Abstracts

Rationale: Anterior temporal lobectomy (ATL) is the most widely performed intervention for drug resistant temporal lobe epilepsy (TLE). However, as many as a third of patients experience seizure recurrence after ATL [1]. Previous studies have reported that ictal EEG patterns, interictal EEG abnormalities, MRI structural abnormalities, and their relationship to the resected area are useful for prognosticating surgical outcomes [2, 3]. Here we investigate whether the spectral characteristics of normal resting state EEG can predict seizure-related outcomes following ATL.

Methods: Our dataset included presurgical scalp EEGs of 54 TLE patients from the Mayo Clinic, Rochester (MCR) and 25 TLE patients from the Cleveland Clinic (CC). All EEGs were recorded in outpatient settings or during the first day of prolonged inpatient studies using the extended 10-20 montage. 34 MCR and 13 CC patients experienced seizure freedom for at least one year following ATL. We preprocessed the EEGs to remove power line noise and common artifacts and selected multiple 10s-long eyes closed awake epochs free of epileptiform activity. For each epoch, we calculated the absolute spectral power (ASP) of delta, theta, alpha, beta, and gamma bands, and aggregated them among the channels located in prefrontal, frontal, temporal, central, parietal, and occipital regions. Similarly, we calculated the interhemispheric spectral coherence (ISC) in each region within the same spectral bands. We first normalized the two datasets separately and statistically analyzed the differences between the patients who experienced seizure freedom and those who did not. We then employed a Naïve Bayes classification approach using the above spectral features to predict surgery outcomes and trained it using a leave-one-patient-out cross-validation within the MCR dataset and tested using the out-of-sample CC dataset (fig 1a).

Results: We found that the ASP of alpha and beta bands across all regions, and ISC in parietal and occipital regions across all frequency bands were positively correlated with good surgical outcomes (p< 0.01, marked in red on fig 1b and 1c). We also found that our Naïve Bayes-based classification approach using those features could predict 1-year seizure freedom following ATL with AUC values of 0.74 and 0.75 for the MCR and CC datasets, respectively (fig 1d).