Topographic components analysis of evoked potentials: parameter estimation and some preliminary results

Analyzing multichannel evoked potentials (EPs) is a difficult problem because of the typically large volumes of data acquired-data which are space- and time-dependent and which must be analyzed in both of these domains. Some preliminary results from using the topographic components (TC) model recently proposed by Mocks (1988) for spatiotemporal EP analysis are presented. A three-stage steepest-descent algorithm is derived for least-squares parameter estimation. The algorithm is used to fit the TC model to actual EP data and to examine two aspects. The model is compared to the standard method of principal components for its ability to fit EP data. For two sample data sets, the TC model is shown to provide a better least-squares fit with many fewer parameters than principal components, suggesting that the seemingly restrictive assumptions made by the model may not be inappropriate. The ability of the model to discriminate between two classes of subjects distinguished by different topographic patterns to their evoked responses for a given stimulus is evaluated