An Automated Seizure Detection Algorithm Using Time and Frequency Analysis with Pattern Recognition Neural Network for FPGA Implementation

Abstract — The purpose of this paper is to investigate a novel algorithm to detect the start and stop time of the seizure occurrence by observing EEG signals. This algorithm has been intended to be implantable in patient-monitoring devices for real time applications. Initially, we selected four frequency bands of EEG signals with the aid of multi-stage FIR filter banks. Subsequently, effective features in time and frequency domains are extracted to build a feature matrix. The time domain features include the number of positive and negative peaks, the summation of absolute values of the voltage and the number of zero crossings. The frequency features include the summation of the FFT values and the number of Fourier coefficients crossing the average FFT coefficient values. The resultant feature matrix is then passed through a pattern recognition artificial neural network to be classified into two categories of normal and seizure activities. A post-processing is then applied on the raw classification results to obtain more accurate and stable outcomes. The performance of the algorithm is evaluated on 13 hours of seven cases collected from six epileptic patients. Experiments show that the proposed method could achieve the accuracy of 98.83 % for detecting the start time of the seizure (in second) and 99.31 % for detecting the end time of the seizure (in second)