BACKGROUND: Advanced intelligent methodologies could help detect and predict diseases from the EEG signals in cases the manual analysis is inefficient available, for instance, the epileptic seizures detection and prediction. This is because the diversity and the evolution of the epileptic seizures make it very difficult in detecting and identifying the undergoing disease. Fortunately, the determinism and nonlinearity in a time series could characterize the state changes. Literature review indicates that the Delay Vector Variance (DVV) could examine the nonlinearity to gain insight into the EEG signals but very limited work has been done to address the quantitative DVV approach. Hence, the outcomes of the quantitative DVV should be evaluated to detect the epileptic seizures.
OBJECTIVE: To develop a new epileptic seizure detection method based on quantitative DVV.
METHODS: This new epileptic seizure detection method employed an improved delay vector variance (IDVV) to extract the nonlinearity value as a distinct feature. Then a multi-kernel functions strategy was proposed in the extreme learning machine (ELM) network to provide precise disease detection and prediction.
RESULTS: The nonlinearity is more sensitive than the energy and entropy. 87.5% overall accuracy of recognition and 75.0% overall accuracy of forecasting were achieved.
CONCLUSIONS: The proposed IDVV and multi-kernel ELM based method was feasible and effective for epileptic EEG detection. Hence, the newly proposed method has importance for practical applications.
