ECG Beat Classification Using Linear Prediction Error Signal
This paper proposes a linear prediction method for beat classification in ECG Holter system. We assume that correlation method is used for recognition of up to 40 QRS templates. Since a classifier has to operate in real time mode a computationally efficient algorithm is used. A three state pulse-code train derived from a linear prediction error signal (LPES) is employed for classification instead of a raw signal.
The paper indicates that linear prediction coefficients do not differ significantly from beat to beat and from patient to patient.
This paper also indicates that the sensitivity of the classifier based on the three state linear prediction error signal to shape changes is sufficient for a Holter system. Its noise immunity is sufficient under condition that ECG signal is band-pass filtered and the threshold for LPES is not symmetrical. These conclusions were possible thanks to a special test signal which was generated using the first three Hermite functions.
KeywordsLinear Prediction Hermite Function Linear Prediction Coefficient Linear Predictive Coefficient Digital Recursive Filter
Unable to display preview. Download preview PDF.
- P.Laguna, P.Caminal, N.V.Thakor, R.Jane, “Adaptive QRS Shape Estimation Using Hermite Model”, IEEE Eng. in Medicine & Biology Society 11th Annual International Coference, 1989.Google Scholar