Predicting epileptic seizure by recurrence quantification analysis of single-channel EEG

This paper presents a promising approach, based on the combination of two different approaches, including recurrence quantification analysis(RQA) of single-channel EEG and artificial neural network (ANN), which predicts seizure onsets in 17 consenting patients with generalized epilepsy. Eight channels of EEG were collected in each patient in Epilepsy Center of Xijing Hospital. The seven measures were extracted from the raw EEGs by RQA, and then a four layer(7-6-2-1) ANN was employed for prediction, using the measures as the inputs. The performance obtained for the proposed scheme in predicting succedent seizures is: sensitivity 50 similar to 72.2%, specificity 61.9 similar to 73.8% and accuracy 58.3 similar to 70%, depending on the different EEG leads.