Masking Artifacts in EEG Time Series

Abstract

SummaryThe removal of artifacts from EEG time series by masking the concerned parts has been studied in order to derive error estimates which apply in general. For this purpose, we selected a sufficiently representative set of artifact-free time series registered from 81 individuals. These time series then served as error-free references. Subsequently, we used the same set of time series, simulated artifacts of 2-12 seconds length (each of which at 5 different positions within each epoch) and investigated the impact of masking procedures on spectrum analysis. Our interest focused particularly on clearing up possble sources of variation (e. g., the actual frequency composition of the signal or the actual position of the masked segment within the time series) in theerrors inherent in the masking process. A detailed error analysis yielded no indication that the inherent errors depend on such factors. Moreover, it turned out that any sufficiently representative set of, say, about 100 epochs of 20 seconds length can be used to reproduce all required estimates at an errorof less than 5 percent. Based on these findings, we developed a correction function which applies in general, thus simplifying the fully computerized processing of EEG time series.