Anticipation of epileptic seizure in advance and localization of seizure onset zone using power spectral density

To have an accurate prediction of epileptic seizure and identification of the epileptogenic region is a difficult task. This paper utilizes scalp electroencephalogram to predict an epileptic seizure and detect an epileptogenic region. To detect epileptogenic region, the signals from five different regions of brain are taken into consideration. Forty-four non-linear features are extracted from eight frequency bands theta, θ, (4–8 Hz), alpha, α, (8–13 Hz), beta, β, (13–30 Hz), gamma1, γ1, (30–50 Hz), gamma 2, γ2 (50–70 Hz), gamma3, γ3 (70–90 Hz), gamma4, γ4 (90–110 Hz) and gamma5, γ5 (110–128 Hz). Features include eight absolute spectral powers, eight relative spectral powers and twenty eight spectral power ratios. These features have been computed for ten seizure cases using a ten minute non overlapping window. From these forty four features the spectral power ratio from gamma band [30–128 Hz] [gamma1 (30–50 Hz) / gamma 3(70–90 Hz)] shows a prominent change for all the seizure cases during pre-ictal duration. The results also show that epileptic seizure is predicted in the second segment i.e. twenty minutes before the onset of seizure. Zone2 (temporal zone in this work) shows the highest change as compared to other zones so it is identified as the epileptogenic region in this work.