spike prior to filtering is shown in Figure 1. The same spike after 30 Hz low-pass filtering is illustrated in Figure 2. The same discharge in the context of alpha background rhythm demonstrating a right temporal preponderance on MEG is shown in Figure 3. The MEG demonstrated only a weak dipolar map over right temporal area. The plot of the calculated ECD on the patient’s MRI, revealing a right mesial temporal localization, is shown in Figure 4. The DE map, with the source and sink time points marked, is shown in Figure 5.

Cortical surface localization of spike source and sink are shown on Figures 6 and 7 respectively, where blue shading of the cortical manifold indicates the outward current flow or source, and red shading indicates the inward current flow or sink. For those figures, the arrow size in each region gives an approximation of relative current magnitude.

The remaining five patients are summarized in Figures 8–12. The first Figure in each set (a) shows the original spike or sharp wave as identified on scalp EEG. For each patient, the best example of a given spike population has been chosen for presentation. The second figure (b) is the HHD source map. The third figure (c) is the HHD sink map. The fourth figure (d) is the combined plot of the DE and GFP.

Table 1 lists the six patients whose interictal activity was analyzed, including: the dipole localization using ECD of most of IEDs recorded during the session, whether or not the HHD topographic plot was concordant with the ECD localization and the pathologic diagnosis if known. For five out of six subjects, the HHD projection was concordant with the ECD localization. With the sixth, the ECD localization was in the left perisylvian region, concordant with the pathology visible on MRI, but the HHD sink and source, while broad and low amplitude, were maximal over the lateral frontal lobe.

For most cases, spikes and sharp waves recorded with simultaneous EEG/MEG, the source and sink areas defined by HHD of the optical flow analysis of the MEG signals appear concordant with the location identified by ECD. Where the two techniques are in disagreement, the divergence may be due to a failure of the OF/HHD to make an accurate localization, or it may be due to an error on the part of the ECD localization. An additional possibility is that neither is in error, but rather the area of cortex identified by each model is different simply because the models represent alternative aspects of the biomagnetic activity (a point dipole is not a current flow). The OF analysis is based on the minimum norm estimate source modeling of the magnetic activity, and so is subject to the limitations of that model. By its nature, OF/HHD describes a region, not a point, and given that the underlying cortical activation generating a discharge visible on a scalp recording is virtually never a point, this may be a more intuitive representation. Compared to the minimum norm estimate, the OF/HHD analysis may have an advantage in the defined direction of current flow, which not only allows for tracking the region of involvement over time, but the characterization of regions of cortex as source and sink for the discharges. At a minimum, this allows for differentiation of stationary versus “moving” sharp waves. Whether the area of involvement reflects source depth or the extent of the epileptic zone remains speculative at this time. As might be expected, the point in time of the traditional spike “peak” frequently occurs at a point of maximal GFP and minimal DE. This was less consistent when other high energy and or high frequency activity occurred concurrent with the spike generation, and a more accurate depiction might result from limiting the analysis to a region of interest, rather than the entire cortical surface.

One limitation of the current study is the relatively small number of patients included in the analysis. While any number of factors might prevent a subject from inclusion during the period of data collection, the most common reason was that no interictal discharges were recorded during the MEG session. A second limitation is the lack of medial cortical surface views, particularly critical for patients with mesial temporal discharges. In the instance of subject 2406, where the HHD localization was discordant from that of the ECD, the absolute amplitude of the source and sink flow vectors are less than maximal. As the magnitude of the flow vectors are normalized for each HHD image, this implies that there sink and source regions of greater amplitude that are not visible with the available set of views. The software used for the current analysis lacks the capacity to present these views, but one may anticipate that increased use in the community of the HHD analysis component of the software will result in this functionality being added.

Many other questions remain to be answered. Defining the relationship (if any) between the source/sink areas and the epileptogenic zone is important, but of greater importance is discovering if OF/HHD can be used in relationship to the eventual surgical resection to predict outcome. Association of particular source/sink patterns with specific pathology should also be investigated, in addition to the potential effects of drugs such as antiepileptic medications.

This study represents the first use of OF/HHD techniques that we are aware of for the localization of epileptic activity. While this approach is novel, it is important to emphasize that despite the appearance of the graphics and the terms source and sink giving the appearance of electrical currents over the cortex, it seems exceedingly unlikely that the actual surface dynamics of the electrical activity of the cortex resemble that which is depicted. The MEG spike data suffers from the same limitations that all MEG recordings do, namely that the MEG is relatively insensitive to superficial radial sources (Nunez and Srinivasan, 2006). While the argument has been made that use of the MNE is the optimal solution for the inverse problem of bioelectromagnetic source localization (Hauk, 2004), the model remains limited by the data upon which it is based. Thus the eventual value of the OF/HHD analysis of spikes will be only determined by the degree to which it demonstrates clinically useful correlations with pathologic brain states.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.