This paper presents an automatic computational system to detect and classify epileptogenic transients in electroencephalogram (EEG), to aid Epilepsy diagnosis. Due to great variability on the morphology of these events, conventional tools of pattern recognition are not able to distinguish between normal and epileptogenic activity, and visual detection is a very time-consuming task. So, tools and methods normaly used to detect these events try to imitate human expertise. False positive detection represents considerable impediment to extensive use of automatic systems by EEG readers. The proposed system applies Wavelet Transform to extract only epileptogenic features from the EEG signals, and a group of specialized Artificial Neural Networks (ANNs) to distinguish spike and sharp wave events from normal background activity. Two ANNs are used cooperatively, allowing greater flexibility in adjusting system's sensibility and specificity, to improve performance. When sensibility and specifity are set to be equal, system's performance achieves 80%.
Automatic Detection; Epilepsy; Neural Networks; Spike; Wavelet Transform
