An adaptive geometrically-complemented approach for ECG signal denoising

Abstract

This paper proposes a geometrical criterion for denoising a single-lead ECG signal. It was designed to ease the use of heuristic procedures for removing the most common types of noises from ANSI/AAMI-compliant ECG signals. However, in this paper, only the system-noise was considered to illustrate how this geometrical criterion is applied to the signal. The proposal here presented relies on a voltage-level slope detector that marks where the signal starts to increase, decrease or remain at the same level in order to perform an abstract segmentation of the ECG signal. The resulting segments are quantitatively classified as significant segments or noisy segments by analyzing their amplitude and time duration according to a previously defined threshold-level with the intention of helping the algorithm to decide its own operational parameters. The system-noise filter proposed here has five different operation modes. The main one is based on the arithmetic mean operation to smooth out short-term fluctuations; additionally, it is complemented with geometrical estimations for preserving the physiological characteristics of the ECG signal. The other operation modes are purely based on geometric estimations to calculate the filter output. The geometrical criterion described here differs from many other approaches presented until now owing to its low mathematical complexity and low computational consumption since all calculations can be performed with raw ADC readings and arithmetical operations, characteristics that make this filter easy to implement on embedded systems. This denoising approach was designed for online processing applications but it also works well with previously recorded signals.