Skip to Main Content
Evoked responses or event related potentials in human EEG have been mostly studied with off-line analog recording and averaging. It is shown here that, at least in some situations, it is possible to detect and classify individual evoked responses or "single epochs" with surprising reliability. To do so, however, requited thinking a new not only the data processing but the whole experimental strategy. The classification is done in real-time by treating the experiments as a signal detection problem in which the computer, in the position of impartial observer, assigns classes to incoming epochs accogding to a predetermined decision rule. Since data collection and processing are interleaved, each classification outcome can be a factor in experiment control as well as in subject feedback. The discrimination performance, expressed in terms of mutual information, is shown to be both a practical index for procedure optimization and a concise and specific descriptor for the experiment results.