Differentiating epileptic from non-epileptic high frequency intracerebral EEG signals with measures of wavelet entropy

Objective: To assess whether there is a difference in the background activity in the ripple band (80-200 Hz) between epileptic and non-epileptic channels, and to assess whether this difference is sufficient for their reliable separation. Methods: We calculated mean and standard deviation of wavelet entropy in 303 non-epileptic and 334 epileptic channels from 50 patients with intracerebral depth electrodes and used these measures as predictors in a multivariable logistic regression model. We assessed sensitivity, positive predictive value (PPV) and negative predictive value (NPV) based on a probability threshold corresponding to 90% specificity. Results: The probability of a channel being epileptic increased with higher mean (p = 0.004) and particularly with higher standard deviation (p < 0.0001). The performance of the model was however not sufficient for fully classifying the channels. With a threshold corresponding to 90% specificity, sensitivity was 37%, PPV was 80%, and NPV was 56%. Conclusions: A channel with a high standard deviation of entropy is likely to be epileptic; with a threshold corresponding to 90% specificity our model can reliably select a subset of epileptic channels. Significance: Most studies have concentrated on brief ripple events. We showed that background activity in the ripple band also has some ability to discriminate epileptic channels. (C) 2016 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.