Elsevier

Information Sciences

Volume 467, October 2018, Pages 99-114
Information Sciences

Computer-aided diagnosis of atrial fibrillation based on ECG Signals: A review

https://doi.org/10.1016/j.ins.2018.07.063Get rights and content

Highlights

  • Existing AF detection techniques are discussed.

  • Building blocks of CADx system are described.

  • Different features explored by researchers are presented.

  • State-of-art CADx system for AF are highlighted.

Abstract

Arrhythmia is a type of disorder that affects the pattern and rate of the heartbeat. Among the various arrhythmia conditions, atrial fibrillation (AF) is the most prevalent. AF is associated with a chaotic, and frequently fast, heartbeat. Moreover, AF increases the risk of cardioembolic stroke and other heart-related problems such as heart failure. Thus, it is necessary to screen for AF and receive proper treatment before the condition progresses. To date, electrocardiogram (ECG) feature analysis is the gold standard for the diagnosis of AF. However, because it is time-varying, AF ECG signals are difficult to interpret. The ECG signals are often contaminated with noise. Further, manual interpretation of ECG signals may be subjective, time-consuming, and susceptible to inter-observer variabilities. Various computer-aided diagnosis (CADx) methods have been proposed to remedy these shortcomings. In this paper, different CADx systems developed by researchers are discussed. Also, the potentials of the CADx system are highlighted.

Introduction

Arrhythmia, or heart rhythm disorder, is a common manifestation of heart disease [3], [74]. An arrhythmia is said to occur if the heart beats too early, late, slowly, quickly, or at irregular intervals. This happens because there is a dysfunction in the electrical conduction pattern of the heart. Among the many different forms of arrhythmias, atrial fibrillation (AF) is the most common. AF induces varying and often very fast or slow heartbeats. In AF, the electrical impulses in the atria (upper chambers of the heart) are disorganized and are unsynchronized or out of coordination with the ventricles (lower chambers of the heart). Examples of signals from a normal and a heart with AF are seen in Figure 1. It can be observed that the electrical impulses in the normal heart proceed in an orderly flow, but the electrical impulses in the AF heart are chaotic. In AF, due to the irregular heart rhythm, the heart may pump over 150 times per minute. In a healthy heart, the heart pumps approximately 60 to 100 times per minute. During AF, the multiple ectopic foci in the atria discharge at an extremely fast rate resulting in multiple irregular stimulations. As the atria cannot respond mechanically to the disorganized electrical signal stimulation, the atria end up twitching rather than contracting properly.

AF is categorized into paroxysmal, persistent, or permanent forms (longstanding persistent AF) [76]. In paroxysmal AF, the ECG signals exhibit irregular heartbeats (AF) that return or spontaneously cardiovert to regular heartbeats (sinus rhythm) within a short time (minutes to a few days). In persistent AF, the AF returns to sinus rhythm only if the patient receives electrical or pharmacological cardioversion treatment to abolish the AF. Lastly, in permanent AF, there is no intention to cardiovert the AF, and the irregular heartbeat is left untreated.

As reported by the World Heart Federation, AF has become a global health interest as the prevalence and associated mortality have grown exponentially in the past decade [77]. According to a study published in Circulation [23], it was reported that 33.5 million people globally have AF. It was also mentioned in the study that in 2005, 3 million Americans are afflicted by AF and the number is expected to increase to 8 million by 2050. Even though AF per se is not a life-threatening condition, it is a critical medical condition that calls for medical attention. This is because the progression of AF can lead to stroke, with its potential for long-term paralysis and disability, congestive heart failure and even death [24]. Typically, the irregular heartbeats in AF do not manifest any symptom until a later stage, when cardioversion success becomes more limited. Consequently, early diagnosis of AF is crucial to halt the further advancement of AF to other heart and stroke complications.

The diagnostic approaches for AF include clinical examination, electrocardiogram (ECG), Holter monitor, an event monitor, stress test, and echocardiogram [95]. Clinical examination to detect AF comprises auscultation, and manual pulse assessment [95]. However, clinical methods are neither adequately sensitive nor specific to secure an accurate diagnosis. The ECG machine is commonly used to confirm the diagnosis. The ECG is a tool that records the electrical activity of the heart. Healthcare professionals often assess the cardiac condition of the patient based on the ECG signals. Similarly, the Holter monitor, event monitor, and stress test produce ECG signals that are recorded over an extended period of time, or when the subject is under physical stress (for example, when running). On the other hand, the echocardiogram utilizes sound waves to construct images of the heart [52]. It can provide an overall visualization of the heart structure. The images are displayed in real-time on a video monitor and these images are recorded for analysis. Still, the gold standard to test for AF is ECG analysis [62].

Section snippets

The morphology of an AF ECG signal

The ECG waveform contains essential information regarding the health conditions of the heart [10]. Figure 2 compares a typical healthy versus an AF ECG signal. A normal ECG waveform is made up of the P wave, QRS complex, and T wave. However, in an AF ECG signal, the P wave is not present but instead, replaced by many and inconsistent fibrillatory waves. Nonetheless, the QRS complex is still present in the waveform [15]. Hence, cardiologists and other healthcare professionals make their

Computer-aided diagnosis system

The CADx encompasses four main processes: pre-processing of ECG signals, extraction of distinctive features from the signals, selection of the most and highly significant features, and lastly, feed the selected features into the classifier for classification [8] (see Figure 3).

  • (i)

    Input ECG signal:

    The ECG signals may be acquired from opensource databases such as the MIT-BIH atrial fibrillation (afdb) and MIT-BIH arrhythmia (mitdb) [37].

  • (ii)

    Pre-processing (Pre):

    The typical frequency of P and T wave

Discussion

Table 4 summarizes a list of published studies of CADx algorithms that could objectively differentiate AF ECG signals from the ECG signals of a healthy person or the ECG signals obtained from other heart conditions. Numerous techniques have been proposed to develop the CADx system. The time-frequency domain and nonlinear analysis approaches were commonly employed. Based upon the studies in Table 4, it can be concluded that the combination of nonlinear and time-frequency domain features can

Future development

Common triggers for AF are usually caused by coronary artery disease (CAD), heart failure (HF), high blood pressure (HBP), and myocardial infarction (MI) [73]. Thus, it is necessary for the CADx system to also be able to detect these cardiac abnormalities. From literature studies, there are existing CADx models developed to diagnose different heart conditions, namely CAD [4], [99], CHF [96], and MI [6], using ECG signals. Therefore, the model will be able to provide preliminary pronouncement of

Conclusion

AF is an irregular and often fast heart rhythm that may lead to serious health problems. Early detection of AF will avert further progression, and with appropriate preventive treatment, reduce the likelihood of patients developing stroke and other heart-related complications. Therefore, it is vital to attend to screening for AF effectively and efficiently. This review paper provided a comprehensive overview of existing CADx systems designed to aid cardiologists in their diagnoses of AF. The

References (110)

  • M. García et al.

    Application of the relative wavelet energy to heart rate independent detection of atrial fibrillation

    Comput. Methods Progr. Biomed.

    (2016)
  • C. Jutten et al.

    Blind separation of sources, part 1: an adaptive algorithm based on neuromimetic architecture

    Signal Process.

    (1991)
  • KangY.D. et al.

    Algorithm for the detection of congestive heart failure index

    J. Mech. Med. Biol.

    (2017)
  • KimT.S. et al.

    Role of echocardiography in atrial fibrillation

    J. Cardiovasc. Ultrasound

    (2011)
  • A. Krizhevsky et al.

    ImageNet classification with deep convolutional neural networks

    Commun. ACM

    (2017)
  • R.J. Martis et al.

    Computer-aided diagnosis of atrial arrhythmia using dimensionality reduction methods on transform domain representation

    Biomed. Signal Process. Control

    (2014)
  • R.J. Martis et al.

    Automated detection of atrial fibrillation using Bayesian paradigm

    Knowl. Based Syst.

    (2013)
  • R.J. Martis et al.

    Application of higher order statistics for atrial arrhythmia classification

    Biomed. Signal Process. Control

    (2013)
  • R.J. Martis et al.

    Automated detection of atrial flutter and fibrillation using ECG signals in wavelet framework

    J. Mech. Med. Biol.

    (2012)
  • G.P. Nason et al.

    The Stationary Wavelet Transform and Some Statistical Applications

    The Stationary Wavelet Transform and Some Statistical Applications

    (1995)
  • J. Pan et al.

    A real-time QRS detection algorithm

    IEEE Trans. Biomed. Eng.

    (1985)
  • J. Sadowsky

    The continuous wavelet transform: a tool for signal investigation and understanding

    John Hopkins APL Tech Digest

    (1994)
  • F. Sufi et al.

    Diagnosis of cardiovascular abnormalities from compressed ECG: a datamining-based approach

    IEEE Trans. Inf. Technol. Biomed.

    (2011)
  • K. Tateno et al.

    Automatic detection of atrial fibrillation using the coefficient of variation and density histogram of RR and ΔRR intervals

    Med. Biol. Eng. Comput.

    (2001)
  • U.R. Acharya et al.

    Automated characterization of arrhythmias using nonlinear features from tachycardia ECG beats

  • U.R. Acharya et al.

    Entropies for automated detection of coronary artery disease using ECG signals: a review

    Biocybern. Biomed. Eng.

    (2018)
  • U.R. Acharya et al.

    Advances in Cardiac Signal Processing

    (2007)
  • P.S. Addison

    Wavelet transforms and the ECG: A review

    Physiol. Meas.

    (2005)
  • N. Ahmed et al.

    Discrete cosine transform

    IEEE Trans.Comput.

    (1974)
  • E.A. Ashley et al.

    Cardiology Explained, From Chapter 3: Conquering the ECG

    (2004)
  • C. Bandt et al.

    Permutation entropy: a natural complexity measure for time series

    Phy. Rev. Lett.

    (2002)
  • R. Bellman

    A markovian decision process

    Indiana Univ. Math. J.

    (1957)
  • C.M. Bishop

    Pattern Recognition and Machine Learning

    (2006)
  • L. Breiman

    Random forest

    Mach. Learn.

    (2001)
  • L. Breiman et al.

    Classification and Regression Trees

    (1984)
  • S. Bochner et al.

    Fourier Transform

    (1949)
  • ChangC.H. et al.

    Cancellation of high-frequency noise in ECG signals using adaptive filter without external reference

  • ChughS.S. et al.

    Worldwide epidemiology of atrial fibrillation: a global burden of disease 2010 study

    Circulation

    (2014)
  • E.J. Ciaccio et al.

    Development of an automaton model rotational activity during atrial fibrillation

    Comput. Biol. Med.

    (2017)
  • T.M. Cover et al.

    Nearest neighbor pattern classification

    IEEE Trans. Inf. Theory

    (1967)
  • CuiX. et al.

    Automated detection of paroxysmal atrial fibrillation using an information-based similarity approach

    Entropy

    (2017)
  • S. Dash et al.

    Automatic real time detection of atrial fibrillation

    Ann. Biomed. Eng.

    (2009)
  • U. Desai et al.

    Diagnosis of multiclass tachycardia beats using recurrence quantification analysis and ensemble classifiers

    J. Mech. Med. Biol.

    (2016)
  • L.P. Devroye et al.

    Distribution-free performances bounds for potential function rules

    IEEE Trans. Inf. Theory

    (1979)
  • C.S. Dhir et al.

    Efficient feature selection based on information gain criterion for face recognition

  • R.O. Duda et al.

    Pattern Classification

    (2001)
  • J.P. Eckmann et al.

    Recurrence plots of dynamical systems

    Europhys. Lett.

    (1987)
  • R.A. Fisher

    Statistical Methods For Research Workers

    (1925)
  • R.A. Fisher, “The use of multiple measurements in taxonomic problems”, Ann. Eugen., vol. 7, no. 2, pp....
  • A.L. Goldberger et al.

    PhysioBank, PhysioToolkit, and PhyisoNet: Components of a new research resource for complex physiologic signals

    Circulation

    (2000)
  • Cited by (0)

    View full text