Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Aug 13.
Published in final edited form as: Clin Neurophysiol. 2025 Aug 13;178:2110971. doi: 10.1016/j.clinph.2025.2110971

Magnetoencephalography distributed source modeling identifies spatiotemporal propagation patterns of interictal epileptiform discharges

Elena Hayday 1, Kaya Scheman 1, C Price Withers 1, Kayla Togneri 1, Kareem A Zaghloul 2, Jeff Stout 3, Antonio Ivano Triggiani 1,*, Sara K Inati 1,*
PMCID: PMC12434894  NIHMSID: NIHMS2107658  PMID: 40845408

Abstract

Objective:

Magnetoencephalography (MEG) is an important adjunctive method used to localize interictal epileptiform discharges (IEDs). The equivalent current dipole (ECD) method, the current gold standard modeling approach, identifies a single point source of activity. However, IEDs propagate widely and may be better represented by distributed source modeling approaches. Here, we investigate how areas of maximal IED-related activity estimated using dynamic statistical parametric mapping (dSPM) compared to dipoles and surgical resections.

Methods:

We analyzed resting-state MEG recordings from 38 NIH patients. We localized areas of IED-related activity along the IED rising phase and peak using spatial clustering of the top 5% of dSPM activations, comparing localizations to ECD and surgical resection areas in seizure free patients.

Results:

We identified dominant primary activation clusters in all patients and non-primary clusters in 24/38 patients. Dipoles localized closer to primary than non-primary clusters. In 12 post-operative seizure-free patients, the primary cluster center of mass was significantly closer to the resected area than dipoles and more stable over time.

Conclusions:

Distributed source modeling adds to our understanding of IED propagation patterns and may be useful in IED localization for epilepsy surgery planning.

Significance:

dSPM enhances pre-surgical planning by refining IED localization, potentially improving surgical outcomes.

Keywords: dSPM, IED, epilepsy, MEG, source localization

Introduction

Surgical interventions are often effective in the treatment of drug-resistant epilepsy, with outcomes highly dependent on accurate seizure focus localization. Magnetoencephalography (MEG) is a non-invasive imaging modality with demonstrated localization utility in complex cases, providing whole brain coverage at relatively high spatial and temporal resolution (Bagic et al., 2011s, Knowlton et al., 1997, Tamilia et al., 2017). Due to limited recording times, MEG typically relies on localization of interictal epileptiform discharges (IEDs), which occur more frequently than seizures (Alarcon et al., 1997). IEDs are brief electrographic events that likely arise from similar mechanisms and spatially overlap with seizures onset zones, although they are less specific at least in part due to rapid propagation (Alarcon et al, 1997, de Curtis et al., 2012). This propagation is thought to arise from focal sources, spreading both locally along the cortical surface and distantly over white matter pathways (Rossi et al., 2017, Diamond et al., 2023, Withers et al., 2023). Both identifying the sources and understanding propagation patterns of this widespread activity are therefore critical to improving the accuracy of non-invasive imaging modalities to localize seizure foci.

Magnetic source imaging (MSI) uses modeling to estimate the neural source(s) of sensor-level activity. The equivalent current dipole (ECD) method is the current clinical “gold standard” for IED localization, validated in large patient cohorts against surgical resection areas and seizure outcomes (Bagic et al., 2011, Knowlton et al., 1997, Stefan et al., 2003). However, the single ECD model assumes and therefore identifies a point source of activity at each timepoint.

Distributed modeling approaches may better reflect the often widespread propagation of interictal activity observed in intracranial EEG recordings (Rossi et al., 2017, Withers et al., 2023, Matarrese et al., 2023). Dynamic statistical parametric mapping (dSPM) is one such approach in which spatial maps of brain activity with millisecond temporal resolution are created, with enhanced spatial accuracy obtained through anatomically constraining sources of activity to the cortical gray matter and incorporating noise-normalized minimum norm estimates (Dale et al., 2000). dSPM was initially introduced to study MEG evoked responses but is also particularly well-suited to identifying brain regions involved in the often highly synchronous and widely propagated activity that occurs during IEDs (Dale et al 2000).

Although several studies have implemented a dSPM-based approach to MEG IED source localization (Shiraishi et al., 2011, Tanaka et al., 2009), there has been limited work to date investigating IED propagation patterns using this approach and comparing the spatial relationships between distributed areas of IED-related activation and the point source localizations obtained using dipole modeling. In this study, we identify cortical areas with maximal IED-related dSPM activation along the rising phase of IED spikes in individual patients. We estimate sources of this activity using the center of mass within and across areas of maximal activation and compare them with dipole localizations and with resection areas in patients who became seizure free after epilepsy surgery. This work helps to increase our understanding of how point source localizations can be reconciled with observations of widespread IED-related activity and also describes a method to localize focal sources leading to widely distributed IED-related areas of activation.

Methods

Study Participants

We retrospectively identified MEG recordings from 223 patients with drug resistant epilepsy undergoing presurgical evaluation at the National Institutes of Health Clinical Center between February 2002 and April 2024. We included patients with a MEG study at the NIH of acceptable quality (no limitations due to artifacts) and a co-registered 3D T1-weighted (T1w) MRI without major anatomical abnormalities. Patients with fewer than 10 recorded interictal epileptiform discharges (IEDs) with similar morphology and distribution were excluded. We identified the subset of patients who underwent surgical resection, and reported their seizure outcomes with at least 12 months postoperative follow up using the Engel classification (Engel Class 1 for good seizure outcomes, >1 for persistant seizures) (Engel 1993). This non-invasive epilepsy imaging protocol 14-N-0061 was approved by the NIH Institutional Review Board, and all patients provided written informed consent.

MEG Acquisition

Resting-state MEG data were recorded in 6-minute eyes closed runs with a sampling rate of 600 or 1200 Hertz (Hz) using a 275-channel whole-head CTF MEG system (CTF Systems, Inc., Canada) with 3rd gradient noise cancellation. Runs with persistent artifact or head movement greater than 5 mm were excluded from analysis. Runs obtained at 1200 Hz sampling rate were downsampled to 600 Hz prior to analysis. Simultaneous low density and in later subjects high density 64-channel EEG was recorded in 32 patients. Electrical source imaging was not performed as part of this study.

MRI Preparation and Head Model

T1w MRI was acquired at either 1.5 or 3 Tesla. MRI co-registration was performed based on identification of three fiducial points mirroring the MEG head position coil locations at the nasion, left and right preauricular points. In 12 patients, this was carried out using Brainsight (Rogue Research; Montreal, Quebec) software prior to MEG acquisition. In all other patients, fiducial points were estimated manually on each patient’s MRI, when available using radio-opaque fiducial markers placed at these three locations prior to the MRI, using Analysis of Functional NeuroImages (AFNI; NIMH, NIH, Bethesda, MD) software (Cox 1996, Cox and Hyde 1997). Cortical surfaces were reconstructed using Freesurfer 6.0.0 (https://surfer.nmr.mgh.harvard.edu/). For dSPM analysis, a boundary element model (BEM) was created from each patient’s T1w MRI using MNE software (Segonne et al., 2004). For ECD analysis, the Nolte method was used to create a realistic head model using CTF software (CTF MEG Neuro Innovation; Coquitlam, British Columbia) and NIMH MEG Core facility in-house code (Nolte 2003). In patients who underwent resective surgery and received a T1w MRI at least one year postoperatively, resection masks were created for spatial comparison (Supplementary Methods).

MEG Pre-Processing

For each subject, we visually selected runs based on signal quality and presence of IEDs. Raw MEG signal was initially inspected using 1 to 100 Hz bandpass filters, and for later analysis was filtered from 5 to 50 Hz. These filter settings were selected based on the expected duration of IEDs from 20 to 200 ms (Kural et al., 2020), and to reflect the 0.1–7 high-pass to 30–40 Hz low-pass filters previously implemented in dSPM IED source localization papers (for example Tanaka et al., 2014, Shiraishi et al., 2011). They are also similar to those recommended in the ACNS guidelines of 1–4 high-pass to 40–70 Hz low-pass filters (Bagic et al., 2011). Prior to dSPM analysis, the filtered signal was decomposed into 30 components using ICA with components containing artifact (max. 3 components) as well as individual time periods with significant artifact visually identified and rejected. For each patient, we then identified a single population of similar-appearing IEDs using a semi-automated in-house ICA-based algorithm. A single ICA component of interest was selected by an investigator (EH, KT) blinded to additional localization information followed by automated peak detection in this component of interest. In included subjects, at least ten of the resulting peaks were clinically validated by an epileptologist reviewer with experience in clinical MEG interpretation (SI) to be included for further analysis (see Supplementary Methods). We created two second “spiking epochs” from the 1.5 seconds preceding to 0.5 seconds following the marked spike peaks, rejecting epochs containing polyspikes or significant artifact. If at least 10 spiking epochs remained after rejection, the remaining epochs were averaged in CTF and in MNE-Python software to obtain a 2-second average spike epoch (Fig. 1A–B) (Gramfort et al., 2013). We visually identified “spike free” two-second epochs to use for background noise removal.

Figure 1. Study methods.

Figure 1.

Open in a new tab

(A) Example interictal epileptiform discharge (IED) in selected MEG sensors. (B) Topographic plot of the MEG activation across all sensors at the peak of the same IED (C) Determination of timepoints of interest from the dipole fit error of a moving dipole fit from 25 milliseconds (ms) to the peak of the averaged spike. (D) Moving dipole fit from “first” to “peak” over the averaged spike in one patient. The legend represents ms before the marked peak. (E) dSPM source estimate at the marked peak of the spike in MNE-Python in one patient. (F) dSPM cluster selection in one patient. The legend represents the dSPM amplitude at each voxel. (left) The dSPM values at every gray matter voxel are shown. (right) Clustered voxels with amplitudes greater than the 95th percentile were selected as the primary cluster, represented by the yellow color. (G) Spatial comparison between source estimates. Primary and secondary clusters are represented by colored circles, each of which has its own center of mass (COM), represented by a dot in each circle. The COM of all dSPM clusters (COMC) is represented by the black star, and the equivalent current dipole (ECD) solution is represented by the blue square.

ECD Analysis

We fit a moving single equivalent current dipole (ECD) along the upslope of the averaged IED using commercial software provided by CTF Systems (VSM MedTech), including all MEG channels that were not removed a priori for noise, with resulting sources unconstrained within the volume space. All included patients had global fit error of less than 30% (or goodness of fit greater than 70%) along the upslope or at the spike peak (Fig. 1C). We identified two timepoints of interest. The “first” dipole localization was obtained at the earliest timepoint within the 25 ms window preceding the spike peak with an overall global fit error of less than 30%, or goodness of fit greater than 70% (Bagic et al., 2011, Shiraishi et al., 2011, Tanaka et al., 2009). “Peak” localization was at the marked spike peak. The 25 ms time window was chosen empirically to be large enough to include the “first” timepoint and small enough to remain on the IED spike upslope without inclusion of preceding waveforms. In accordance with previous studies (Kobayashi et al 2005), we also calculated metrics at the timepoint with the lowest dipole fit error (“best”). The median difference between “best” and “peak” was 0 ms; further, the dipole and cluster metrics for “best” did not significantly differ from “peak” and were thus left out of the main analysis (Supplementary Fig. 1).

dSPM Analysis and Clustering

A noise covariance matrix was constructed in MNE-Python using spike-free epochs. The dSPM inverse solution was calculated and applied to the averaged spiking epoch smoothed with a moving average with fixed window of 30 samples (Fig. 1E) (Gramfort et al., 2013, Gramfort et al., 2014). To identify brain regions with maximal IED-related activation at the first and peak timepoints, virtual sensor locations were spatially smoothed on the white matter surface using a Gaussian filter (full width at half maximum [FWHM] of 20 mm) using AFNI’s SurfSmooth. We mapped dSPM amplitudes into the volume on the white matter surface using AFNI’s 3dSurf2Vol using two iterations of modal smoothing using @ROI_modal_grow to fill in resulting gaps. We identified regions with maximal IED-related activation by thresholding dSPM amplitudes at the 95th percentile and clustering using AFNI’s 3dclust, with a minimum cluster size of 75 mm3 (Fig. 1F). If multiple clusters were present, we identified the primary cluster (P) using a weighted measure of average dSPM amplitude multiplied by cluster volume. All other clusters were designated as non-primary (NP).

Spatial Comparison

The overall dSPM activation-related center of mass (COM) across all resulting clusters (COMC) was obtained by the AFNI command 3dCM, weighted by the dSPM values within each cluster. Similarly, COM was found for each individual cluster (COMP or COMNP). At each timepoint, the Euclidean distance between the dipole localization and COMs were determined. Additionally, the Euclidean distance between the dipole localization and COMs and every point in the resection mask was calculated, and the shortest value was reported as the distance between the resection area and each localization metric (Fig. 1G). Between-cluster distance was calculated as the average Euclidean distance between COMC and the COM of each cluster. For patients with multiple NP clusters, distance was calculated to each COMNP and the average of those distances was used for statistical analysis. In a supplementary analysis, we performed similar spatial comparisons using the location of the voxel with the maximal or peak dSPM amplitude instead of the COM both within and across clusters (see Supplementary Methods and Results, Supplementary Figures 2–4). Unless otherwise noted, paired t-tests were used to determine the difference in localization distance between two metrics.

Results

Study Participants

Thirty-eight subjects were identified that met the inclusion and exclusion criteria (22 male; 25.18 ± 12.49 years of age) (Table 1, Supplementary Table 1). Of these, 50% had temporal and 50% had extratemporal lobe epilepsy. In this cohort, 21 (55%) underwent surgical resection, with 12 (57%) obtaining good surgical outcomes (Engel class 1) at their most recent follow-up appointment at least 12 months post-operatively (24 ± 15 months) (Engel 1993).

Table 1.

Patient Demographics.

All patients Patients with resection
Sex
Male 22 (58%) 14 (66%)
Female 16 (42%) 7 (33%)
Age at MEG (mean ± SD) 25.18 ± 12.49 22.24 ± 11.77
Epilepsy duration (mean ± SD) 14.13 ± 11.74 13.05 ± 10.85
Localization
TLE 19 (50%) 12 (57%)
ETLE 19 (50%) 9 (43%)
Pathology
FCD 5 (13%) 3 (14%)
MTS 12 (32%) 10 (48%)
Tumor 1 (3%) 1 (5%)
Other 8 (21%) 4 (19%)
Non-specific 12 (32%) 3 (14%)
Resective Surgery
Y 21 (55%) 21 (100%)
N 17 (45%) -
Surgical Outcome
Seizure-free - 12 (57%)
Seizure-persist - 9 (43%)
Open in a new tab

ETLE: extratemporal lobe epilepsy; FCD: focal cortical dysplasia; MEG: magnetoencephalography; MTS: mesial temporal sclerosis; TLE: temporal lobe epilepsy

Volume of IED-related dSPM activation increases over time to the IED peak

We hypothesized that the volume of IED-related activation would increase along the rising phase of the averaged IED. To test this, we compared the dSPM areas of activation at the first and peak timepoints using the same activity threshold, defined as 95% of dSPM amplitudes at the first time point (Fig. 2A). As predicted, we found a significantly greater volume of activation at the peak (mean 60.9 cm3) as compared to the first timepoint (mean 23.1 cm3; p<0.001, paired t-test) (Fig. 2B).

Figure 2. IED-related dSPM cluster characteristics.

Figure 2.

Open in a new tab

(A) Timepoint selection for patient 10. Left Dipole fit error at every timepoint in the moving dipole fit. The blue circle represents the selected “first” timepoint, and the green circle represents the selected “peak” timepoint. Right Dipole position over the course of the moving dipole fit. The legend to the right represents the timepoint for each dipole position, in milliseconds (ms) before the spike peak. (B) Total cluster volume (mm3) is larger at the peak versus the first timepoint (N=38). (C) Number of patients with each number of clusters at both the first and the peak timepoints. (D) Primary cluster (P) volumes (mm3) are greater than non-primary (NP) cluster volumes at both timepoints in patients with both cluster types (N=24). Primary cluster volumes are shown on the left and the average volume of each patient’s non-primary clusters are shown on the right at both timepoints. Significance bars represent the results of paired t-tests *p≤0.05; **p≤0.01; ***p≤0.001; ****p≤0.0001.

IED-related dSPM activation often involves spatially discrete brain regions

We then explored whether the identified areas of maximal IED-related activation at each time point were contiguous, as would be expected with radial spread of activity along the cortical sheet, or spatially discrete, most likely occurring through white matter propagation from a common source. For this analysis, we identified the top 5% of dSPM activations at the first and peak time points independently, then used spatial clustering to identify regions of maximal IED-related activation. At both timepoints, 24 patients (63%) displayed more than one spatially discrete area of activation, with similar numbers of clusters at each time point (Fig. 2C). In each patient, we designated a primary cluster with the largest volume (after thresholding for the top 5% at each time point, averaging 21.3 cm3 at the first and 23.6 cm3 at the peak timepoint), with additional significantly smaller spatially discrete non-primary clusters observed in a majority of patients (average volume 1.1 cm3 at first and 1.6 cm3 at peak; both p<0.0001) (Fig. 2D). Given that the number of IED-related activation clusters did not change over time while the overall volume of IED-related activation increased, we sought to determine whether this increase in volume of activation occurred in the primary cluster, non-primary clusters, or both. The ratio of primary cluster volume to the next-largest cluster was 12:1 at the first timepoint and 12.5:1 at the peak. If the observed IED-related activity is indeed arising from a single focal source, these findings are most consistent with white matter propagation leading to epileptiform activity observed in multiple discrete brain regions at each time point, with subsequent radial expansion occurring to a similar extent in both primary and non-primary clusters.

ECDs localize closest to the primary activation cluster COM

We then sought to determine the spatial relationship between the ECD localizations (dipoles) and the dSPM activation clusters over time. At each time point of interest, we assessed the Euclidean distance between the dipole and the center of mass across clusters (COMC), as well as the primary (COMP) and non-primary (COMNP) clusters. A mixed-effects analysis was carried out to account for patients who did not have any non-primary clusters and thus no COMNP value. We found that the Euclidean distance between the dipole and the three COMs differed significantly across timepoints (p<0.0001). Using Tukey’s multiple comparisons testing following the mixed-effects analysis, we found that the dipole was located significantly closer to the COMP (mean 8.6 mm at first; 3.2 mm at peak) than the overall COMC (mean 31.8 mm at first; 20.39 mm at peak) at both timepoints (p<0.0001). The distance from the dipole to COMP was also significantly shorter than the distance to COMNP at both timepoints (first: mean 28.1 mm; p=0.0001; peak: mean 29.9 mm; p<0.0001). Additionally, the dipole localization was significantly closer to COMC at the spike peak versus the first timepoint (Fig. 3A; p=0.03).

Figure 3. Dipole and cluster position over time.

Figure 3.

Open in a new tab

(A) Distance from the dipole localization to cluster localization metrics at the first and peak timepoints. The center of mass (COM) of all clusters is represented by COMC, the COM of the primary cluster is represented by COMP, and the COM of all non-primary clusters is represented by COMNP. The significance bars represent the results of Tukey’s multiple comparisons test following a mixed-effects analysis to account for patients who did not have COMNP (p<0.0001). (B) COMC moves less than dipole localizations from the first to peak timepoint (paired t-test p=0.04). (C) The average distance between the individual clusters (COMP and COMNP) and the overall COMC is similar at the first and peak timepoints. The significance bar represents the results of a paired t-test (p>0.05). *p≤0.05; **p≤0.01; ***p≤0.001; ****p≤0.0001.

IED-related dSPM localizations are more spatially stable than dipole localizations

Comparing movement of localizations over the IED rising phase, we found that the dipole moved significantly more (average 1.13 cm) than the overall COM across dSPM activation clusters (COMc) (0.52 cm, p=0.002, paired t-test; Fig. 3B). Further, the mean distance from each cluster to the overall COM (cluster-to-COMC) did not significantly differ between the first and the peak timepoint (first: 25.2 mm, peak: 22.9 mm; p=0.79, paired t-test). These results indicate that dipole movement, rather than a change in dSPM cluster localization, underlies the change in dipole-to-COMC distance between the first and peak timepoints (Fig. 3C).

IED-related dSPM clusters lie closer to the resected area than the ECD localization

Concordance of localization results with the area of resection in patients who become seizure free (N=12 in our cohort) after epilepsy surgery remains the gold standard for identification of the epileptogenic zone (Fig. 4A). In these patients, we found that the shortest Euclidean distance from the area of resection to the dipole, COMC, or COMP differed significantly at both the first and the peak timepoints (p=0.0007 and p=0.0018 at first and peak, respectively; mixed-effects analysis). At the first timepoint, the dipole localized significantly further from the resection (mean 40 mm) versus the COMC (mean 5.8 mm; p=0.008) and COMP (mean 0 mm; p=0.0014) (Fig. 4B). Similarly, at the spike peak, the dipole localized further from the resection (mean 34.1 mm) than the COMC (mean 11.1 mm; p=0.0018) and COMP (mean 4.1 mm; p=0.002) (Fig. 4C). The COMp overlapped with the resection at both timepoints in all but one patient, suggesting similar accuracy along the spike upslope for this metric for most patients.

Figure 4. Distance to resection in seizure-free patients.

Figure 4.

Open in a new tab

(A) Example results for patient 10. Left The primary cluster (P) is shown in yellow, and the non-primary cluster (NP) is shown in orange. The center of mass of each cluster (COMP or COMNP) is represented by the dot in the middle of each cluster, and the total cluster COM (COMP) is represented by the black star. Right The surgically resected area is represented by the blue mask. (B) In seizure-free post-operative patients, Euclidean distance from the resection mask is less for COMC and COMP than COMNP and dipole localizations (N=12). (C) Euclidean distance at the peak timepoint to any point in the resection mask is less for COMC and COMP than COMNP and dipole localizations. Significance bars represent the results of Tukey’s multiple comparisons tests following a mixed-effects analysis to account for missing values in patients who did not have a non-primary cluster. *p≤0.05; **p≤0.01; ***p≤0.001; ****p≤0.0001.

When comparing primary to non-primary dSPM cluster activations, we found that the resection-to-COMP distance was significantly less than the resection-to-COMNP distance at both the first (mean 0.0 mm versus 17.8 mm; p=0.03; Fig. 4B) and the peak timepoints (mean 4.1 mm versus 19.1 mm; p=0.03; Fig. 4C). These metrics demonstrate that the primary cluster COM is most often overlapping or located within 5 mm of the resected area in seizure-free patients.

Discussion

In focal epilepsy, interictal epileptiform activity has been shown to propagate from focal cortical sources, often leading to widely distributed activity observed in EEG and MEG recordings. Dipole modeling of point sources of activity has remained the clinically validated gold standard for MEG IED source localization. In contrast, distributed source modeling approaches can allow for improved mapping of propagation of IED-related activity at the cost of lack of specificity in localization of the source of the observed activity. Here, we observed increasing propagation in the form in increased volume of IED-related cortical activity along the rising phase of IEDs using dSPM source modeling. Using spatial clustering of areas of maximal IED-related activation at an early and peak timepoint, we found a dominant largest “primary” cluster, as well as additional non-contiguous smaller areas of activity in a majority of patients at each time point, consistent with early propagation through white matter pathways. Estimating focal sources using the centers of mass (COM) of our dSPM clusters, we found that dipoles localize closer to the COM of the primary cluster than to the overall COM across clusters or to non-primary cluster COMs. In comparison to dipole localizations, both the primary and overall dSPM COMs localized more closely to the resection area in seizure free patients, with less movement over time along the upslope of the IED, suggesting that the dSPM approach may be a highly accurate and stable marker of the epileptogenic zone.

Previous investigations in animals and humans have demonstrated widespread propagation of interictal activity in focal epilepsy. IEDs visible on scalp EEG recordings have been estimated to require simultaneous involvement of at least 6–10 cm2 of cortex, as observed in simultaneous iEEG recordings (Tao et al., 2005). Local propagation has been proposed to occur from focal sources both contiguously along the cortical surface as traveling waves and non-contiguously following functional white matter pathways (Rossi et al., 2017, Diamond et al., 2023, Withers et al., 2023). The increasing volume of individual dSPM IED activation clusters from the first timepoint to the spike peak is consistent with local contiguous propagation, while the presence of spatially discrete non-contiguous areas of activation is most consistent with white matter propagation. Shiraishi et al. (2011) similarly observed discrete activation areas in MEG using dSPM in patients with diffuse spikes, attributed to multifocal IED onsets in some cases and propagation in others. Matarrese et al. (2023) similarly showed spatiotemporal propagation of IED activity outward from more circumscribed sources in iEEG recordings using a dSPM-based source localization approach, similarly finding that resection of the area of onset and not spread predict surgical outcomes. In this work, we analyzed an averaged waveform for each patient derived from a single IED population with similar appearance and distribution. Therefore, our multiple activation clusters likely also represent propagation from a single focal source rather than multiple independent sources. This is also supported by our surgical resection data, in which the primary COM nearly always overlapped with the surgical resection area in patients achieving seizure free outcomes, while the non-primary cluster COMs were more often located outside of the resection areas in these patients.

As the single dipole model remains the gold standard for clinical IED source localization, it is important to understand the spatial relationship between dipole point source localizations and areas of IED-related dSPM activation. To our knowledge, our study is the first to investigate the relationship between dipoles and multiple discrete regions of activation identified by distributed source modeling. We found that dipoles most often localized within 1–2 cm of the primary cluster COM, and significantly further from the overall or non-primary cluster COM. This suggests that dipole modeling tends to approximate the location of the largest region of IED-related activation, despite the presence of additional, more distant regions of activation. Previous studies have similarly found that dSPM and ECD solutions are often quite similar, particularly in the setting of relatively localized IEDs (Shiraishi et al., 2011, Tanaka et al., 2009). However, in the setting of more widespread or multifocal activity, dipole modeling may be misleading or fail to localize the activity altogether (Shiraishi et al., 2011, Kobayashi et al., 2005). In these cases, distributed methods such as dSPM may be of particular clinical utility. In our cohort, we found that both the COM across dSPM clusters and the primary dSPM cluster COM more closely localized to, and most often overlapped with, the area of resection in seizure-free patients as compared to the dipole solution. This result suggests that dSPM localizations are highly accurate in localizing the epileptogenic zone even in complex cases where dipole solutions may be less effective.

As dipole point source localizations may be less accurate in the setting of more widely propagated activity, there is ongoing debate regarding the optimal time points that should be used for IED source localization. Dipole localizations have been shown to move significantly over the IED rising phase (Lantz et al., 2003). Previous studies have suggested that earlier localizations may best reflect intracranial EEG findings at seizure onset, although lower goodness of fit due to low signal-to-noise ratios may lead to mislocalization, particularly in the setting of deep sources (Kobayashi et al., 2005). Midway through the rising phase of IEDs, localizations may most closely approximate the source of the activity due to better signal-to-noise ratios and more limited propagation (as in Lantz et al., 2003). Localization at the IED peak (as in Reinsberger et al., 2010, Oishi et al., 2006) benefits from maximal signal-to-noise ratios but may also mislocalize due to more widespread areas of propagation (Kobayashi et al., 2005, Lantz et al., 2003, Plummer et al., 2019). Here, we defined our “first” time point based on the IED rising phase crossing a relatively liberal goodness of fit threshold of 70% across all sensors (as in Reinsberger et al., 2010), noting that studies using subsets of MEG sensors often rely on more stringent goodness of fit thresholds. Others have used the timepoint with the smallest dipole fit error (Kobayashi et al., 2005, Shiraishi et al., 2011). In our data, we found no difference between localizations at this time point and the IED peak.

Given the importance of timing to dipole localizations, it was somewhat unexpected that using our dSPM approach we found very little movement of the overall dSPM activation COM across time points, with no significant change in number of activation clusters or in the mean distance between individual clusters and the overall COM. Similarly, both the overall and primary cluster COM nearly always overlapped with the area of resection in seizure free patients at both time points, which is the gold standard for seizure focus localization. In contrast and in accordance with previous studies, we found that the dipole localizations moved more than the dSPM COMs over time and moved closer to the dSPM overall COM over the spike course. Furthermore, the dipole localizations were located farther from the resection area at both time points. Our findings support that the increased signal to noise at the peak may lead to more accurate dipole localizations, particularly since propagation of IED activity to discrete spatial regions appears to have already occurred even at the earliest time point localizable using the ECD approach. These findings are consistent with those of Kanamori et al. (2013), who reported that dSPM localizations were more accurate than dipoles in the earlier time points compared to intracranial EEG recordings.

There are several limitations to our study. While our sample size is larger than many similar studies, this remains a carefully selected patient population consisting of patients with frequent, reproducible IEDs that were able to be averaged and modeled as dipoles. The dSPM method may be less robust in the setting of fewer spikes. Additionally, we excluded patients in whom dipole modeling failed to localize the activity with at least 70% goodness of fit. Advantages of dSPM modeling may be more apparent in these cases. In keeping with our clinical practice, we did not model multiple dipoles and did not constrain dipole localizations to the cortical surface, as we did for dSPM clusters. Similarly, we did not select subsets of channels for dipole modeling, using all sensors as is our typical clinical practice and as used in other studies (for example, see Pellegrino et al. (2018). These modeling choices may explain some of the differences compared to previous work, particularly in assessing accuracy compared to the distance to the resection.

Specifically, use of subsets of sensors may be more accurate in some cases, therefore localizing closer to the resection, while inappropriate channel selection may at times lead to mislocalization (Bagic et al. 2011). Similarly, deeper often white matter localizations permitted in the dipole localizations in our study may not be included in topectomies largely limited to gray matter and therefore may disadvantage dipole compared to dSPM localizations in these cases. Finally, our analyses were based solely on magnetic source imaging and did not include electrical source imaging (ESI). ESI has been shown to provide complementary localizing information to MSI alone, particularly when the earliest localizable source solution is observed in EEG (Plummer et al. 2019). However, overall localizations based on MSI alone tend to localize closer to the irritative zone using iEEG than ESI localizations (Tamilia et al. 2019), Finally, Plummer et al. suggest that independent modeling of MSI and ESI yield better results that combined EMSI. For this study, however, we focused on the use of only MSI to compare the results of dipole and distributed source modeling results. ESI could be used in the future to study propagation patterns in a similar fashion. Future work could also evaluate the efficacy of these approaches in subpopulations, such as temporal versus extratemporal lobe epilepsy, or focal versus broadly distributed IEDs.

Conclusion

Our findings suggest that dSPM is a useful tool to study areas of IED-related activation over time. Spatially discrete areas of activation suggestive of white matter propagation were observed in most patients and were already apparent at the earliest valid dipole localization. Increases in volume of activation over time did not appreciably change the locations of the primary cluster’s center of mass or overall center of mass across clusters. This stability likely accounts for the high concordance with the area of resection at each time point studied. In contrast, dipole localizations exhibited more movement over time and were further from the resection area at both timepoints. These findings suggest that distributed source modeling using dSPM may present advantages over single dipoles for localization of seizure foci, particularly in the setting of more widespread areas of activation, and that the use of multiple source localization methods is important to better understand brain networks activated during interictal epileptiform activity.

Supplementary Material

1

Highlights.

Activation area increases along interictal spike (IS) rising phases, often with early spread to spatially discrete brain regions.

Dipole sources localize near the center of mass (COM) of the largest IS distributed source-modeled activation cluster.

Activation cluster COMs are relatively stationary and overlap with surgical resections in seizure-free postoperative patients.

Funding

This research was supported by the Intramural Research Program of the NIH, NINDS, where the clinical research was carried out. The opinions expressed in this publication are those of the author(s) and do not necessarily reflect the views of the NIH.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Conflict of interest statement

None.

References

  1. Alarcon G, Garcia Seoane JJ, Binnie CD, Martin Miguel MC, Juler J, Polkey CE, et al. Origin and propagation of interictal discharges in the acute electrocorticogram. Implications for pathophysiology and surgical treatment of temporal lobe epilepsy. Brain 1997;120 (Pt 12):2259–82. 10.1093/brain/120.12.2259. [DOI] [PubMed] [Google Scholar]
  2. Bagić AI, Knowlton RC, Rose DF, Ebersole JS. American Clinical Magnetoencephalography Society Clinical Practice Guideline 1: Recording and Analysis of Spontaneous Cerebral Activity*. Journal of Clinical Neurophysiology 2011:1. 10.1097/WNP.0b013e3182272fed. [DOI] [PubMed] [Google Scholar]
  3. Cox RW. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res 1996;29:162–73. 10.1006/cbmr.1996.0014. [DOI] [PubMed] [Google Scholar]
  4. Cox RW, Hyde JS. Software tools for analysis and visualization of fMRI data. NMR Biomed 1997;10:171–8. 10.1002/(sici)1099-1492(199706/08)10:4/5<171::aid-nbm453>3.0.co;2-l. [DOI] [PubMed] [Google Scholar]
  5. de Curtis M, Avanzini G. Interictal spikes in focal epileptogenesis. Prog Neurobiol 2001;63:541–67. 10.1016/s0301-0082(00)00026-5. [DOI] [PubMed] [Google Scholar]
  6. de Curtis M, Jefferys JGR, Avoli M. Interictal Epileptiform Discharges in Partial Epilepsy: Complex Neurobiological Mechanisms Based on Experimental and Clinical Evidence. In: Noebels JL, Avoli M, Rogawski MA, Olsen RW, Delgado-Escueta AV, editors. Jasper’s Basic Mechanisms of the Epilepsies. 4th ed., Bethesda (MD): National Center for Biotechnology Information (US); 2012. [PubMed] [Google Scholar]
  7. Dale AM, Liu AK, Fischl BR, Buckner RL, Belliveau JW, Lewine JD, et al. Dynamic statistical parametric mapping: combining fMRI and MEG for high-resolution imaging of cortical activity. Neuron 2000;26:55–67. 10.1016/s0896-6273(00)81138-1. [DOI] [PubMed] [Google Scholar]
  8. Diamond JM, Withers CP, Chapeton JI, Rahman S, Inati SK, Zaghloul KA. Interictal discharges in the human brain are travelling waves arising from an epileptogenic source. Brain 2023;146:1903–15. 10.1093/brain/awad015. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Engel J Surgical Treatment of the Epilepsies. 2nd ed. Raven Press; 1993. [Google Scholar]
  10. Gramfort A, Luessi M, Larson E, Engemann DA, Strohmeier D, Brodbeck C, et al. MNE software for processing MEG and EEG data. NeuroImage 2014;86:446–60. 10.1016/j.neuroimage.2013.10.027. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Gramfort A, Luessi M, Larson E, Engemann DA, Strohmeier D, Brodbeck C, et al. MEG and EEG data analysis with MNE-Python. Front Neurosci 2013;7. 10.3389/fnins.2013.00267. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Kanamori Y, Shigeto H, Hironaga N, Hagiwara K, Uehara T, Chatani H, et al. Minimum norm estimates in MEG can delineate the onset of interictal epileptic discharges: A comparison with ECoG findings. Neuroimage Clin 2013;2:663–9. 10.1016/j.nicl.2013.04.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Knowlton R, Laxer K, Aminoff M, Roberts T, Wong S, Rowley H. Magnetoencephalography in partial epilepsy: Clinical yield and localization accuracy. Ann Neurol 1997;42:622–31. 10.1002/ana.410420413. [DOI] [PubMed] [Google Scholar]
  14. Kobayashi K, Yoshinaga H, Ohtsuka Y, Gotman J. Dipole modeling of epileptic spikes can be accurate or misleading. Epilepsia 2005;46:397–408. 10.1111/j.0013-9580.2005.31404.x. [DOI] [PubMed] [Google Scholar]
  15. Kural MA, Tankisi H, Duez L, Sejer Hansen V, Udupi A, Wennberg R, Rampp S, Larsson PG, Schulz R, Beniczky S. Optimized set of criteria for defining interictal epileptiform EEG discharges. Clin Neurophysiol. 2020. Sep;131(9):2250–2254. doi: 10.1016/j.clinph.2020.06.026. [DOI] [PubMed] [Google Scholar]
  16. Lantz G, Spinelli L, Seeck M, de Peralta Menendez RG, Sottas CC, Michel CM. Propagation of interictal epileptiform activity can lead to erroneous source localizations: a 128-channel EEG mapping study. J Clin Neurophysiol 2003;20:311–9. 10.1097/00004691-200309000-00003. [DOI] [PubMed] [Google Scholar]
  17. Matarrese MAG, Loppini A, Fabbri L, Tamilia E, Perry MS, Madsen JR, Bolton J, Stone SSD, Pearl PL, Filippi S, Papadelis C. Spike propagation mapping reveals effective connectivity and predicts surgical outcome in epilepsy. Brain. 2023. Sep 1;146(9):3898–3912. doi: 10.1093/brain/awad118. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Nolte G The magnetic lead field theorem in the quasi-static approximation and its use for magnetoencephalography forward calculation in realistic volume conductors. Phys Med Biol 2003;48:3637–52. 10.1088/0031-9155/48/22/002. [DOI] [PubMed] [Google Scholar]
  19. Oishi M, Kameyama S, Masuda H, Tohyama J, Kanazawa O, Sasagawa M, et al. Single and multiple clusters of magnetoencephalographic dipoles in neocortical epilepsy: significance in characterizing the epileptogenic zone. Epilepsia 2006;47:355–64. 10.1111/j.1528-1167.2006.00428.x. [DOI] [PubMed] [Google Scholar]
  20. Pellegrino G, Hedrich T, Chowdhury RA, Hall JA, Dubeau F, Lina J-M, et al. Clinical yield of magnetoencephalography distributed source imaging in epilepsy: A comparison with equivalent current dipole method. Hum Brain Mapp 2018;39:218–31. 10.1002/hbm.23837. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Plummer C, Vogrin SJ, Woods WP, Murphy MA, Cook MJ, Liley DTJ. Interictal and ictal source localization for epilepsy surgery using high-density EEG with MEG: a prospective long-term study. Brain 2019;142:932–51. 10.1093/brain/awz015. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Reinsberger C, Tanaka N, Cole AJ, Lee JW, Dworetzky BA, Bromfield EB, et al. Current dipole orientation and distribution of epileptiform activity correlates with cortical thinning in left mesiotemporal epilepsy. Neuroimage 2010;52:1238–42. 10.1016/j.neuroimage.2010.04.264. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Rossi LF, Wykes RC, Kullmann DM, Carandini M. Focal cortical seizures start as standing waves and propagate respecting homotopic connectivity. Nat Commun 2017;8:217. 10.1038/s41467-017-00159-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Ségonne F, Dale AM, Busa E, Glessner M, Salat D, Hahn HK, et al. A hybrid approach to the skull stripping problem in MRI. Neuroimage 2004;22:1060–75. 10.1016/j.neuroimage.2004.03.032. [DOI] [PubMed] [Google Scholar]
  25. Shiraishi H, Ahlfors SP, Stufflebeam SM, Knake S, Larsson PG, Hämäläinen MS, et al. Comparison of Three Methods for Localizing Interictal Epileptiform Discharges with Magnetoencephalography. J Clin Neurophysiol 2011;28:431–40. 10.1097/WNP.0b013e318231c86f. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Stefan H, Hummel C, Scheler G, Genow A, Druschky K, Tilz C, et al. Magnetic brain source imaging of focal epileptic activity: a synopsis of 455 cases. Brain 2003;126:2396–405. 10.1093/brain/awg239. [DOI] [PubMed] [Google Scholar]
  27. Tamilia E, Madsen J, Grant PE, Pearl PL, Papadelis C. Current and Emerging Potential of Magnetoencephalography in the Detection and Localization of High-Frequency Oscillations in Epilepsy. Front Neurol 2017;8. 10.3389/fneur.2017.00014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Tanaka N, Cole AJ, von Pechmann D, Wakeman DG, Hämäläinen MS, Liu H, et al. Dynamic statistical parametric mapping for analyzing ictal magnetoencephalographic spikes in patients with intractable frontal lobe epilepsy. Epilepsy Res 2009;85:279–86. 10.1016/j.eplepsyres.2009.03.023. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Tao JX, Ray A, Hawes-Ebersole S, Ebersole JS. Intracranial EEG substrates of scalp EEG interictal spikes. Epilepsia 2005;46:669–76. 10.1111/j.1528-1167.2005.11404.x. [DOI] [PubMed] [Google Scholar]
  30. Withers CP, Diamond JM, Yang B, Snyder K, Abdollahi S, Sarlls J, et al. Identifying sources of human interictal discharges with travelling wave and white matter propagation. Brain 2023;146:5168–81. 10.1093/brain/awad259. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS2107658-supplement-1.docx (507KB, docx)

RESOURCES