Intracranial EEG spike detection based on rhythm information and SVM

Spike detection plays a key role in clinical diagnosis of epilepsy. Intracranial EEG is mainly used to locate the lesion according to the location and number of spikes before the epilepsy surgery. Many spike detection methods have been adopted for scalp EEG, but few of them aimed at intracranial EEG. So this paper proposes a novel spike detection algorithm using frequency-band amplitude feature and kernel support vector machine classifier for intracranial EEG data. The algorithm consists of two steps. In the first step, a fast Fourier transform algorithm computes the discrete Fourier transform of intracranial EEG, which includes the spikes and its locations marked by two expert neurologists. The total amplitude of the delta, theta, alpha, beta and gamma frequency-band is extracted as the different features, respectively. In the second step, those features are selectively used, and the kernel support vector machine is used as a classifier for training a detection model to detect spikes on the training sets. The performance of algorithm is shown to be efficient and accurate on the testing sets, and the average performance is obtained with 98.44% sensitivity, 100% selectivity and 99.54% accuracy.