Mathematical classification of evoked potential waveforms

Abstract

It is shown that evoked potential waveforms can be classified into clinically significant patterns in computerised EEG analysis and other medical fields through application of a pattern-recognition technique which is presented. Fourier descriptors are used to characterize each waveform. Since the feature sets of Fourier descriptors contain information about the shape of the waveforms, they are used in the structural stage of the algorithm. In this stage, discriminant analysis is used to design a measure of dissimilarity between pairs of waveforms. Only the resulting interdistance matrix is preserved in the construction of the vector representation, making the dissimilarity measure crucial to the success of the remainder of the procedure. An algorithm using Fourier descriptors as feature vectors and the sum of squared error criterion to indicate improvement in percent of correctly classified waveforms has yielded accuracies of at least 90%.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>