Epileptic seizure detection using adaptive Gabor wavelet filter bank

Epileptic seizure identification is crucial in early diagnosis and treatment of epilepsy, a neurological disorder that strikes millions of people worldwide. In this work, an Adaptive Gabor Wavelet Filter Bank (AGWFB)-based method has been presented, that boosts seizure identification accuracy from an electroencephalogram (EEG) signal. Gabor wavelet, owing to its best time-frequency locality, has been implemented here to extract discriminating attributes from an EEG signal. AGWFB selects an optimum parameter of orientation dynamically based on signal attributes, boosting seizure classification through feature representation. Support vector machine (SVM) classification utilizes these resultant features. Comparisons from several benchmark EEG databases reveal that AGWFB outshines other conventional wavelet-based strategies as well as deep-learning strategies in terms of sensitivity, specificity and accuracy. The adaptive property of the filter bank also confers stability against both inter-subject variability, as well as intra-subject variability, so it has the possibility of application in seizure detection contexts. The method has the benefits of being an efficient, interpretable, as well as accurate one, for seizure monitor systems in the clinic. Improving efficiency, as well as deploying AGWFB in continuous, low-power, embedded systems, are future work areas.