Multirate Adaptive Filter Banks Based On Automatic Bands Selection For Epilepsy Detection

In this paper, a novel methodology that automatically identifies segments of EEG recordings with epileptic activity based on multi-rate adaptive filter banks is presented. As an advantage, the proposed approach accurately tracks parameters variability in the specific frequency band of each filter according to its energy in the spectrogram. To this end, the Shannon energy is used as information criteria for filter variability computed over the time-frequency information of the EEG data. Hence, both time and frequency data variability are considered. The proposed approach is evaluated in a real scenario where the information extracted from the obtained filter bank is used to feed a support vector machine to discriminate between normal and epileptic events. In the obtained results are validated that the proposed adaptive approach outperforms fixed filter bank alternatives.