Resting-State SEEG May Help Localize Epileptogenic Brain Regions

Abstract

BACKGROUND

Stereotactic electroencephalography (SEEG) is a minimally invasive neurosurgical method to localize epileptogenic brain regions in epilepsy but requires days in the hospital with interventions to trigger several seizures.

OBJECTIVE

To make initial progress in the development of network analysis methods to identify epileptogenic brain regions using brief, resting-state SEEG data segments, without requiring seizure recordings.

METHODS

In a cohort of 15 adult focal epilepsy patients undergoing SEEG, we evaluated functional connectivity (alpha-band imaginary coherence) across sampled regions using brief (2 min) resting-state data segments. Bootstrapped logistic regression was used to generate a model to predict epileptogenicity of individual regions.

RESULTS

Compared to nonepileptogenic structures, we found increased functional connectivity within epileptogenic regions (P < .05) and between epileptogenic areas and other structures (P < .01, paired t-tests, corrected). Epileptogenic areas also demonstrated higher clustering coefficient (P < .01) and betweenness centrality (P < .01), and greater decay of functional connectivity with distance (P < .05, paired t-tests, corrected). Our functional connectivity model to predict epileptogenicity of individual regions demonstrated an area under the curve of 0.78 and accuracy of 80.4%.

CONCLUSION

Our study represents a preliminary step towards defining resting-state SEEG functional connectivity patterns to help localize epileptogenic brain regions ahead of neurosurgical treatment without requiring seizure recordings.

ABBREVIATIONS

AUC

area under the curve

EMU

epilepsy monitoring unit

MEG

magnetoencephalography

RNS

responsive neurostimulation system

ROC

receiver operator characteristic

SD

standard deviation

SEEG

stereotactic electroencephalography

Successful epilepsy surgery depends critically on accurate localization of epileptogenic brain regions for targeted therapies.^1,2 Stereotactic electroencephalography (SEEG) is a safe, minimally invasive neurosurgical approach which is increasingly being used to obtain intracranial recordings.^3-5 SEEG was first introduced in France over 60 yr ago but has only gained popularity outside of Europe in recent years.^6,7 In this procedure, several depth electrodes are placed using stereotactic techniques without a craniotomy.^8,9 However, localization of epileptogenic regions using SEEG still requires an inpatient hospital admission of many days to weeks for video-EEG monitoring. Various interventions are utilized to trigger and capture several seizures, and adverse events may occur.¹⁰ Alternatively, the ability to localize epileptogenic brain regions using brief epochs of resting-state SEEG recordings could potentially reduce hospitalization, eliminate the need to provoke seizures, reduce medical costs, and improve patient care.

In a previous magnetoencephalography (MEG) study of focal epilepsy patients, it was found that high resting-state functional connectivity (estimated by imaginary coherence in the alpha frequency band) predicted epileptogenicity in the region of surgical resection.¹¹ However, MEG spatial resolution and accuracy are limited in deep brain structures that can easily be sampled with SEEG.^12,13 In the present study, we sought to determine whether network analysis using alpha-band imaginary coherence measurements from resting-state SEEG can help distinguish epileptogenic from nonepileptogenic brain regions in focal epilepsy patients. Our ultimate goal is to develop an electrographic functional connectivity biomarker to identify epileptogenic structures for surgical treatment without necessitating prolonged inpatient ictal recordings.

METHODS

Subjects

This cohort study included 16 consecutive patients with medically refractory epilepsy who underwent SEEG and video-EEG monitoring at our institution in 2017 or 2018. One patient did not have any seizures during the hospitalization and was excluded from the study. Demographics and disease characteristics of the remaining 15 patients are summarized in the Table. For patients who underwent subsequent surgery for treatment, postoperative seizure outcome was determined at last follow-up (>1 yr in all but one patient). This study and all procedures were approved by the Vanderbilt University Medical Center Institutional Review Board, and written informed consent was obtained from all patients.

TABLE.

Summary of Patient Characteristics

Demographics
Age, years	35.2 ± 11.4
Gender, female	11 (73.3)
Handedness, right	9 (60.0)
Disease information
Epilepsy duration, years	21.1 ± 15.5
Seizure frequency, monthly
Focal with spared consciousness	6.7 ± 15.0
Focal with impaired consciousness	7.2 ± 5.9
Focal to bilateral tonic-clonic	0.4 ± 0.5
Seizure onset region
Mesial temporal, unilateral	6 (40.0)
Mesial temporal, bilateral	6 (40.0)
Focal neocortical	3 (20.0)
Details of recordings
Days recorded	9.9 ± 4.3
Electrodes implanted	9.3 ± 1.8
Regions sampled	13.5 ± 3.1

N = 15. Data are mean ± SD for continuous variables or N (%) for counts.

SEEG Data Collection

Intracranial electrodes (Ad-Tech, Oak Creek, Wisconsin) were implanted after using CRAnialVault Explorer (CRAVE; Vanderbilt University, Nashville, Tennessee) to plan their anatomic trajectories.¹⁴ Patients in our study had 9.3 ± 1.8 (mean ± standard deviation [SD]) electrode leads implanted (0.86-mm diameter) with 10 contacts per lead. SEEG recordings were collected at a 1-kHz sampling rate in the epilepsy monitoring unit (EMU) using a JE-209 clinical EEG system (Nihon Kohden America, Irvine, California). Twenty minutes of interictal data during the “resting-state” (lying awake with eyes closed, as defined previously¹⁵) were collected one full day after admission to EMU. Raw resting-state SEEG data were preprocessed (bandpass filter, 1-119 Hz; notch filter, 60 Hz) in EEGLab (https://sccn.ucsd.edu/eeglab/index.php) and FieldTrip (http://www.fieldtriptoolbox.org/). The anatomical region associated with each electrode contact was determined using each patient's preoperative magnetic resonance imaging and postoperative computed tomography coregistered in CRAVE, with brain regions defined using the Harvard-Oxford Atlas (Harvard Center of Morphometric Analysis, Cambridge, Massachusetts). Data were re-referenced using bipolar montage, and electrode pairs completely in white matter were excluded from analysis. A clean data segment of 120 s, which was free of interictal spike activity or artifact was selected for each patient.

Each anatomical region sampled in each patient was defined as “epileptogenic” or “nonepileptogenic” using traditional clinical interpretation by the treating epileptologist. This determination occurred at the end of the full SEEG recording session, and was completed prior to any functional connectivity analyses. The “epileptogenic zone” was defined using the criteria first proposed by Lüders and colleagues¹⁶ in 1993, which includes the “area of cortex that is necessary and sufficient for initiating seizures and whose removal is necessary for complete abolition of seizures.” Consistent with Lüders and colleagues,¹⁷ epileptogenic areas included the “seizure onset zone” and also the “potential seizure outset zone,” which may include areas of seemingly simultaneous seizure onset and those with very frequent interictal spike activity. Specifically, some of the electrographic patterns the epileptologists used to help localize epileptogenic regions included but were not limited to high-frequency oscillations, rhythmic temporal theta, rhythmic alpha, spikes, and spike and wave discharges. Overall, the goal of our investigation was to dichotomize regions into those representing potential surgical targets vs those that were not potential surgical targets, in order to compare functional connectivity analysis with the traditional clinical EEG interpretation.

Functional Connectivity Measurements

Functional connectivity was estimated using imaginary coherence in the alpha-band (8-12 Hz), which ignores zero-time lag signals and minimizes artifact and volume conduction effects.^18,19 Furthermore, imaginary coherence in the alpha-band has the highest test-retest reliability, likely due to the resting-state alpha EEG peak, and has been well established as a measure of functional connectivity^11,20 including in intracranial EEG studies.¹⁹ For further comparison, we also utilized individualized alpha-bands based on the alpha peak seen on the 1/f power spectrum plot for each patient, and we compared these results to the traditional 8 to 12 Hz alpha-band. Imaginary coherence in the delta (1-4 Hz) and theta (4-8 Hz) bands was also measured. Coherence was calculated using the multitaper method by Thomson²¹ with time-frequency bandwidth of 3 (unitless), 1-s window, and 5 tapers. The coherency analysis outputs an electrode contact-by-contact imaginary coherence functional connectivity matrix. These values were averaged into an anatomical region-by-region matrix for further analyses. “Between connectivity” was defined as the mean functional connectivity between all electrode pairs within a region with all other regions sampled across the brain, whereas “within connectivity” was defined as the mean functional connectivity of only those electrode contact pairs within a particular region to other electrode pairs in that same region. Imaginary coherence values across regions within each patient were normalized to a standard z-score for comparison across patients.

We also examined the relationship between functional connectivity and distance between electrode contact pairs. To facilitate this analysis, we extracted each electrode contact position in 3D space using CRAVE and created a matrix of the Euclidean distance between all electrode contact pairs. It was noted that the relationship between functional connectivity and distance was typically negative, but it was often not linear. Therefore, using MATLAB (version 2017a; MathWorks Inc, Natick, Massachusetts), an exponential fit to each data plot (functional connectivity * distance) for each electrode contact pair was calculated as:

(1)

where c represents functional connectivity, e is a natural constant, λ represents the decay constant, d is distance, and y represents the value at which the exponential fit crosses the y-axis. The average decay constant for all electrode pairs in each individual region was then calculated.

To further analyze functional connectivity between brain regions, we utilized 2 centrality measures (nodal betweenness centrality and edge betweenness centrality) and the clustering coefficient. Betweenness centrality is defined as the fraction of all shortest paths in the network that contain a given node (nodal) or connection between nodes (edge), whereas clustering coefficient reflects the degree to which nodes tend to cluster together.²² These analyses were performed using an undirected contact-by-contact imaginary coherence association matrix. These 3 measures were calculated using the Brain Connectivity Toolbox (https://sites.google.com/site/bctnet/).²² As above, functional connectivity results within patients were converted to a standard z-score for comparison across patients. For visualization of betweenness centrality results, we utilized BrainNet Viewer (http://www.nitrc.org/projects/bnv/).²³

Statistical Analyses

The Anderson-Darling test for normal distribution was employed before applying parametric tests. In epileptogenic vs nonepileptogenic regions, 2-tailed paired t-tests were used to compare functional connectivity measures and the functional connectivity-distance decay constant. The mean functional connectivity across epileptogenic regions and across nonepileptogenic was calculated within individual patients (n = 15) prior to statistical testing. To predict epileptogenicity of individual regions, we generated a binary classification logistic regression model incorporating 6 variables (between connectivity, within connectivity, distance-connectivity decay constant, clustering coefficient, nodal betweenness centrality, and edge betweenness centrality), using bootstrapping methods to train the model with subsamples of data. Receiver operator characteristic (ROC) curves were generated to evaluate the performance of the model in predicting epileptogenicity, and evaluate the area under the curve (AUC). For patients with postoperative outcome data, functional connectivity in the presumed epileptogenic regions was compared in patients with favorable vs unfavorable seizure outcome using chi square. Statistical analyses were performed using MATLAB and SPSS 23 (Armonk, New York), with significance assessed at P < .05. Tests were corrected for multiple comparisons using the Bonferroni-Holm method, where appropriate.

RESULTS

Patient and Data Characteristics

SEEG and video data for all 15 patients, which were collected over 9.9 ± 4.3 d (mean ± SD), were reviewed and interpreted by the treating epileptologist prior to functional connectivity analyses. Overall, 13.5 ± 3.1 brain regions were sampled per patient, and 20.0% of regions across all patients were epileptogenic, whereas 80.0% were determined to be nonepileptogenic. There was no difference in the number of electrode contact pairs in epileptogenic regions (4.7 ± 1.7) vs nonepileptogenic regions (4.5 ± 0.9) across patients (P = .7, paired t-test). Mesial temporal lobe epilepsy was identified in 12 patients (6 patients with bilateral epileptogenicity), whereas 3 individuals had focal neocortical epilepsy. Patient characteristics are summarized in the Table.

Overall Increased Functional Connectivity in Epileptogenic Regions

We noted that epileptogenic regions often demonstrated higher functional connectivity with other regions (“between connectivity”) and functional connectivity within the region (“within connectivity”) in the alpha, theta, and delta frequency bands, as demonstrated in an example patient (Figure 1A and 1B). Across all 15 patients, between connectivity (Figure 1C) and within connectivity (Figure 1D) were higher in epileptogenic compared to nonepileptogenic brain regions. As the largest differences in functional connectivity were observed in the alpha-band, we utilized alpha-band imaginary coherence measurements for the remainder of our analyses. Furthermore, we repeated this analysis utilizing individualized alpha-bands for each patient (using power spectra to identify the individual alpha peak), to ensure that possible alpha slowing in epilepsy patients was not influencing our results. We observed very similar results using either 8 to 12 Hz alpha-band or individualized alpha-band (Figure, Supplemental Digital Content). Of note, the majority of patients in this study had mesial temporal lobe epilepsy; however, no differences in alpha-, theta-, or delta-band imaginary coherence were noted between nonepileptogenic mesial temporal lobe structures vs nonepileptogenic neocortical regions (P = .47−.66, unpaired t-tests, uncorrected), suggesting that our functional connectivity findings were not driven by anatomic location alone.

FIGURE 1.

Epileptogenic regions exhibit high overall functional connectivity. A and B, Connectivity patterns in an example patient, with “between connectivity” A representing mean imaginary coherence of all electrode contacts in a region to all other regions sampled and “within connectivity” B signifying imaginary coherence between individual electrode contacts within each region. In this patient, seizures originated in the left precentral gyrus (†), and this region demonstrates high between brain and within region functional connectivity in the alpha, theta, and delta frequency bands. C and D, Across all patients, between brain functional connectivity and within region functional connectivity are higher in epileptogenic vs nonepileptogenic regions, with statistical comparison performed across patients ***P < .01, paired t-test, Bonferroni-Holm correction. ctx, cortex; gyr, gyrus; inf, inferior; L, left; lat, lateral; lob, lobule; R, right; sup, superior; suppl, supplementary.

Relationship Between Functional Connectivity and Distance

Next, we sought to understand the relationship between functional connectivity and distance in epileptogenic vs nonepileptogenic regions. Examining functional connectivity seeded from electrode contact pairs located within epileptogenic regions, we often noted a negative nonlinear relationship between functional connectivity and distance, with high short-range functional connectivity, and rapid decay of long-range functional connectivity (example in Figure 2A). This trend was observed less often in nonepileptogenic regions (example in Figure 2B). Evaluating summary data across the 15 patients, a larger decay constant of the relationship between distance and functional connectivity was found in epileptogenic compared to nonepileptogenic regions (Figure 2C). Overall, this suggests larger increases in short-range functional connectivity than long-range functional connectivity in epileptogenic regions.

FIGURE 2.

Epileptogenic regions demonstrate greater decay of functional connectivity over distance compared to long-range functional connectivity. A, Example distance vs functional connectivity plot seeded from an electrode contact pair in an epileptogenic region of a single patient, with data points demonstrating alpha-band imaginary coherence to other electrode contact pairs. High functional connectivity to nearby contact pairs is observed, with rapid decay of functional connectivity to more distant contact pairs, as demonstrated by the line of exponential fit. B, Example distance vs functional connectivity plot seeded from an electrode contact pair in a nonepileptogenic region of the same patient, with little decay of functional connectivity over distance, as demonstrated by the line of exponential fit. C, In examining the relationship between distance and functional connectivity across all patients, there is a larger decay constant (λ; see eq. 1, Methods) in epileptogenic compared to nonepileptogenic regions, with statistical comparison performed across patients (n = 15). The boxplot represents the median, upper quartile, and lower quartile, whereas error bars demonstrate extreme data points. *P < .05, paired t-test.

Altered Network Properties in Epileptogenic Regions

We then asked whether epileptogenic regions and their connections might possess different network properties using graph theory. As noted in an example patient (Figure 3A), and in summary data across all patients (Figure 3B), epileptogenic areas demonstrated higher clustering coefficient, nodal betweenness centrality, and edge betweenness centrality than nonepileptogenic regions. As shown in network maps of 2 example patients with mesial temporal lobe epilepsy and unilateral (Figure 4A) or bilateral (Figure 4B) electrode coverage, epileptogenic regions appeared to show greater connectedness than nonepileptogenic regions. In particular, epileptogenic hippocampi showed strong functional connectivity to the ipsilateral amygdala and other limbic structures.

FIGURE 3.

Epileptogenic regions exhibit increased clustering and betweenness centrality. A, Clustering coefficient, nodal betweenness centrality, and edge betweenness centrality values in an example patient measured using alpha-band imaginary coherence data. In this patient, seizures originated from the left hippocampus and left amygdala, and these regions (^†) demonstrated high functional connectivity in all measures. B, Across all patients (n = 15), functional connectivity is higher in epileptogenic regions vs nonepileptogenic regions using each measure. The boxplot represents the median, upper quartile, and lower quartile, whereas error bars demonstrate extreme data points **P < .05, ***P < .01, paired t-test, Bonferroni-Holm correction. ant, anterior; ctx, cortex; gyr, gyrus; inf, inferior; L, left; lat, lateral; mid, middle; post, posterior; R, right; sup, superior; suppl, supplementary; temp, temporal.

FIGURE 4.

Betweenness centrality maps in example patients demonstrating functional connectivity of epileptogenic regions. Patients are A 30-yr-old left-handed female with left mesial temporal lobe epilepsy and seizures for 3 yr with unilateral electrode placement, and B a 30-yr-old left-handed female with right mesial temporal lobe epilepsy and seizures since infancy with bilateral electrode placement. Overall, epileptogenic regions show relatively high nodal betweenness centrality (size of sphere) and high-edge betweenness centrality (thickness of line connecting 2 spheres) with strong connections to other structures. Edge connections below an arbitrary threshold (2) are not shown for simplicity. Alpha-band imaginary coherence is utilized for all measures. Images were created using BrainNet Viewer.²³ Ant, anterior; ctx, cortex; gyr, gyrus; inf, inferior; L, left; lat, lateral; lob, lobule; mid, middle; Post, posterior; R, right; sup, superior; suppl, supplementary.

Predicting Regional Epileptogenicity Using Functional Connectivity Modeling

Finally, our goal was to create a prediction model that might help identify epileptogenic regions in an individual patient ahead of definitive neurosurgical treatment. We therefore utilized a logistic regression model incorporating the 6 functional connectivity measures examined above. Comparing ROC curves of this model to the individual functional connectivity measures (Figure 5), an AUC of 0.78 was noted for the model, compared to an AUC range of 0.64 to 0.75 for the individual measures (with the connectivity-distance decay constant demonstrating the highest AUC [0.75] of any individual measure). This suggests that the model performs slightly better than any individual measure in predicting epileptogenicity of a brain region. Overall, the accuracy of the model was 80.4%, and at a cutoff value (0.14) corresponding to maximum specificity plus sensitivity, the model demonstrated a sensitivity of 82.5% and specificity of 60.4% in correctly predicting epileptogenic vs nonepileptogenic brain regions.

FIGURE 5.

Receiver operator characteristic (ROC) curves of epileptogenicity predictors. The ROC curves demonstrate sensitivity and 1 (specificity) of 6 individual functional connectivity measures (from Figures 1–3) in predicting epileptogenic vs nonepileptogenic brain regions across all 15 patients. The model resulting from binary logistic regression analysis, which incorporates all 6 functional connectivity measures, is also shown (solid red line). The area under the curve (AUC) across the 6 individual measures ranged from 0.64 to 0.75, whereas the model demonstrated an AUC of 0.78 and overall accuracy of 80.4%. At the cutoff value (0.14) corresponding to maximum sensitivity plus specificity (dashed red line), the model demonstrated a sensitivity of 82.5% and specificity of 60.4% in predicting epileptogenic vs nonepileptogenic brain regions. All functional connectivity measures shown utilized alpha-band imaginary coherence measurements.

Postoperative Seizure Outcomes

After SEEG monitoring, 4 (26.7%) of 15 patients with bilateral temporal lobe (3) or eloquent motor cortex (1) epilepsy either declined responsive neurostimulation system (RNS) or are still being considered for RNS. Eleven (73.3%) patients underwent surgery and had postoperative follow-up of 12.8 ± 2.0 mo (mean ± SD; range 10-17), with one patient having less than 1 yr postoperative follow-up. Surgeries included selective amygdalohippocampectomy in 4 patients, temporal lobectomy in 3 individuals, bilateral temporal lobe RNS in 2 patients, insular resection in 1 patient, and midfrontal lobe resection in 1 individual. At last follow-up, 6 (66.7%) patients who underwent resection had a favorable seizure outcome with seizure freedom, whereas 3 (33.3%) had an unfavorable seizure outcome with persistent seizures, and the 2 patients who underwent RNS experienced >80% decrease in seizure frequency, which was deemed a favorable response (>50%) to this treatment. Given the small number of patients receiving surgery, the heterogeneous procedures performed, and short follow-up duration, we did not perform detailed analysis to relate functional connectivity patterns to seizure outcome. Of note, however, whereas 8 (88.9%) of 9 patients with high (Z > 0) overall functional connectivity (alpha-band imaginary coherence) at their suspected epileptogenic regions had a favorable seizure outcome, 0 (0%) of 2 patients with low (Z < 0) overall functional connectivity at their suspected epileptogenic regions had a favorable seizure outcome, suggesting a possible moderate relationship between high epileptogenic zone functional connectivity and favorable outcome (chi square = 6.5, P = .055).

DISCUSSION

Summary of Functional Connectivity in Epileptogenic vs Nonepileptogenic Regions

In the present study, we observed higher functional connectivity in epileptogenic regions than nonepileptogenic areas in 15 focal epilepsy patients, using alpha-band imaginary coherence measurements in resting-state SEEG data. Compared to nonepileptogenic structures, epileptogenic regions demonstrated increased overall functional connectivity to other brain regions sampled, increased functional connectivity within the region, higher betweenness centrality, and a larger clustering coefficient. Using various functional connectivity measures, we also created a model to help predict epileptogenicity in individual brain regions, which demonstrated an AUC of 0.78 and overall accuracy of 80.4%. The predictive value of this model was somewhat higher than any individual functional connectivity measure alone. At a cutoff value that maximized specificity plus sensitivity, the model demonstrated a sensitivity of 82.5% and specificity of 60.4% in predicting epileptogenicity. These results suggest that our current methods remain insufficient to replace traditional SEEG interpretation, which may ultimately require an accuracy of more than 90% corroborated by postoperative seizure outcomes. However, our findings do demonstrate an important proof of principle: connectivity analysis utilizing brief resting-state recording epochs may contribute useful localizing information ahead of neurosurgical treatment, even in the absence of ictal or interictal epileptiform activity. Once improved, functional connectivity analysis may be appropriate to consider in the clinical setting.

Increased Short-Range Functional Connectivity With Rapid Decay in Epileptogenic Regions

We also observed that epileptogenic regions demonstrated increased short-range functional connectivity, but that functional connectivity to other regions decreased with distance. Overall, the functional connectivity-distance decay constant was the single best predictor of epileptogenicity (AUC = 0.75) out of all the individual measures we studied. We also specifically observed high functional connectivity between the hippocampus and amygdala in mesial temporal lobe epilepsy patients. Several other resting-state intracranial EEG studies have reported increased network functional connectivity within epileptic networks that dissipates with distance.²⁴ For instance, investigations of mesial temporal lobe epilepsy patients have uncovered regional increases in limbic network functional connectivity that resemble our findings.^25,26 However, increased regional functional connectivity may not be specific to mesial temporal lobe epilepsy, as one SEEG study of neocortical epilepsy patients with focal cortical dysplasia similarly noted preferential coupling between structures within the epileptogenic network and a gradual decrease in functional connectivity distal to the epileptogenic.²⁷ In other studies of neocortical epilepsy, examination of slow oscillatory intracranial EEG signals has suggested increased functional connectivity within seizure propagation networks, with functional isolation of epileptogenic regions.^28,29 Higher synchronization likelihood and clustering index were also observed in another study of both neocortical and mesial temporal lobe epilepsy patients.³⁰ Furthermore, an SEEG study of patients with focal cortical dysplasia also found increased synchronization at electrode contacts near epileptogenic zones, particularly in higher frequencies.³¹ Overall, human intracranial EEG studies to date suggest increased local functional connectivity related to epileptogenic networks, with fewer changes in distal functional connectivity.

The Role of SEEG Functional Connectivity Analysis in Surgical Decision Making

Although we did observe that patients with high overall functional connectivity in epileptogenic regions were somewhat more likely to achieve a favorable seizure outcome than those with low functional connectivity, this was a preliminary finding in only a subset of patients with insufficient postoperative follow-up duration. Future studies of a larger cohort with long-term outcomes should relate detailed functional connectivity patterns to postoperative outcome, to determine whether network analysis may ultimately be used as a tool to improve outcomes in epilepsy surgery. Other groups have related intracranial EEG functional connectivity patterns to seizure outcomes after epilepsy surgery. For instance, one group described worse seizure outcomes after resection in patients with a greater number of high-betweenness centrality nodes.³² Although these nodes were interpreted as being potentially protective against seizures, explaining poor outcomes after their resection, it is also possible that patients with a greater number of highly connected nodes are more likely to harbor multiple epileptogenic regions. Other groups have also used intracranial EEG measures of global synchrony,³³ nodal transition time,³⁴ or phase amplitude coupling³⁵ to help predict seizure outcome. Regardless of analysis method, increased connectedness of the area of resection appears to be associated with more favorable seizure outcome.

Study Limitations

There are several other limitations of this study to consider. Although our functional connectivity results were relatively consistent across the 15 patients examined, this is a preliminary investigation in a small number of patients. We also did not differentiate other important parts of the epilepsy network, such as irritative zone or propagation regions that are not part of the epileptogenic zone, and there will be value to examining these in future studies. Next, we collected SEEG data the day after implantation because medication wean and seizures have typically not yet begun at that point, but data collection at a later time might decrease the risk of network disruption from implantation. Furthermore, our analyses utilized only a single connectivity metric (imaginary coherence). It may be useful to incorporate other network analysis metrics (eg, coherence, phase amplitude coupling) in future studies.

CONCLUSION

By measuring alpha-band imaginary coherence analysis in resting-state SEEG data, we found increased functional connectivity in epileptogenic brain regions compared to nonepileptogenic areas in focal epilepsy patients. Epileptogenic areas demonstrate higher short-range functional connectivity, betweenness centrality, and clustering coefficients, which may be used to help localize these areas for targeted surgical treatment. Optimizing network analysis methods for localization of epileptogenic structures without necessitating long-term recordings to capture several seizures may ultimately aid the identification of neurosurgical targets in focal epilepsy surgery.

Disclosures

The authors have no personal, financial, or institutional interest in any of the drugs, materials, or devices described in this article. Dr Englot has received one consulting honorarium from Medtronic. This work was supported in part by the National Institutes of Health awards R01 NS112252 (to Dr Englot), R00 NS097618 (to Dr Englot), R01 NS075270 (to Dr Morgan), T32 EB021937 (to Mr González), T32 GM07347 (to Mr González), KL2 TR002539 (to Dr Rolston), and R01 NS095291 (to Dr Dawant) and by the Vanderbilt Institute for Surgery and Engineering (VISE).

Supplemental Digital Content. Figure. Epileptogenic regions exhibit high overall functional connectivity using either standard or individualized alpha peaks. Connectivity patterns in an example patient, with “between connectivity” A representing mean imaginary coherence of all electrode contacts in a region to all other regions sampled and “within connectivity” B signifying imaginary coherence between individual electrode contacts within each region. In this patient, seizures originated in the left precentral gyrus (†), and this region demonstrates high between brain and within region functional connectivity, using either alpha-band calculation. C and D, Across all patients, between brain functional connectivity and within region functional connectivity are higher in epileptogenic vs nonepileptogenic regions using either alpha-band calculation, with statistical comparison performed across patients (n = 15). The boxplot represents the median, upper quartile, and lower quartile, whereas error bars demonstrate extreme data points. **P < .05, ***P < .01, paired t-test, Bonferroni-Holm correction. ctx, cortex; gyr, gyrus; inf, inferior; L, left; lat, lateral; lob, lobule; R, right; sup, superior; suppl, supplementary.