Towards real-time in-implant epileptic seizure prediction

We present an architecture of an epileptic seizure prediction system suitable for an implantable implementation. The microsystem comprises a neural interface, a spectral analysis processor and an artificial neural network (ANN). The neural interface and the spectral analysis processor have been prototyped in a 0.35 mu m CNOS technology with experimental results are presented. The wavelet-based artificial neural network predicts the onsets of seizure up to two minutes before their occurrence in an in-vitro epilepsy model using a mouse hippocampal brain slice with recurrent spontaneous seizures.