Missing data problem in random electrocardiogram signal processing

Abstract

Basically, signals are the entities that convey information and biomedical signals are the signals that carry information about the physiological process of organisms. Electrocardiogram (ECG) signal or known as heart signal is the signal that contains information about electrical activities in the heart. Since physiological signal are generated at low values and devices advancements are not sufficient to detect these small values perfectly, these signal tends to be missing from the record. As the noise interferes the signal at the same time, raw signal is practically unreliable to be interpreted directly. Hence, the random signal processing is required to obtain the signal as precise as possible. In this study, the missing probabilities of signal missingness were set to 0.1 at high values and 0.3 at low values. The noise to be reduced is Gaussian noise with zero mean and standard deviation 0.01 mV. A few methods have been applied to estimate the missing signal, including single mean imputation, empirical conditional mean imputation and Holt-Winters exponential smoothing. For noise filtering, the approach used is the Finite Impulse Response (FIR) Wiener filter. The study finds that the empirical conditional mean imputation is the best method among the three to estimate missing signal due to its accuracy, adequacy and simplicity. However, it appears that the FIR Wiener filter does not compatible with the estimation from empirical conditional mean imputation and does not further improve the signal quality by removing noise in general.