Piecewise linear correction of ECG baseline wander: a curve simplification approach

doi:10.1016/j.cmpb.2004.10.008

Computer Methods and Programs in Biomedicine

Volume 78, Issue 1, April 2005, Pages 1-10

https://doi.org/10.1016/j.cmpb.2004.10.008 Get rights and content

Summary

In this paper, we suggest a novel method for ECG baseline correction, exclusively based on pattern recognition tools, namely, dominant points (DPs). The DPs are computed by the Douglas–Peucker curve simplification algorithm. The so-computed DPs include peak and baseline points, the discrimination of which yields gradual piecewise linear estimation of the baseline wander (BaselineW) in two iterations. At each iteration, the current BaselineW is subtracted from the input signal according to the decomposition scheme: ECG ≈ ECG_ZBLW + BaselineW, where ECG_ZBLW is the underlying baseline wander free ECG. The method targets many types of baseline deviations in a unified approach: baseline drift due to respiration, amplitude modulation due to perspiration and abrupt potential change due to electrode loose contact. We tested the developed method on a variety of ECG records including half synthesized records contaminated with different types of baseline deviations (simulated) noise, and on records from the MITBIH database presenting important baseline deviations, including normal and abnormal heart beats cases. The method showed good performance in computing a piecewise linear estimation of the baseline deviation and in extracting the ECG_ZBLW, which represents the clinically significant electrocardiogram information.

Introduction

Electrocardiogram (ECG) baseline wander removal is an important preprocessing phase for features detection, analysis and diagnosis, that not only eases clinician visual inspection, but also makes the automation process of the above tasks more accurate. For example, clinicians proceed to measurements of ST, RR and QT segments to depict any abnormalities in the cardiac activity. This difficult task is even harder for them in the case of ECG records contaminated with artifacts and noise. Common sources of such disturbances include power-line interference, electrode contact noise, motion artifacts, muscle contraction (EMG), respiration and also a combination of more than one of these sources. In this study, we propose a unified method capable of linearly estimating a variety of baseline deviations: baseline drift due to respiration, amplitude modulation due to perspiration and ECG abrupt potential change in case of electrode loose contact, due to patient movement. Targeting of these kinds of noise is motivating since they are ubiquitous in electrocardiograms, especially in ambulatory ECG (Holter) and exercise ECG.

The rest of this paper is organized as follows. In Section 2, some previous works related to the proposed method are cited. In Section 3, design considerations of the proposed method are presented. In Section 4, a general description as well as the details of important aspects of the method are presented. In Section 5, some of the experiments we have performed on the proposed algorithm are reported. In Section 6, results of the conducted experiments are discussed. In Section 7, the authors emit concluding thoughts and remarks.

Section snippets

Background

Previous works on ECG baseline correction include filtering methods (e.g. [1], [2]). This approach, basically, proceeds to decomposition of the ECG signal in its various participating frequencies, then elimination (filtering) of noisy frequencies. This type of processing requires a tremendous number of operations to perform, as the used filters are of narrow bandwidth type. The computational complexity could be reduced in such approach by the technique of sampling rate decimation (e.g. [1], [2]

Design considerations

In this study, we use the Douglas–Peucker algorithm [8] for the ECG curve simplification. This algorithm is known to be the most accurate curve simplification procedure at selecting perceptually dominant points of a given curve [9]. With regard to this quality, the algorithm possesses an interesting O(n log₂(n)) temporal complexity order. Besides, this algorithm is one of the most widely used algorithms in cartography [10]. In the ECG case, the accordingly computed DPs include peak (PIC) and

Algorithm main steps

The used approach is based on the thesis that input ECG signal (ECG_in) is composed of an underlying “true ECG” denoted hereinafter: Zero-Baseline-Wander ECG (ECG_ZBLW), the baseline deviation with respect to the isoelectric level, denoted hereinafter: BaselineW, and other high frequency noise, denoted hereinafter: Noise. This corresponds to the equation: $EC G_{in} = EC G_{ZBLW} + BaseLineW + Noise$ Since, in this study, large magnitude high frequency noise is not considered, setting: $EC G_{out} = EC G_{ZBLW} + Noise$ yields: $EC G$

Simulated baseline wander

In order to assess a baseline correction algorithm, one would need to acquire an ECG segment possessing an isoelectric level baseline, add a known signal simulating one or a combination of the above cited baseline deviation types, compute the baseline deviation from the (artificially) contaminated ECG and finally compute the error associated to the baseline extraction method. This evaluation scheme had been adopted by previous authors for baseline deviation extraction [2] and other ECG related

Method evaluation

The conducted experiments show the proposed method ability to correct most important baseline deviations of noisy ECG records, in a piecewise linear fashion. In particular, for half synthesized ECG experiments, both NRMSE for gold standard ECG and added baseline wander show the capacity of the developed algorithm to eliminate most added noise. Two observations can be issued from Table 2. First, the NMRSE_g value is almost constant. This is due to using the same minimal inter-DPs distance (57

Future plans

In this study, we suggested a piecewise linear baseline correction method for ECG signal, as a preprocessing phase to later phases, such as modeling, feature recognition and analysis. The method uses the Douglas–Peucker algorithm for curve simplification. The study targeted many ECG ubiquitous types of disturbances: baseline drift due to respiration, amplitude modulation due to perspiration, abrupt potential change in case of electrode loose contact due to patient movement for example, and

Acknowledgments

The authors would like to thank the anonymous referees for their constructive comments and suggestions.

References (17)

Y. Ferdi et al.
Un filtre numérique basé sur la dérivation non-entière pour l’analyse du signal électrocardiographique
ITBM-RBM
(2000)
J.A. Van Alsté et al.
Removal of base-line wander and power-line interference from the ECG by an efficient FIR filter with a reduced number of taps
IEEE Trans. Biomed. Eng.
(1985)
L. Sörnmo
Time-varying digital filtering of ECG baseline wander
Med. Biol. Eng. Comput.
(1993)
C.R. Meyer et al.
Electrocardiogram baseline estimation and removal using cubic splines and state space computation
Comp. Biomed. Res.
(1977)
Q. Ji et al.
Breakpoint detection using covariance propagation
IEEE Trans. Pattern Anal. Mach. Intell.
(1998)
G. Fernandez et al.
A new plant image segmentation algorithm
(1995)
H. Imai et al.
An efficient encoding method for electrocardiography using Spline functions
Syst. Comput. Jpn.
(1985)
C. Li et al.
Detection of ECG characteristic points using wavelet transforms
IEEE Trans. Biomed. Eng.
(1995)

There are more references available in the full text version of this article.

Cited by (27)

Development of ECG beat segmentation method by combining lowpass filter and irregular R-R interval checkup strategy
2010, Expert Systems with Applications
We have developed a long-term cardiorespiratory sensor system that includes a wearable sensor probe with adaptive hardware filters and data processing algorithms (Choi and Jiang, 2006, Choi and Jiang, 2008). However, the data processing algorithm proposed for the R–R interval (RRI) information extraction did not work well in the case of ECG signals with baseline shifts or muscle artifacts. Furthermore, many false ECG beats were extracted due to a weak decision-making scheme. Then, those false beats produced irregular RRI information and erroneous heart rate variability results. Modification of data processing algorithm was strongly needed. Therefore, this work presented an efficient ECG beat segmentation method using an irregular RRI checkup strategy into five sequential RRI patterns. This algorithm was comprised of signal processing stage and ECG beat detector stage. The signal processing included the wavelet denoising, the baseline shift elimination by 20 Hz lowpass filter and the envelope curve extraction by a single degree of freedom analytical model. The ECG beat detector included the candidate ECG beat detection and segmentation by one threshold and by irregular RRI checkup strategy, respectively. In particular, four abnormal RRI patterns were proposed to find out false ECG beats. The MIT-BIH arrhythmia database was selected as the dataset for testing the proposed algorithm. The proposed irregular RRI checkup strategy estimated 5463 beats to the suspected false beats and succeeded in segmenting 96.19% (5255 beats) of them. The performance results showed that our algorithm had very good results such as the detection error of 0.54%, sensitivity of 99.66% and positive predictivity of 99.80%. Furthermore, our algorithm showed very high accuracy as the mean time error between the beat annotations of the database and our obtained beat occurence times was 7.75 ms.
Reduced data similarity-based matching for time series patterns alignment
2010, Pattern Recognition Letters
Citation Excerpt :
There are also other sources causing local disturbances in the measured ECG samples: High frequency local noise due to the local electromyography (EMG), mainly due to perspiration, and the baseline wander which is mostly due to interference of low frequency respiration signal with the true ECG. If the non-natural perturbation is possible to deal with, for instance through different proposed techniques for ECG noise removal, e.g. (Boucheham et al., 2005), the local natural variability of the heart cycles is a persistent aspect of the problem. To deal with these difficulties and hence take confident decisions, we came to the conclusion that long enough time series were necessary.
We propose a similarity-based matching technique for the purpose of quasi-periodic time series patterns alignment. The method is based on combination of two previously published works: a modified version of the Douglas–Peucker line simplification algorithm (DPSimp) for data reduction in time series, and SEA for pattern matching of quasi-periodic time series. The previously developed SEA method was shown to be more efficient than the very popular DTW technique. The aim of the obtained ASEAL method (Approximate Shape Exchange ALgorithm) is reduction of the space and time necessary to accomplish alignments comparable to those of the SEA method. The study shows the effectiveness of the proposed ASEAL method on ECG signals taken from the Massachusetts Institute of Technology – Beth Israel Hospital (MIT–BIH) database in terms of the correlation factor and alignment quality, for savings up to 90% in used samples and processing time reduction up to 97% with respect to those of SEA. Particularly, the method is able to deal with very complex alignment situations (magnitude/time axis shift/scaling, local variabilities, difference in length, phase shift, arbitrary number of periods) in the context of quasi-periodic time series. Among other possible applications, the proposed ASEAL method is a novel step toward resolution of the ‘person identification using ECG’ problem.
A novel application of sample entropy to the electrocardiogram of atrial fibrillation
2010, Nonlinear Analysis: Real World Applications
Citation Excerpt :
The ECG recordings were preprocessed in order to improve later analysis. First, baseline wander was removed making use of bidirectional high pass filtering with a 0.5 Hz cut-off frequency [14,15]. Second, high-frequency noise was reduced with eight-order bidirectional IIR Chebyshev low-pass filtering, whose cut-off frequency was 70 Hz [16,17].
Nowadays, several non-linear regularity estimators have been successfully applied to invasive atrial electrograms in order to characterize the atrial electrical activity organization during atrial fibrillation (AF). This arrhythmia is the most common encountered in clinical practice, accounting for approximately one-third of all the hospitalizations for cardiac rhythm disturbances. However, from a clinical point of view, it would be more desired to evaluate atrial activity (AA) organization from surface electrocardiographic (ECG) recordings, since they can be obtained easily and cheaply and the risks associated with invasive recordings could be avoided. In this work, Sample Entropy (SampEn) is proposed to assess the organization degree of the AA obtained from surface ECGs. To this respect, a reliable and non-invasive organization estimator would allow the prediction of spontaneous AF termination, since invasive studies have shown more organized electrical activity signals during the preceding instants of AF termination. The proposed method computed SampEn over the AA obtained from TQ segments, free of QRST complexes, and was validated with a database containing a training set of 20 AF recordings, with known termination properties, and a test set of 30 recordings. A simulation study showed that patients with heart rates of 130 bpm and above must be handled with care, because TQ intervals could be considerably reduced (<50 ms). As an overall result, spontaneous AF termination in 90% of the learning and test recordings was correctly predicted through this novel approach. As a conclusion, this work introduces the application of a non-linear regularity index able to assess significative differences in AA organization from surface ECG recordings during AF.
Surface ECG organization analysis to predict paroxysmal atrial fibrillation termination
2009, Computers in Biology and Medicine
Citation Excerpt :
These ECG recordings were preprocessed in order to improve later analysis. Firstly, baseline wander was removed making use of bidirectional high pass filtering with 0.5 Hz cut-off frequency [19,20]. Secondly, high frequency noise was reduced with an eight-order bidirectional IIR Chebyshev low pass filtering, whose cut-off frequency was 70 Hz [21,22].
The aim of this work is to predict non-invasively if an AF episode terminates spontaneously or not by analyzing the increase of atrial activity organization prior to paroxysmal atrial fibrillation (PAF) termination. Sample entropy was selected as non-linear organization index. Synthetic PAF signals were used to evaluate the notable impact of noise in AA organization estimation. Three strategies to reduce noise, ventricular residues and enhance the atrial activity main features were proposed. The best prediction results were obtained through main atrial wave (MAW) organization estimation. The MAW can be considered as the fundamental waveform associated to the AA. The 92% of the terminating and non-terminating analyzed PAF episodes were correctly classified. Thereby, it can be concluded that the MAW non-linear analysis from the surface ECG is a reliable and useful tool to predict spontaneous PAF termination.
Noise-tolerant electrocardiogram beat classification based on higher order statistics of subband components
2009, Artificial Intelligence in Medicine
This paper presents a noise-tolerant electrocardiogram (ECG) beat classification method based on higher order statistics (HOS) of subband components.
Five levels of discrete wavelet transform (DWT) were applied to decompose the signal into six subband components. Higher order statistics proceeded to calculate four sets of HOS features from the three midband components, which together with three RR interval-related features constructed the primary feature set. A feature selection algorithm based on correlation coefficient and Fisher discriminality was then exploited to eliminate redundant features from the primary feature set. A feedforward backpropagation neural network (FFBNN) was employed as the classifier. Two sample selection strategies and four categories of noise artifacts were utilized to justify the capacity of the method.
More than 97.5% discrimination rate was achieved, no matter which of the two sampling selection strategies was used. By using the feature selection method, the feature dimension can be readily reduced from 30 to 18 with negligible decrease in accuracy. Compared with other method in the literature, the proposed method improves the sensitivities of most beat types, resulting in an elevated average accuracy. The proposed method is tolerant to environmental noises; as high as 91% accuracies were retained even when contaminated with serious noises, 10 dB signal-to-noise ration (SNR), of different kinds.
The results demonstrate the effectiveness and noise-tolerant capacities of the proposed method in ECG beat classification.
Matching of quasi-periodic time series patterns by exchange of block-sorting signatures
2008, Pattern Recognition Letters
We propose a novel method for quasi-periodic time series patterns matching, through signature exchange between the two patterns. The signature is obtained through sorting of the time series on magnitude. The advantage is that the difficult task of comparing the two patterns can be easily performed as a result of this exchange: The original time series is compared (point to point matching) to the reconstructed time series obtained through the reverse process, using the other time series signature. The matching is such that periods in one time series are put into correspondence with periods in the other time series, even if the time series is of different basic patterns and/or different lengths. The method is simple to implement and requires no parameters. It was compared to the very appreciated DTW algorithm on execution time, space and accuracy. Due to the quasi-periodic nature of the electrocardiogram, the tests were performed on ECG traces, selected from the Massachusetts Institute of Technology – Beth Israel Hospital (MITBIH) public database. Results show that the proposed method outperforms DTW on all aspects. This suggests that our method could be a good alternative to the classical DTW technique for quasi-periodic signals comparison. Specific applications are foreseen for our method: Novelty detection and person identification using ECG.

View all citing articles on Scopus

View full text

Piecewise linear correction of ECG baseline wander: a curve simplification approach

Summary

Introduction

Section snippets

Background

Design considerations

Algorithm main steps

Simulated baseline wander

Method evaluation

Future plans

Acknowledgments

ITBM-RBM

Removal of base-line wander and power-line interference from the ECG by an efficient FIR filter with a reduced number of taps

IEEE Trans. Biomed. Eng.

Time-varying digital filtering of ECG baseline wander

Med. Biol. Eng. Comput.

Electrocardiogram baseline estimation and removal using cubic splines and state space computation

Comp. Biomed. Res.

Breakpoint detection using covariance propagation

IEEE Trans. Pattern Anal. Mach. Intell.

A new plant image segmentation algorithm

An efficient encoding method for electrocardiography using Spline functions

Syst. Comput. Jpn.

Detection of ECG characteristic points using wavelet transforms

IEEE Trans. Biomed. Eng.