Multichannel Fusion Models for the Parametric Classification of Differential Brain Activity

Abstract

This paper introduces parametric multichannel fusion models to exploit the different but complementary brain activity information recorded from multiple channels in order to accurately classify differential brain activity into their respective categories. A parametric weighted decision fusion model and two parametric weighted data fusion models are introduced for the classification of averaged multichannel evoked potentials (EPs). The decision fusion model combines the independent decisions of each channel classifier into a decision fusion vector and a parametric classifier is designed to determine the EP class from the discrete decision fusion vector. The data fusion models include the weighted EP-sum model in which the fusion vector is a linear combination of the multichannel EPs and the EP-concatenation model in which the fusion vector is a vector-concatenation of the multichannel EPs. The discrete Karhunen-Loeve transform (DKLT) is used to select features for each channel classifier and from each data fusion vector. The difficulty in estimating the probability density function (PDF) parameters from a small number of averaged EPs is identified and the class conditional PDFs of the feature vectors of averaged EPs are, therefore, derived in terms of the PDFs of the single-trial EPs. Multivariate parametric classifiers are developed for each fusion strategy and the performances of the different strategies are compared by classifying 14-channel EPs collected from five subjects involved in making explicit match/mismatch comparisons between sequentially presented stimuli. It is shown that the performance improves by incorporating weights in the fusion rules and that the best performance is obtained using multichannel EP concatenation. It is also noted that the fusion strategies introduced are also applicable to other problems involving the classification of multicategory multivariate signals generated from multiple sources.

A Feature Ranking Strategy to Facilitate Multivariate Signal Classification

“…The classification accuracies for the ERPs of the 3 subjects, averaged across 200 trials, are summarized in Table II. The best results from the previous study for the 3 subjects, using a weighted multi-channel ERP fusion technique [16], are included in the last column of the table. Because the previous study did not include single-trial classification, the results for r=1 are labeled NA.…”

“…This, however, is not unexpected because ERPs are known to have very poor SNRs [15], [16]. In a relative sense, TMEP signals are not as noisy as ERP signals.…”

“…In the usual ERP model, it is assumed that the background EEG noise is statistically equivalent across the ERP classes [15], [16]. Consequently, it can be assumed that the covariance matrix is the same across the match and mismatch conditions.…”

“…Signal averaging is the most often used operation to improve the signal-to-noise ratios in ERPs [15], [16] and classification results can be presented as a function of the number of single-trials averaged.…”

“…Table III shows the classification accuracies averaged across the 3 subjects as a function of r. On the average, the PCT gave the best classification accuracies and the 2-dimensional DCT gave the worst accuracies across the 4 selection criteria and as well as r. The classification accuracy of the PCT classifiers using the magnitude criterion was slightly higher than that of the PCT classifier using the variance selection criteria for r=2 and 4. In the previous decision and data fusion study [16] involving the same multi-channel ERP data, the best results were obtained using a data fusion strategy in which the ERPs of all 14 channels were concatenated into a multi-channel ERP vector. In order to develop a multivariate classifier, the dimension of the multi-channel ERP vector was decreased to satisfy the dimensionality condition using a two-step procedure.…”