A dynamic multi-channel decision-fusion strategy to classify differential brain activity

A strategy is developed to dynamically fuse classification information from multiple channels in order to accurately classify brain activity elicited by external stimuli. The strategy is dynamic in the sense that different channels are selected at different time-instants. The channels are ranked at different time-instants according to their classification accuracies. Although the brain signals are multivariate signals, the classifiers are simple univariate classifiers. A rule is formulated to dynamically select different channels at different time-instants during the testing phase. The independent decisions of the selected channels are fused into a decision fusion vector. The resulting decision fusion vector is optimally classified using a discrete Bayes classifier. The dynamic decision fusion strategy is tested on 3 evoked potential (EP) data sets of 2 different paradigms using univariate mean and Gaussian classifiers. It is shown that the strategy yields high classification accuracies especially for high noise cases. Furthermore, the generalized formulation of the strategy makes it applicable to a wide range of multi-category classification problems involving multivariate signals collected from multiple sensors.