Dynamic EEG modeling and single-evoked potential extraction using real-time recurrent neural network

Evoked potentials (EPs) of the brain are very meaningful for clinical diagnosis. The EPs are usually embedded in an ongoing electroencephalogram (EEG). The traditional method of EP extraction is ensemble averaging. In this study, for the investigation of evoked potentials in single segment measurements, a method that separates the measured activity into a spontaneous part and evoked potentials was used. The spontaneous part of the measured activities was estimated by artificial neural network (ANN). Since EEGs are time-varying signals, dynamic approaches must be used to obtain accurate results. Therefore, it was considered that post-stimulus EEG activity might be estimated by a dynamic ANN which is trained by pre-stimulus data. In this approach, EPs have successfully been extracted in single segment and results compared with the ensemble averaging in time and frequency domain.