Proposal for patient-specific automatic on-line detection of spike-and-wave discharges utilizing an artificial neural network

We aimed to develop an automatic on-line detection system of spike-and-wave discharges (SWDs), which are peculiar EEG waveforms in epileptic patients. In this study, an artificial neural network (ANN) was utilized for automatic online detection of SWDs for an epileptic patient. Upon detection, 100% specificity was intended for the safety of the patient during possible future magnetic stimulation therapy. Fifty-four samples of SWD and fifteen samples of pseudo-SWD, extracted from thirty minutes of four-channel EEG signals of an epileptic patient, were employed. The ANN was trained and examined by a standard backpropagation algorithm and a leave-one-out cross-validation, respectively. Results in the off-line classification section showed both the SWDs and the pseudo-SWDs were classified perfectly. In the on-line detection section, the undetected ratio for the SWDs increased, however, a 0% false-alarm ratio was obtained. Therefore, it is suggested that the proposed method is effective for automatic on-line detection of SWDs.