Abstract
Epilepsy diagnosis is commonly performed by a neurologist through visual inspection of electroencephalography (EEG) signals. Computer aided diagnosis (CAD) system has a great potential to assist neurologist or medical expert therefore improving the accuracy and shortening the diagnosis time. In this article, we present an adaptive learning approach for EEG-based CAD system for epilepsy diagnosis. With adaptive learning, the CAD system is able to reinforce new knowledge based on the neurologist feedback to improve its performance over the time. A combination of discrete wavelet transform (DWT) and Shannon entropy is used to extract feature from the EEG signal. K-nearest neighbors )kNN) clasifies the EEG signal based on normal and epileptic baseline. Both baselines are continuously updated based on the most recent classification or diagnosis result. Our proposed method shows promising results tested using publicly available University of Bonn EEG dataset with overall accuracy up to 100%.