Abstracts

Rationale: Epileptogenicity is a very complex brain property depending on the interplay between altered excitability and connectivity. In children with drug-resistant epilepsy (DRE), identifying the epileptogenic zone can be challenging, even when using intracranial EEG (icEEG). Recent evidence shows that, to treat focal DRE, we must localize pathological regions (depicted by altered excitability) but also understand how they interact within the epileptogenic network (identifying altered connections). We hypothesize that connections between highly-excitable and highly-connected regions will denote “epileptogenic hubs,” which must be disrupted for seizure-freedom.

To this purpose, we propose a novel twofold approach to optimize icEEG interpretation, that is able to quantify both brain excitability (via phase-amplitude coupling, PAC) and functional connectivity (FC), using interictal data independently of the presence of frank epileptiform patterns. Our main goal is to develop a new computer-aided tool to boost icEEG reading and improve surgical planning in children with DRE, without requiring evocation of seizures.

Methods: We studied icEEG data (5-min epochs) from children who had resective epilepsy surgery at Boston Children’s Hospital with known outcome (Engel). Figure1A-B shows how, for each electrode, we computed measures of excitability (PAC) and connectivity (through FC and graph analysis) for four frequency bands. We quantified the ability of each individual measure to identify the brain tissue resection that predicts seizure-freedom (Fisher’s-exact-test), by estimating positive and negative predictive value. Finally, we designed a fuzzy-inference-system that combined connectivity and excitability measures into one “epileptogenicity-index” per icEEG contact (Figure2A) and identified the “most epileptogenic hubs.” We tested whether the output of our fuzzy-inference-system predicted postsurgical outcome—based on each patient’s resection—and compared it with a support vector machine (SVM), which was trained to identify the tissue to resect using data from good outcome patients only.

Results: We included 32 children (21 good outcome, Engel 1) who had stereotactic-EEG and/or electrocorticagraphy (ECoG) before surgery. Using individual measures of FC and PAC, we predicted patient’s outcome with a maximum accuracy of 75% (p-value < 0.05). Through these results, we identified the most predictive measures and designed our fuzzy inference system (Figure 2A): we found that removing the most epileptogenic hubs, identified by our newly designed system, predicts postsurgical outcome with 88% accuracy (p-value < 0.001). This outperformed the SVM, which presented 66% accuracy and failed to predict poor outcome (negative predictive value=50%), as Figure2B