ニューラルネットワークを用いた睡眠脳波のスペクトル時間推移パターン認識による特徴波の検出

抄録

We proposed a new type Neural Network model by which we could detect a several kinds of important characteristic waves in EEG that is needed for the diagnosis of sleep stages. In this paper, we compared the proposed method with two conventional type Neural Network medels and likelihood ratio method as to the performance of detecting the characteristic waves from sleep EEG. Experimental results indicated that the proposed Neural Network model had much more capacity for detecting the isolated and transient characteristic waves than other conventinal methods and little depended on the duration of characteristic waves. Analyzing the connection weights of the Neural Network model revealed that XOR operation was carried out for detecting the isolated characteristic waves.