A rational strategy for the automated detection of artifacts in heart period data is outlined and evaluated. The specific implementation of this approach for heart period data is based on the distribution characteristics of successive heart period differences. Because beat-to-beat differences generated by artifacts are large, relative to normal heart period variability, extreme differences between successive heart periods serve to identify potential artifacts. Critical to this approach are: 1) the derivation of the artifact criterion from the distribution of beat differences of the individual subject, and 2) the use of percentile-based distribution indexes, which are less sensitive to corruption by the presence of artifactual values than are least-squares estimates. The artifact algorithms were able to effectively identify artifactual beats embedded in heart period records, flagging each of the 1494 simulated and actual artifacts in data sets derived from both humans and chimpanzees. At the same time, the artifact algorithms yielded a false alarm rate of less than 0.3%. Although the present implementation was restricted to heart period data, the outlined approach to artifact detection may also be applicable to other biological signals.
Berntson et al. (1990) studied this question.