Abstracts

Rationale: One of the most debilitating aspects of epilepsy is the seemingly unpredictable occurrence of seizures (ictal states). The ability to predict epileptic seizures well prior to their occurrences may lead to the treatment of epilepsy along the lines of neuromodulation, that is, via timely electromagnetic stimulation and /or administration of anti-epileptic drugs, tens of minutes prior to a seizure onset (preictal period), to disrupt the observed entrainment of the dynamics of normal brain sites with the epileptogenic focus. In the past, Iasemidis et al. first reported results from testing a prospective seizure prediction algorithm using the nonlinear measure of short-term maximum Lyapunov exponent to measure the stability of the EEG in the time domain. Herein, we introduce a novel approach to detect preictal transitions in brain dynamics, and hence further assist seizure prediction, that measures the stability of the EEG in the time-frequency domain. Methods: Intracranial long-term EEG recordings from five patients with temporal lobe epilepsy (TLE) were analyzed. First, a time-frequency transform was generated per electrode site and EEG segment. Second, a state space was created from the time-frequency transform of each EEG segment and the measure of the respective brain site s stability of dynamics was estimated in that state space. Thus, the stability of the dynamics of the spectral probability distribution of the brain s electrical activity per EEG segment and electrode site is captured. Next, the spatio-temporal synchronization between the stability measures at critical cortical sites over time was quantified using a student s t-test. The predictability of each seizure was then estimated as the period before a seizure s onset during which synchronization between critical sites is highly statistically significant (?=0.01). The mean predictability time across seizures from this new method was estimated and then compared to a frequency-domain-only-based approach for reconstruction of the state space. Results: For the vast majority of the recorded seizures (>90%), the new algorithm detected preictal periods and estimated long predictability periods in the order of tens of minutes. The corresponding preictal entrainment of dynamics profiles of the involved critical brain sites were similar to the ones reported in the past from the time domain, using the method of delays per electrode for state space reconstruction, with a more abrupt and discernible postictal disentrainment. Finally, the measure of time-frequency stability introduced herein resulted to longer predictability times than the ones from a frequency-domain-only-based approach. Conclusions: The seizure predictability and prediction results from the new algorithm imply the robustness of the existing algorithms for seizure prediction with respect to the parameters involved in the creation of the state space they work on. In addition, the new methodology may contribute to further improvement of the sensitivity and specificity of existing seizure prediction algorithms.