Discrimination of Alzheimer's disease and normal aging by EEG data

Quantitative EEG results in Alzheimer's disease may be summarized by the term ‘slowing’, i.e. slow frequencies (delta, theta) are increased and fast frequencies (alpha, beta) are decreased. But how can EEG data be used to discriminate AD patients from controls by means of EEG data? Discriminant analysis may produce false predictions using too many predictors, as is often the case in EEG studies. We studied 4 approaches to this problem: Classification by group means, stepwise discriminant analysis, a neuronal network using back propagation and discriminant analysis preceded by principal components analysis (PCA). A maximum of 86.6% correct classifications was reached using the last mentioned approach with EEG data alone. Including age as a moderator variable in a subgroup, 95.9% correct classifications were reached.