Instantaneous Measurement of SNR in Electrocardiograms Based on Quasi-continuous Time-Scale Noise Modeling

Proper measurement of signal quality is a common problem in biomedical applications, including the electrocardiogram interpretation. The need for a reliable signal-to-noise estimate raises with the common use of telemedical recordings performed in the home care conditions and interpreted automatically. The currently used techniques perform noise measurements on the baseline and extrapolate the results to a whole heart beat. Main drawback of this method lies in irregular occurrence and short duration of the baseline. This paper presents a new ECG-dedicated noise estimation technique based on a time-frequency model of noise computed in a quasi-continuous way. The proposed algorithm uses the temporarily adapted local bandwidth variability of cardiac electrical representation to recognize cardiac components. The part of the time-frequency plane remaining above the local bandwidth of the signal, represents background activities of any origin (muscle, mains interference etc.). This noise estimate in each particular scale is available as non-uniformly sampled and has to be interpolated to the regions where components of cardiac representation are expected. In lower scales, the noise contribution is computed with use of square polynomial extrapolation. The algorithm was implemented and tested with use of the CSE Database records with the addition of the MIT-BIH Database noise patterns. The differences between the added and estimated noise show similar performance of baseline-based and noise model-based methods (0.69 dB and 0.64 dB respectively) as long as the noise level is stable. However when dynamic noise occurs, the baseline-based method is outperformed by the proposed algorithm (2.90 dB and 0.95 dB respectively) thanks to consideration of multiple measurement points and accurate noise tracking.