Epileptic seizure detection by combining robust-principal component analysis and least square-support vector machine

摘要

The feature extraction from electroencephalogram (EEG) signals is widely used for computer-aided epileptic seizure detection. However, multiple channels of EEG signals and their correlations have not been completely harnessed. In this article, a novel automatic seizure detection approach is proposed by analyzing the spatiotemporal correlation of multi-channel EEG signals. This approach combines the maximum cross-correlation, robust-principal component analysis, and least square-support vector machine to detect the events. Our proposed method delivers higher detection sensitivity, specificity, and accuracy than the state-of-the-art approaches based on the 19 channels' EEG signals of 37 absence epilepsy patients experiencing 57 seizure events.