Abstracts

Rationale: For children undergoing epilepsy surgery, locating the seizure onset zone (SOZ) and area to resect is often nontrivial. Many diagnostic techniques are used to this purpose; among them intracranial EEG (iEEG) is regarded the best to record and localize epileptic brain activity. Though, finding potential sites for resection through iEEG requires an expert to visually screen the data; plus, the process is hindered if no seizures occur while monitoring. The possibility to estimate SOZ and tailor the resection using iEEG characteristics that are imperceivable to the human reader would boost the iEEG value, adding to the traditional human review. _x000D_ _x000D_ Here, we study subtle iEEG changes using entropy analysis at various frequencies and identify potentially epileptic iEEG contacts. We hypothesize that low iEEG entropy (high regularity) is an interictal biomarker of epileptogenicity and aim to develop a new tool to aid clinicians estimate SOZ and area to resect by locating Low-Entropy-Zones (LEZ) from short interictal iEEG with, or even without, clear epileptic patterns._x000D_
Methods: We studied iEEG data (5-min) from 40 children (12.7±5.5 yrs) who had epilepsy surgery with known Engel outcome. Data were segmented in 3-s epochs that were grouped into epileptic or silent (Figure 1A) if containing or not epileptic spikes.  _x000D_ _x000D_ For each contact, we computed normalized Shannon Entropy in 5 frequency bands (Figure 1B) and compared it between iEEG contacts inside and outside SOZ or resection. We identified the iEEG contacts with lowest entropy (LEZ) via patient-specific thresholds defined by piecewise linear regression. Finally, we narrowed down to new regions of interest (LEZ) and assessed if their resection (Figure 1C) was predictive of postsurgical outcome (ROC curve; Fisher’s test). We conducted all the analyses on silent and epileptic data separately to assess if the value of our entropy measures depends on having visible epileptic patterns on iEEG. _x000D_
Results: We found that low entropy, in multiple frequencies, characterizes the SOZ (Fig 2A): lower entropy is seen inside than outside SOZ (p < 10-3) using both epileptic and silent iEEG._x000D_ _x000D_ Lower entropy is also seen inside than outside resection in good-outcomes (p < 10-2) in any frequency (Figure 2B); interestingly, this is not seen in poor outcomes when analyzing silent iEEG (p>0.05, Figure 2C). In addition, the overlap of LEZs with resection is higher in good than poor outcomes when analyzing either epileptic or silent iEEG data as Figure 2D shows.  _x000D_ _x000D_ Finally, removing what we identified as LEZ is predictive of outcome in our cohort (Figures 2E and 2F) with 74% accuracy (using either epileptic or silent data), and positive and negative predictive value of 85-86% and 61-63%.  _x000D_
Conclusions: We presented a new presurgical tool able to delineate a patient’s LEZs in various frequencies via short iEEG data. Our findings show that such a tool can locate potential regions of interest to target during surgery (independently from the presence of spikes on the iEEG that is analyzed) and provides outcome prognostication for epilepsy surgery with positive and negative predictive values above 85% and 61%. 

Funding: BCH/Harvard Faculty Career Development Fellowship