A novel approach for epileptic seizure detection

A new on-line adaptive methodology is introduced for detecting a suspected epileptic seizure from an electroencephalogram (EEG). This is achieved by using the angle between two progressive oblique spaces. Two seizure indices are introduced. In the absence of seizure, the EEG remains short time stationary, and hence the seizure indices remain approximately unity. If and when abnormality or epileptic seizure activity occurs, the EEG becomes non-stationary causing the seizure indices to drop in value. The dips in the seizure indices indicate the time at which the seizure activity occurs, while the magnitude of the dips indicate the strength of the abnormality. Simulation is carried out to show that the methodology can be used to detect the signal burst in a stationary or a short time stationary environment adaptively. Probability of error detection is given for both seizure indices. Results for real data collected from epileptic patients are given to substantiate the methodology. The proposed methodology provides an objective criterion for detecting suspected epileptic seizure. It can be used to analyze routine and 24 hour EEG records. Since the detection is on-line and adaptive, if and when a possible epileptic seizure activity is detected, the methodology can be applied to activate appropriate neuro-transmitters (thalamic, cerebellar, or vagal) instantly as a preventive measure. The algorithm can be implemented in VLSI form as an implantable device