Discrimination between demented patients and normals based on topographic EEG slow wave activity: comparison between z statistics, discriminant analysis and artificial neural network classifiers

Abstract

The topographic distributions of absolute delta and theta powers were used to classify demented patients and normals by means of z statistics, discriminant analysis and artificial neural networks (NN). The data were taken from two psychopharmacological studies in mildly to moderately demented patients (111 and 96 patients for studies I and II, respectively) and from 56 normal healthy controls. All patients were diagnosed according to DSM-III criteria and were free of medication for at least 2 weeks. The NN used was a strictly layered feed-forward network with complete connections. The z-transformed absolute power values in the combined delta and theta frequency range at 17 electrodes, recorded in a 3 min vigilance-controlled EEG with eyes closed, were used as input. After having trained the NN successfully by backpropagating of errors, the generalization test with independent data results in a classification performance of 90% determined by "relative operating characteristic" analysis. The NN out-performed z statistics and discriminant analysis. This high percentage of correct classifications may justify the development of further application of NNs based on topographic EEG data.