Computationally-efficient algorithms for real-time ECG baseline removal

Abstract

ECG signal processing has been one of the major studied topics in the biomedical field. The introduction of new techniques and the extensions to the previous keep constantly evolving the span of the ECG research, providing a true realisation of the problems specific to each new approach. For this reason, thorough data analysis and accuracy evaluation have been the most significant tools in effective quantifying of ECG noise elimination techniques.

The ECG signal is generally defined to have a spectral content between 50 mHz and 150 Hz with a few millivolts in amplitude, and identified as susceptible to physiological and environmental interferences. The elimination of noise interferences, in particular the baseline wander, is a major concern in preserving the ECG signal integrity (i.e the ST segment) due to the overlapping spectral content of noise sources with this segment. The inherent complexity of such a problem has led to computationally-intensive algorithms in the literature (i.e. Empirical Mode Decomposition (EMD), Independent Component Analysis (ICA), Wavelet Transforms (WT) and others) and the removal of the baseline drift is acquired off-line with powerful simulation tools. The adaptations of these methods for ambulatory designs, on the other hand, demonstrate substantial accuracy degradation due to scaling. Therefore, real-time approaches to match comparable accuracy to the computational intensive algorithms are yet to be proposed.

This research investigates a computationally-efficient baseline wander removal technique and targets comparable performance to its computational off-line counterparts reported in the literature while preserving the signal integrity of the ECG.

This work introduces a novel hardware-efficient real-time baseline estimation method based on three distinctive ``isoelectric'' fiducial point detections per heart beat. These detected points are cubic spline interpolated to achieve a realistic representation of the baseline estimation, and removed from the noisy signal to provide an ``undistorted'' ECG signal representation. Computational efficiency of this approach is further improved with a novel weighted piecewise linear interpolation technique. This approach targets non-uniformly sampled systems with less computational requirements compared to the higher order polynomial interpolation. The MCU-based real-time hardware system realisation of these algorithms demonstrates accurate ambulatory system response and this is the first tested system level design addressing baseline wander removal with detailed analysis. The validated tests have presented original contributions for baseline wander detection and removal by tackling one of the most crucial challenges currently present in clinically valid ECG signal processing. The accuracy and the computational requirements of the developed algorithms show real-time capabilities of the overall system and challenge its computational ECG signal processing counterparts.