Heart Rate Variability Analysis for Epileptic Seizures Onset Detection

This study proposes the new method for epileptic seizure onset detection based on heart rate variability. The one- class support vector machine technique is used as a core part of the proposed approach. To extract RR intervals from ECG signal Pan-Tompkins method is used. Then six time domain features are calculated: number of RR intervals, mean value, standard deviation, root-mean-square of difference, variance and number of RR pairs whose lengths differ more than 50 ms. The preprocessing phase is based on singular value decomposition and consists of four main steps: covariance matrix calculation, singular values and singular vectors extraction, principal components computation and data scaling. The singular vector that belongs to the largest singular value is used as data point in a dataset for one-class support vector machine, which acts as anomaly detector. To calculate features D2 FPGA board is used. ECG was recorded from 31 subjects with total duration of more than 50 hours. The number of the detected seizures is 153. The results achieved are: sensitivity - 97%, specificity - 96%, precision - 95%, accuracy 95%, F1 score - 95%.Ref. 17, fig. 2, tabl. 2.