Data mining of Intracranial Interictal EEG recordings of Epilepsy patients with focal cortical dysplasia.

Abstract

Epilepsy is a group of neurological diseases that affects up to 1% of the world’s population. About a third of the patients diagnosed with epilepsy are considered with difficult treatment (refractory), this group of patients can benefit from a resective surgery, that removes the epileptogenic tissue of the brain. Nowadays, the exams for the delineation of the areas for resection are still imprecise, and one of the techniques for a better definition of these brain areas require electrophysiological examination with invasive intracranial long-term electroencephalography monitoring (iEEG). One of the strategies for determining the epileptogenic zone (EZ) is to analyze the interictal data of patients with favorable outcomes and unfavorable outcomes with respect to the surgery resected areas and determine the statistical significance between them. A detection, analysis, and clustering data mining algorithm was used in order to extract information of 52 patients with focal cortical dysplasia (FCD) epilepsy. The detection algorithm identifies the interictal epileptiform discharges (IEDs) and arranges the detected activities into clusters given the patterns of spreading. For the statistical analysis, a comparison of the clustered data from three different vigilance epochs (sleep, awake and the combination of both) identified the most relevant epoch for identifying the epileptogenic areas and extract additional parameters. The results showed that the combined epoch of awake and sleep showed strong statistical significance in relation to the outcomes, followed by sleep and awake, respectively. Furthermore, given the positive results of the first analysis, an additional data mining was done in order to utilize the algorithm’s outputs to predict the patient’s FCD group, a predictive model was trained and displayed accuracy greater than 80% when tested with non-trained data.