Epileptic seizure detection with linear and nonlinear features

Automatic seizure detection is significant in both diagnosis of epilepsy and relieving the heavy workload of inspecting prolonged EEG. This paper presents a new seizure detection method for multi-channel long‐term EEG. The fractal intercept derived from fractal geometry is extracted as a novel nonlinear feature of EEG signals, and the relative fluctuation index is calculated as a linear feature. The feature vector, consisting of the two EEG descriptors, is fed into a single-layer neural network for classification. Extreme learning machine (ELM) algorithm is adopted to train the neural network. Finally, post-processing including smoothing, channel fusion, and collar technique is employed to obtain more accurate and stable results. Both the segment-based and event-based assessments are used for the performance evaluation of this method on the 21-patient Freiburg dataset. The segment-based sensitivity of 91.72% and specificity of 94.89% were achieved. For the event-based assessment, this method yielded a sensitivity of 93.85% with a false detection rate of 0.35/h.

► The fractal intercept and fluctuation index are extracted as EEG features. ► Extreme learning machine (ELM) is employed to train a neural network classifier. ► Post-processing is used to obtain more accurate and stable detection results. ► Experiments with long term EEG of 21 patients demonstrate the effectiveness.