RESUMEN
We investigated changes of the scaling exponent alpha estimated by detrended fluctuation analysis (DFA) of electroencephalograms (EEG) in patients with dementia including Alzheimer's disease(AD), and attempted to apply a method of pattern recognition using the alpha value-based feature vector to classify dementia. In 9 patients with AD, 8 patients with other types of dementia (vD), and 7 patients without dementia(C), DFA was performed for approximately one minute with background EEG data recorded at 16 different scalp monopoles. The alpha values were significantly higher in patients with AD at electrodes F7, C3, P3, P4, T3, and T5 than in patients without dementia. No significant difference in alpha values was found between patients with vD and without dementia. Then, an artificial neural network (ANN) was trained on the alpha value-based feature vector of EEG to classify patients with dementia into AD and vD. The trained ANN successfully diagnosed all four new test cases of AD. From these observations, it is suggested that AD has a specific pattern in the alpha value-based feature vector. Thus, pattern recognition using alpha value-based feature vector may be useful for the classification of dementia.