Novel noninvasive identification of patient-specific epileptic networks in focal epilepsies: Linking single-photon emission computed tomography perfusion during seizures with resting-state magnetoencephalography dynamics

Abstract

Single-photon emission computed tomography (SPECT) during seizures and magnetoencephalography (MEG) during the interictal state are noninvasive modalities employed in the localization of the epileptogenic zone in patients with drug-resistant focal epilepsy (DRFE). The present study aims to investigate whether there exists a preferentially high MEG functional connectivity (FC) among those regions of the brain that exhibit hyperperfusion or hypoperfusion during seizures. We studied MEG and SPECT data in 30 consecutive DRFE patients who had resective epilepsy surgery. We parcellated each ictal perfusion map into 200 regions of interest (ROIs) and generated ROI time series using source modeling of MEG data. FC between ROIs was quantified using coherence and phase-locking value. We defined a generalized linear model to relate the connectivity of each ROI, ictal perfusion z score, and distance between ROIs. We compared the coefficients relating perfusion z score to FC of each ROI and estimated the connectivity within and between resected and unresected ROIs. We found that perfusion z scores were strongly correlated with the FC of hyper-, and separately, hypoperfused ROIs across patients. High interictal connectivity was observed between hyperperfused brain regions inside and outside the resected area. High connectivity was also observed between regions of ictal hypoperfusion. Importantly, the ictally hypoperfused regions had a low interictal connectivity to regions that became hyperperfused during seizures. We conclude that brain regions exhibiting hyperperfusion during seizures highlight a preferentially connected interictal network, whereas regions of ictal hypoperfusion highlight a separate, discrete and interconnected, interictal network.

Keywords: MEG; SPECT; cerebral blood perfusion; epilepsy; functional connectivity.

FIGURE 1

Pipeline used for comparing ictal single‐photon emission computed tomography (SPECT) and magnetoencephalography (MEG) functional connectivity. We first perform cortical reconstruction of the patient's magnetic resonance imaging (MRI) using Freesurfer. We then co‐register the ictal SPECT data to the MRI and parcellate the resulting SPECT data. Next, we import the temporally extended signal space separation (tSSS)‐processed MEG time series, the reconstructed cortical surfaces, and the regions of interest (ROIs) generated from SPECT data to Brainstorm. We perform source modeling analysis of the MEG data and generate a time series per ROI. We then use ICA decomposition to remove any residual EKG artifacts that were not rejected during tSSS processing. Next, we perform connectivity analysis of the resulting time series using coherence and phase locking value. Finally, we perform statistical analysis of the ROI connectivity and SPECT perfusion values.

FIGURE 2

(a) Ictal single‐photon emission computed tomography (SPECT) pattern of hyperperfusion (red blobs) and hypoperfusion (blue blobs) for Patient P05. (b) Parcellation of the cortical surface using subtraction ictal SPECT co‐registered to MRI (SISCOM) z value. (c, d) Relationship between SISCOM perfusion z value and phase‐locking value for: (c) a hyperperfused node and (d) a hypoperfused node. (e) Relationship between SISCOM z value and ${\beta }_1$‐coefficient for Patient P05. Each grey shaded circle represents 1 out of 200 regions of interest (ROIs). ROIs with statistically significant ${\beta }_1$‐coefficients are denoted using green borders.

FIGURE 3

Schematic visualization of the hyperperfused regions of interest (ROIs) that were resected (in red), hyperperfused ROIs that were not resected (in orange), and hypoperfused ROIs (in blue). The dashed black curve denotes the surgical resection margin.

FIGURE 4

Distribution of the cross‐correlation coefficient between ${\beta }_1$‐coefficients and subtraction ictal SPECT co‐registered to MRI (SISCOM) z values across patients and frequency bands when estimating connectivity using (a) phase‐locking value, and (b) coherence. Asterisks denote statistically significant (p < .001) and positive correlation between ${\beta }_1$ distribution and SISCOM z score across patients.

FIGURE 5

Distribution of ${\beta }_1$‐coefficients for the (a) hypoperfused regions of interest (ROIs), (b) baseline perfused ROIs—that is, ROIs that showed no significant change in perfusion during the seizure, and (c) hyperperfused ROIs. Single (p < .05), double (p < .01) or triple (p < .001) asterisks denote a statistically significant difference between the fractions of ROIs that have negative versus positive ${\beta }_1$‐coefficients. Phase‐locking value (PLV) was used as the connectivity estimator.

FIGURE 6

Distribution of phase‐locking value (PLV)‐based intrinsic and extrinsic connectivity for the five frequency bands across all patients. Statistically significant differences between connectivity estimates are denoted with single (p < .05), double (p < .01), and triple (p < .001) asterisks.

FIGURE 7

Distribution of coherence‐based intrinsic and extrinsic connectivity for the five frequency bands across all patients. Statistically significant differences between connectivity estimates are denoted with single (p < .05), double (p < .01), and triple (p < .001) asterisks.