Prediction coding of ECG by nonlinear autoregressive model

Abstract

In linear prediction coding methods for ECGs using linear autoregressive models, the prediction accuracy of QRS waves is poor, which is not improved even when the prediction degree is set higher than second or third degree. In this paper, this is attributed to the fact that a QRS wave is produced by a nonlinear occurrence mechanism and ECGs contain nonlinear components that cannot be predicted by linear models. A nonlinear prediction coding method for ECGs using a layered neural network or a Volterra functional series, such as are used frequently to identify nonlinear systems, is proposed as a nonlinear autoregressive model. The accuracy of prediction of QRS complex is improved by using a nonlinear model, the average code length in the bit rate region greater than 3 bits is improved by about 0.1 to 0.3 bit, and superior coding efficiency is realized. This paper shows that the proposed method using a nonlinear model is especially effective in improving the efficiency of coding of ECGs since the improvement of the coding efficiency is at most 0.1 bit with ECG coding methods using linear transforms, such as linear prediction, orthogonal wavelet transforms, and the like. © 2000 Scripta Technica, Syst Comp Jpn, 31(7): 66–74, 2000