Source-sink connectivity: A novel interictal EEG marker for seizure localization

Abstract

Over 15 million epilepsy patients worldwide have medically refractory epilepsy (MRE), i.e., they do not respond to anti-epileptic drugs. Successful surgery is a hopeful alternative for seizure freedom but can only be achieved through complete resection or disconnection of the epileptogenic zone (EZ), the brain region(s) where seizures originate. Unfortunately, surgical success rates vary between 30%-70% because no clinically validated biological markers of the EZ exist. Localizing the EZ has thus become a costly and time-consuming process during which a team of clinicians obtain non-invasive neuroimaging data, which is often followed by invasive monitoring involving days-to-weeks of EEG recordings captured intracranially (iEEG). Clinicians visually inspect iEEG data, looking for abnormal activity (e.g., low-voltage high frequency activity) on individual channels occurring immediately before seizures. They also look for abnormal spikes that occur on iEEG between seizures ("resting-state"). In the end, clinicians use <1% of the iEEG data captured to assist in EZ localization (minutes of seizure data versus days of recordings), missing opportunities to leverage these largely ignored data to better diagnose and treat patients. Intracranial EEG offers a unique opportunity to observe rich epileptic cortical network dynamics but waiting for seizures to occur increases patient risks associated invasive monitoring. In this study, we aim to leverage iEEG data in between seizures by developing a new networked-based resting-state iEEG marker of the EZ. We hypothesize that when a patient is not seizing, it is because the EZ is inhibited by neighboring nodes. We develop an algorithm that identifies two groups of nodes from the resting-state iEEG network: those that are continuously inhibiting a set of neighboring nodes ("sources") and the inhibited nodes themselves ("sinks"). Specifically, patient-specific dynamical network models (DNMs) are estimated from minutes of iEEG and their connectivity properties reveal top sources and sinks in the network, with each node being quantified by a source-sink index (SSI). We validate the SSI index in a retrospective analysis of 65 patients by using the SSI of the annotated EZ as a metric to predict surgical outcomes. SSI predicts with an accuracy of 79%, compared to the accuracy of clinicians being 43% (successful outcomes). In failed outcomes, we identify regions of the brain with high SSIs that were untreated. When compared to high frequency oscillations, the most common resting-state iEEG feature proposed for EZ localization, SSI outperformed in predictive power (by a factor of 1.2) suggesting SSI as a resting-state EEG fingerprint of the EZ.

A Taxonomy of Seizure Spread Patterns, Speed of Spread, and Associations With Structural Connectivity

“…Other regions can have quick spread but low connectivity strengths because spread may come from a third node activating both regions. Spread time between meso-scale regions may be better predicted by the interaction between seizure generating regions and regulatory/inhibitory regions from models that incorporate these interactions such as diffusion models, source sink models 28 , push pull network models 29 , Epileptor 30 , and others.…”

“…Spread time between meso-scale regions may be better predicted by the interaction between seizure generating regions and regulatory/inhibitory regions from models that incorporate these interactions such as diffusion models, source sink models 28 , push pull network models 29 , Epileptor 30 , and others. We propose a naming of each cluster based on the the timing of activity of each region averaged across the seizures in each We hypothesize that Cluster 1 represents a focal, or localized, spread pattern because the average of all patients within that cluster does not indicate an early activation time of any one region.…”

Diagnosing Epilepsy with Normal Interictal EEG Using Dynamic Network Models

“…Both methods were validated across multiple epilepsy centers and developed in the Sarma lab. 24,25,32…”

“…Two network-based metrics, fragility 24 and source-sink 25 , have been proposed to aid in the localization of the epileptogenic zone from ictal and interictal intracranial EEG recordings, respectively. Both methods operate on the assumption that there are alterations in the neural network, specifically in the epileptic tissue, of an epilepsy patient.…”

“…EpiScalp aims to diagnose epilepsy from a patient‘s scalp EEG by i) estimating patient-specific dynamical network models from the EEG recording 24,25 and ii) analyzing the network properties to detect whether pathological patterns, inherent to an epileptic brain, are present in the EEG network.…”