Abstracts

Rationale: Intracranial EEG (iEEG) is often performed in patients with drug-resistant epilepsy (DRE) to estimate the seizure onset zone (SOZ) and tailor epilepsy surgery. While ictal iEEG data are not always captured and can require long recording periods, interictal data are largely available immediately. Interictal iEEG data are typically reviewed to identify traditional biomarkers, such as epileptic spikes, though these are not specific to the area that generates seizures and thus provide limited presurgical utility. Novel accurate interictal biomarkers are needed to boost the presurgical value of iEEG and add to human review.
Here, we propose a multifrequency entropy-based analysis to extract iEEG characteristics that are linked to underlying epileptogenicity although imperceivable to the human eye. We aim to (i) identify the Interictal-Entropy-Zone (IENZ) in children with DRE using iEEG data with or without epileptic spikes and (ii) assess its potential to estimate SOZ and predict postsurgical outcome. We hypothesize that the IENZ is an estimator of the area responsible for generating seizures that can be estimated through multifrequency analysis of brief iEEG traces with or without spikes.

Methods: We studied iEEG (5-min) from 49 children who had epilepsy surgery with known Engel outcome. Data were segmented (3-s epochs) and grouped into epileptic (EPI) or non-epileptic (NEPI) if containing or not spikes (Figure 1A). For each contact, we computed Shannon Entropy in six frequency bands (delta to fast-ripples, Figure 1B) and compared it inside and outside SOZ. We modeled the entropy distribution across all iEEG electrodes and defined the IENZ at each frequency as the contacts standing out from the rest (patient-specific thresholds, Figure 1C). Effectiveness of the IENZ to predict postsurgical outcomes was tested based on the resection overlap (logistic regression followed by ROC curve and Fisher Exact test). NEPI and EPI data were studied separately. We computed positive and negative predictive value (PPV, NPV) of all frequency-specific entropies (separately and together) and traditional spike detection.

Results: Entropy, in multiple frequencies, was lower inside than outside the SOZ (Figure 2A) using both EPI and NEPI data. The overlap between the IENZ (spike and gamma frequency bands) and resection was higher in good than poor outcomes either using epileptic or NEPI data (Figure 2B). Removal of the frequency-specific IENZs (delta to gamma bands) predicted outcome (p< 0.05; 75% accuracy with NEPI data), but highest performance was obtained using all the frequency bands: 82% accuracy (PPV, NPV= 90%, 70%) using EPI data and 77% with NEPI data (PPV, NPV= 83%, 71%, Figure 2C). These outperformed the predictive accuracy of removing areas with high spike rates (51%, Figure 2D).