Adaptive modelling of biological time series for artifact detection

The authors propose a method for artifact detection based on linear modelling of biological time series. An artifact, coming from a different "source", generally does not fit in the model and can be detected. Biological time series are not stationary, so that adaptive filtering is used for model estimation. Real time constraints warrant the use of predictive models only past input values are used to predict the current sample values. A set of thresholds or the prediction errors is used to detect artifacts. The authors model each time series by means of an adaptive prediction filter and, when a priori knowledge or the relation between two measurements, is available, they model this specific cross-channel relation with an adaptive filter. They applied this method to sequences of cardiovascular measurements from ICU and from Holter monitoring. The results obtained are fully satisfactory.<<ETX>>

[1]  S. Haykin,et al.  Adaptive Filter Theory , 1986 .

[2]  F. Conforti,et al.  A two-channel adaptive filtering approach for recognition of the QRS morphology , 1991, [1991] Proceedings Computers in Cardiology.

[3]  W. Y. Chen,et al.  A variable step size LMS algorithm , 1990, Proceedings of the 33rd Midwest Symposium on Circuits and Systems.

[4]  Peter M. Clarkson,et al.  A class of order statistic LMS algorithms , 1992, IEEE Trans. Signal Process..