Signal Processing Driven Machine Learning Techniques for Cardiovascular Data Processing

Signal Processing Driven Machine Learning Techniques for Cardiovascular Data Processing features recent advances in machine learning coupled with new signal processing-based methods for cardiovascular data analysis. Topics in this book include machine learning methods such as supervised learning, unsupervised learning, semi-supervised learning, and meta-learning combined with different signal processing techniques such as multivariate data analysis, time-frequency analysis, multiscale analysis, and feature extraction techniques for the detection of cardiovascular diseases, heart valve disorders, hypertension, and activity monitoring using ECG, PPG, and PCG signals.

In addition, this book also includes the applications of digital signal processing (time-frequency analysis, multiscale decomposition, feature extraction, non-linear analysis, and transform domain methods), machine learning and deep learning (convolutional neural network (CNN), recurrent neural network (RNN), transformer and attention-based models, etc.) techniques for the analysis of cardiac signals. The interpretable machine learning and deep learning models combined with signal processing for cardiovascular data analysis are also covered.

• Provides details regarding the application of various signal processing and machine learning-based methods for cardiovascular signal analysis
• Covers methodologies as well as experimental results and studies
• Helps readers understand the use of different cardiac signals such as ECG, PCG, and PPG for the automated detection of heart ailments and other related biomedical applications

• Language: English
• Publisher: Academic Press
• Release date: Jun 12, 2024
• ISBN: 9780443141409

Book preview

Chapter 1: Introduction to cardiovascular signals and automated systems

Dhanhanjay Pachoria; Shaswati Dashb; Rajesh Kumar Tripathyb; Tapan Kumar Jaina    aDepartment of ECE, Indian Institute of Information Technology (IIIT), Nagpur, India

bDepartment of EEE, Birla Institute of Technology and Science, Pilani, Hyderabad, India

Abstract

Cardiovascular signals such as the electrocardiogram (ECG), phonocardiogram (PCG), and photoplethysmograph (PPG) provide valuable information for the automated and early detection of different heart diseases and other related biomedical applications. This chapter reviews different cardiovascular signals and the information associated with these signals for different biomedical signal-processing applications. The frequency content of each of these cardiac signals is also discussed. The steps used in the automated diagnosis system (ADS) for detecting cardiac and other diseases using ECG, PCG, and PPG signals are provided. The role of signal processing and machine learning techniques in developing ADS is discussed in this chapter.

Keywords

cardiovascular signals; automated diagnosis systems; machine learning; signal processing; biomedical applications

Chapter Outline

1.1  Heart conduction system and ECG signal

1.1.1  Features of ECG signals

1.1.2  Heart diseases and morphological changes in ECG signals

1.1.3  Automated disease diagnosis system using ECG

1.1.3.1  Recording of ECG signals

1.1.3.2  Preprocessing of ECG data

1.1.3.3  ECG feature extraction and selection

1.1.3.4  Machine learning and deep learning

1.2  Cardiac auscultation and PCG signal

1.2.1  Heart valve diseases and changes in PCG

1.2.2  Automated detection of HVDs using PCG

1.3  PPG signal and cardiorespiratory activity

1.3.1  Automated analysis of PPG signals

1.4  Future scope of cardiac data processing

1.5  Conclusion

References

1.1 Heart conduction system and ECG signal

The heart is mainly an autonomous organ, and its function is to provide oxygen-rich blood to the entire body [1]. It comprises four chambers, namely, the left atrium (LA), left ventricle (LV), right atrium (RA), and right ventricle (RV). Similarly, it also contains four valves, namely, the tricuspid valve (TV), mitral valve (MV), pulmonary valve (PV), and aortic valve (AV), which help to prevent the backward flow of blood [2]. The flow of blood happens in the heart in a unidirectional manner. Initially, the heart receives deoxygenated blood through the RA, which passes from the RA to the RV through the TV. The function of the RV is to pump the deoxygenated blood to the lungs for purification through the PV. Similarly, the LA collects the oxygen-rich blood from the lungs, which is passed to the LV. The function of the LV is to pump the oxygenated blood through the AV to the aorta, which further transfers the blood to the entire body [2]. The heart thus acts as a two-way pump for transferring blood. The heart's function is based on an electrical conduction path. The pacemaker cells, also known as cardiac myocytes, establish the conduction path. These pacemaker cells form the (a) sino-atrial (SA) node, (b) atrioventricular (AV) node, (c) His bundle, and (d) Purkinje fibers [3]. The firing rate of each pacemaker cell is different. The SA and AV nodes have firing rates of 60 to 100 beats per minute (bpm) and 40 to 60 bpm, respectively. Similarly, for the His bundle and Purkinje fibers, the firing rates vary between 20 and 45 bpm [4].

The electrocardiogram (ECG) is a graphical representation of the heart's electrical activity and captures information regarding the depolarization and repolarization of heart chambers. A typical lead I ECG signal for a normal sinus rhythm (NSR) is shown in Fig. 1.1(a). The ECG signal is collected from the PTB diagnostic 12-lead ECG database [5] [6]. It consists of clinical patterns such as the P-wave, QRS-complex, T-wave, and baseline. Features such as duration and amplitude of the clinical patterns (P-wave, QRS-complex, T-wave) are evaluated from the ECG signal [2]. These features are widely used in clinical studies for cardiovascular disease diagnosis and other biomedical applications [7] [8] [9]. Similarly, the Fourier spectrum of the ECG signal is shown in Fig. 1.1(b). It is evident that significant spectral energy is observed between 0.5 Hz and 45 Hz in the spectrum of the ECG signal. The time-domain and spectral characteristics of ECG signals vary for different heart diseases [10]. The details regarding the features of ECG signals are described in the following subsection.

Figure 1.1 (a) ECG signal with marked clinical patterns (the sampling frequency of the ECG signal is 1000 Hz). (b) The Fourier spectrum of the ECG signal.

1.1.1 Features of ECG signals

The P-wave represents the electrical activity regarding atrial depolarization [2]. Typically, the amplitude of the P-wave is 0.25 mV for an NSR. Similarly, the duration of the P-wave varies between 80 ms and 100 ms for a normal person. During atrial hypertrophy-based cardiac ailments, the ECG signal has a higher (amplitude more than 0.25 mV) and wider P-wave (duration longer than 100 ms) [2]. Similarly, abnormal and unordered P-waves are seen in the ECG signal during atrial fibrillation (AF) [2]. The QRS-complex reveals the electrical activity for the depolarization of the left and right ventricles of the heart. The QRS-complex duration varies between 80 ms and 100 ms for a normal heart rhythm in healthy individuals. The increase in the duration of QRS-complex beyond 100 ms indicates branch bundle block (BBB) [2]. The ST-segment in the ECG represents the interval between the end of ventricular depolarization and the beginning of ventricular repolarization. Typically, the slope of the ST-segment is used in clinical settings as an important biomarker for the diagnosis of myocardial infarction (MI). The T-wave in the ECG reveals the electrical activity for the repolarization of ventricles. For a healthy heart or NSR, the amplitude and duration of the T-wave are less than 0.5 mV and between 100 ms and 250 ms, respectively [2]. The change in the morphology of the T-wave is an important parameter for the diagnosis of MI and cardiomyopathy-based heart diseases [2].

1.1.2 Heart diseases and morphological changes in ECG signals

Various heart diseases are diagnosed based on morphological changes in the ECG signal [2]. These cardiac diseases include (a) MI, (b) BBB, (c) AF, and (d) ventricular tachycardia (VT) [10]. MI-based cardiac diseases mainly happen due to obstruction in one of the coronary arteries of the heart [11]. This disease progresses into three main stages: the ischemic stage, acute stage, and necrosis stage [8]. The ischemic stage corresponds to the reduction of blood flow in the coronary artery due to atherosclerosis plaque formation [12]. The inverse T-wave elevated ST-segment and abnormal Q-wave morphology are interpreted as the morphological changes in ECG signals due to MI-based heart disease [11]. The ECG signal for the MI case is depicted in Fig. 1.2(a). Similarly, the spectrum of MI pathology-based ECG signal is shown in Fig. 1.2(b). It can be observed that both temporal and spectral information of ECG data changes due to MI pathology as compared to the NSR case in Fig. 1.1(a). Similarly, BBB occurs due to a delay in the heart's conduction system [2]. This type of delay can occur during the depolarization of the heart's ventricles. The morphological changes such as a wider QRS-complex and an increased R-wave amplitude are observed in ECG signals due to BBB-based heart disease. AF is interpreted as an irregular and rapid atrial rhythm, and it can increase the chances of heart failure and MI [13]. The fibrillatory (F)-waves and varying RR intervals are seen in the ECG signal due to AF [14] [2]. The ECG signals for BBB and AF-based heart disease cases are shown in Fig. 1.2(c) and Fig. 1.2(e), respectively. The spectra of the ECG signals for BBB and AF cases are depicted in Fig. 1.2(d) and Fig. 1.2(f), respectively. It is evident that the morphologies of ECG data for AF and BBB differ from the NSR case. There are variations in the spectra of AF- and BBB-based ECG signals [10]. Similarly, VT occurs due to damage in the heart muscle and causes heart failure, or sudden cardiac death [15]. An abnormal QRS-complex with a duration longer than 0.14 s occurs in the ECG signal due to VT [2].

Figure 1.2 (a) ECG signal for MI. (b) The spectrum of MI pathology-based ECG signal. (c) ECG signal for BBB. (d) The spectrum of BBB pathology-based ECG signal. (e) ECG signal for AF. (f) The spectrum of AF pathology-based ECG signal.

1.1.3 Automated disease diagnosis system using ECG

The continuous monitoring of ECG data from subjects at the intensive care unit (ICU) and wearable devices generates a huge volume of data [16]. It is a time-consuming process for medical staff to manually investigate all morphological features of ECG data in diagnosing cardiac diseases [10]. The manual diagnosis procedure is prone to error, and hence the automated diagnosis system (ADS) is used to detect cardiac diseases from ECG data. The flow chart of the ADS is depicted in Fig. 1.3. It mainly consists of three important stages: (a) recording of ECG data, (b) preprocessing of ECG data, and (c) the use of signal processing-domain machine learning (ML)- and deep learning (DL)-based methods for automated detection of cardiac diseases [8].

Figure 1.3 Automated diagnosis system for heart disease detection using ECG signals.

1.1.3.1 Recording of ECG signals

The ECG data can be recorded in the resting state [2] and when the subject is moving [17]. The 12-lead ECG (with lead numbers I, II, III, aVR, aVL, aVF, V1, V2, V3, V4, V5, and V6) is used for the recording of resting-state ECG signals [11]. The 12-lead ECG data provide more information as the heart's electrical activity is viewed from distinct angles. For diagnosing MI, BBB, hypertrophy, and other cardiac arrhythmias, the 12-lead ECG recording is normally recommended by medical professionals [2]. Similarly, the single-channel-based ECG recording is performed in wearable devices for the daily monitoring of heart rate and morphological features of ECG data for different applications like detection of AF [18], stress monitoring [19], and human activity recognition [20]. Furthermore, the ECG signals recorded either in the resting state or from wearable devices are transmitted for telehealthcare monitoring and telemedicine applications