A new sensitivity-specificity product-based automatic seizure detection algorithm

Epilepsy is a disorder of the human brain function affecting 1% of the world's population. Automatic epileptic seizure detection is important to help neurologists to interpret the electroencephalogram signal readings, particularly the signals recorded in the ictal or seizure attack, which are more crucial than those recorded in the inter-ictal (between the attacks). Time-frequency (t-f) analysis methods, wavelet transform, and linear discriminant analysis are the most common modalities used for epileptic seizure detection. The main objective of this work is to compare between ten different test cases of the EEG signal detection methods over twenty patients considering the sensitivity, specificity, and the accuracy. The analysis has been conducted in three levels: Firstly, the EEG is filtered by a discrete wavelet transform (DWT); Secondly, five features which are relative energy, fluctuation index, variance, energy and autocorrelation are calculated; and finally, these features are applied as inputs to the support vector machine (SVM) to detect the occurrence of epilepsy. Due to the trade-off between sensitivity and specificity (i.e. as a sensitivity is improved, the specificity is degraded and vice versa), a new technique which is sensitivity-specificity product is proposed in this work. Simulation results on different test cases have shown that the maximum sensitivity-specificity product occurs when only four features are included (i.e. relative energy, fluctuation index, energy, and autocorrelation) and the fifth feature (i.e. the variance) is excluded.