Abstracts

Rationale: The diagnosis of epilepsy relies on surrogate markers for epileptogenicity such as the presence of repeated seizures or spikes. The aim of this study is to test the usefulness of a type of non-linear mathematical analysis: recurrence quantitative analysis (RQA) of electroencephalogram (EEG) signals as a direct biomarker of epileptogenicity in the brain. Methods: We enrolled 76 patients that met the following inclusion criteria: 1) age 1 month to 21 years, 2) normal neuropsychological development. 38 patients (cases) had several episodes of electro-clinically documented absence seizures. The remaining 38 patients (controls) met the following criteria: 1) at least one EEG study performed because of the clinical suspicion of seizures, 2) normal EEG study results, 3) an electro-clinical presentation that was not consistent with epilepsy based on a thorough evaluation. We identified and collected sample EEG segments of 10-30 seconds duration each: 1) in the cases, three segments were collected: baseline without spikes, hyperventilation without spikes, and spikes; 2) in controls, two segments were collected: baseline without spikes, and hyperventilation without spikes. RQA was used to compute 19 different nonlinear measures for each of the 19 EEG channels, resulting in 19X19 or 361 features. From these, 20 of the most statistically relevant values were automatically selected and used for classification. These nonlinear features were used in two different machine learning algorithms to classify the EEG segments as belonging to cases (epileptic patients) or controls (non-epileptic patients). The concordance of this diagnostic classification was compared to the classification made by the clinical epileptologist (gold standard). Results: The results of the RQA and of the human epileptologist were highly correlated as shown in table 1. Hyperventilation increases the levels of epileptogenicity of both cases and controls, making them more difficult to differentiate. In order to control for this potential confounder, we performed a subgroup analysis eliminating the tracings during hyperventilation in both cases and controls. The resulting classification of patients into cases and controls was even more correlated with the diagnosis of the human epileptologist (Table 2). Conclusions: Analysis of EEG signals with RQA was highly correlated with the classification of the human epileptologist (gold standard). Non-linear analysis of short EEG signals of patients with epilepsy, even in epilepsy patients with normal EEG on visual inspection, may be a useful method to monitor and diagnose epilepsy.