Elsevier

ISA Transactions

Volume 66, January 2017, Pages 362-375
ISA Transactions

QRS detection using adaptive filters: A comparative study

https://doi.org/10.1016/j.isatra.2016.09.023Get rights and content

Highlights

  • This paper presents an improved QRS detection algorithm which is based on adaptive filtering principle.

  • In comparative study, Variable-Leaky LMS, Variable Step-size LMS, Leaky-LMS, RLS and Fractional-LMS are novel combinations.

  • The leaky-LMS algorithm gives the best performance with sensitivity of 99.68% and positive predictivity of 99.84%.

  • According to study, performance of other LMS-variants gets affected due to low SNR, but Leaky-LMS still gives better results.

Abstract

Electrocardiogram (ECG) is one of the most important physiological signals of human body, which contains important clinical information about the heart. Monitoring of ECG signal is done through QRS detection. In this paper, an improved QRS detection algorithm, based on adaptive filtering principle, has been designed. Enumeration of the effectiveness of various LMS variants used in adaptive filtering based QRS detection algorithm has been done through fidelity parameters like sensitivity and positive predictivity. Whole family of LMS algorithm has been implemented for comparison. Sign-sign LMS, sign error LMS, basic LMS and normalized LMS are re-implemented, while variable leaky LMS, variable step-size LMS, leaky LMS, recursive least squares (RLS), and fractional LMS are novel combination presented in this paper. After analysis of the obtained results, performance of leaky-LMS algorithm is found to be the best with sensitivity, positive predictivity, and processing time of 99.68%, 99.84%, and 0.45 s respectively. Reported results are tested and evaluated over MIT/BIH arrhythmia database. Presented study also concludes that the performance of most of the variants gets affected due to low SNR but the Leaky LMS performs better even under heavy noise conditions.

Introduction

ECG is the electrical activity of heart, measured by placing electrodes at several positions on human body. ECG signal is a quasi-periodic wave that comprises of a few individual characteristic waves, given as P, Q, R, S, and T. These waves initiate with P wave that originates from the atrial depolarization of heart muscles; the Q, R, and S waves originate from the ventricular depolarization, and are together interpreted as the QRS complex. The T wave is due to the re-polarization of heart. This cycle repeats itself with the next P wave [1]. ECG signal monitoring and analysis is widely used to explore and identify various heart diseases like atrial and ventricular premature contractions, atrial fibrillations, bradicardia, and tachycardia etc. [1]. Variation from normal sinus rhythm (standard heart cycle) are the symptoms of arrhythmia that represents some form of cardiovascular disease (CVD). The CVD has been identified as one of the dominant causes of deaths all over the world by World Health Organization (WHO) [2]. CVD is primary health concern in most of the countries, which spend huge expenditure on research and development of CVD monitoring and control equipments. Such equipments and devices rely on digital ECG signal processing methods. Digital ECG also plays an important role in compressing the ECG signal for tele-porting to tertiary health care centers under low bandwidth requirements. To enhance ECG monitoring for easy identification of abnormalities, automated and adaptive computerized methods have been developed and adopted [3], [4].

Mostly, computer and electrical engineering researches have contributed various methods and algorithms for ECG signal processing. Initial steps in these researches are preprocessing and detection of QRS complex. Generally, ECG preprocessing involves several analog and digital filtering methods [5], [6]. In [6], a review of ECG de-noising methods is presented. In [7], [8], importance of ECG signal processing in terms of better recognition of ST changes during myocardial ischemia is specified. The QRS complex detection is basically the detection of R peak (the highest peak); and then based on this location, the detection of Q and S peaks. The QRS detection methods developed so far are based on approaches of artificial neural network [9], signal derivatives [10], digital filtering [11], [12], linear data prediction [13], [14], transforms [15], [16], level-cross sampling [17], integrate and fire sampler [18], dynamic plosion index [19], and many more. Amongst them, signal derivatives method, the oldest one, proposed by Pan and Tompkins in 1985, is implemented in real-time ECG monitoring halter systems. At present, the research trend is in the direction of microprocessor based methods for storing purpose as well as for automated real-time ECG signal processing [20].

In comparison with methods based on signal derivatives and other approaches, the adaptive filtering based approach, which has been used in this paper, is advantageous for real time ECG signal processing as the ECG signal is quasi-periodic in nature, and its quality also degrades with various random artifacts (noises) during recording. The advantage of using adaptive filtering approach is that it estimates the underlying signal even in absence of prior information about statistical properties of the signal and noise [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], due to which, this approach is resistant to a sound degree to all types of noises and gives better detection results. This adaptive filtering based method of QRS detection also offers an advantage of multitasking such that the prediction error (PE), calculated only once, can be used for QRS detection as well as ECG compression purpose. With adaptive filtering artifacts due to formation of extra ventricular waveform during recording are also removed [21]. Adaptive filtering also performs better detection of late potentials [22]. Adaptive filtering is generally implemented through least mean squares (LMS) and recursive least squares (RLS) algorithms [23].

In this paper, an improved adaptive linear prediction based QRS detection algorithm, incorporating novel combination of LMS variants with adaptive filter, is proposed. The improved QRS detector through comparative analysis of various LMS variants for adaptive filtering yields an efficient QRS detection method that achieves good sensitivity and positive predictivity in comparison with other competing methods, and offers an advantage of low computational complexity and real-time implementable design. In its implementation, various LMS variants have been analyzed to select the best variant amongst all that gives optimum results with low computational complexity and processing time. The new LMS variants incorporated for adaptive filtering are the fractional LMS (FLMS), leaky-LMS (LLMS), variable-leaky LMS (VLLMS), variable step-size LMS (VSLMS), and recursive least squares (RLS). These newly incorporated variants have been compared with previously implemented variants of [14], which are given as the basic LMS, normalized LMS (NLMS), signed LMS (SLMS), and sign-sign LMS (SSLMS).

The proposed QRS detection methodology is based on the principle that a low order predictor can successfully predict the slow varying (steady) components of a signal but produces large PE for the transient parts [14]. Analogically, in ECG prediction, the PE corresponding to QRS complex regions is higher, which is in the form of impulses, whereas for other signal portions, the PE values are lower. These impulses at the places of QRS complexes are used as features for R peak detection, which is achieved by comparing the processed PE against adaptive threshold. The incorporation of new LMS variants for selecting the best performing LMS variant for QRS complex detection is the novelty of this paper.

The paper is organized in 5 sections. In Section 1, overview and importance of ECG signal processing is presented. Section 2 explores the principle of QRS detection through adaptive prediction, different variants of algorithms used for adaptive prediction and details of data set used. Results and discussion are presented in 3 ECG database, 4 Results respectively. In 5 Discussion, 6 Conclusion, final conclusion and future direction to extend this study have been presented.

Section snippets

Principle of QRS detection process

The process of QRS detection starts with mean subtraction of ECG signal after which adaptive filter also known as adaptive linear prediction (in context of ECG processing) is applied to get the instantaneous prediction error (PE), which undergoes Savitzky–Golay (SG) filtering for peak enhancement and noise suppression. It is basically the smoothening of a signal by its local approximation with polynomials in least square error sense. Further signal enhancement is provided with squaring and

ECG database

The performance of proposed QRS detection algorithm is evaluated on standard MIT-BIH Arrhythmia Database for a data-length of 60 s (21,600 samples). The database is obtained from Physionet Bank ATM collected from Boston׳s Beth Israel hospital (BIH) arrhythmia laboratory. The database contains two channel ECG recordings in digital format from 47 subjects that include 25 men and 22 women of age groups 32–89 and 23–89 respectively. Approximately 60% of the total subjects were inpatients. The

Optimized initialization of predictor-coefficients

The optimum filter order for comparison of all variants is empirically found to be 4, since a very low order predictor predicts even the low frequency components with high error, thus it produces indistinguishable features. On the other hand, a very high order predictor precisely predicts even the QRS complex portion, thus producing very less error. These indistinguishable and low amplitude features result in false positive and false negative beats respectively.

In all the variants, filter taps

LMS variants versus fidelity parameters

Every variant has its benefits and limitations in terms of detection results. Detection results of variants FLMS, LMS and VLLMS deteriorate due to severe baseline wandering. For some variants, the performance also deteriorates due to the appearance of high amplitude sharp T-waveforms in ECG signal because of production of QRS equivalent features after preprocessing. They are misinterpreted as features due to QRS complexes. NLMS variant suffers from this demerit.

Noises due to Electromyogram

Conclusion

This paper presents an improved QRS detection system that works on the method of adaptive linear prediction. The improvement of QRS detection system is done by utilizing new variants of LMS algorithm; and a novel combination of few variants has been tested and analyzed. The performances of all variants are compared to find out the best variant to be used in adaptive filtering based QRS detection system. Analytically, the LLMS variant based QRS detection system achieves the best detection

Future scope

The positive predictivity (+P) of LLMS based algorithm can be further improved if its limitation of producing FP beats under conditions of heavy noises can be overcome by adopting some suitable noise removal method keeping the signal features unaffected. If the proposed algorithm is utilized in offline ECG processing (i.e. processing of stored ECG signals), then its sensitivity (Se) can further be improved by preventing FN beats by applying some search-back algorithm with reduced average

References (27)

  • M. Okada

    A digital filter for the QRS complex detection

    IEEE Trans Biomed Eng

    (1979)
  • V.X. Afonso et al.

    ECG beat detection using filter banks

    IEEE Trans Biomed Eng

    (1999)
  • K.P. Lin et al.

    QRS feature extraction using linear prediction

    IEEE Trans Biomed Eng

    (1989)
  • Cited by (0)

    View full text