1.1 Overview of PCG signals
A phonocardiogram or PCG is a graphical interpretation of the cardiac sounds produced by the heart. Phonocardiography is the study of the sounds acquired using a standard tool such as an electronic stethoscope. Phonocardiograms (or heart sounds) acquired during cardiac auscultation contain bio-acoustic information related to the proper operation of the heart.
Stethoscope was first invented by R.T.H. Laënnec 12 in the year 1819. A perforated wooden cylinder was used to transmit the human heart sound from the patient’s chest to the physician’s ear. This new device helped Laënnec to diagnose diseases such as tuberculosis at an earlier stage than was previously possible. Later, this monaural stethoscope was modified into the binaural type stethoscope consisting of two flexible rubber tubes that attaches the chest piece to spring-connected metal tubes with earpieces.
The present day stethoscope uses a dual bell-shaped, open-ended chest piece, which transmits low-pitched sounds well, and the flat diaphragm based chest piece that detects sounds of higher frequency. Both types of chest piece are arranged so that they can be rapidly interchanged by turning a valve.
Heart sounds gives medical information regarding the health of the human beings. It also serves many purposes. It has recently been used as a biometric tool, for education purpose, for long term monitoring of patient health and telemetry. Heart sound is usually considered as a valid biomedical tool alternative to finger print or face recognition tool. Also various types of phonocardiogram instruments are currently in use for long time monitoring of patients’ health. Continuous time monitoring of the health of a patient helps the doctors to understand the cause of the disease better in real time and updates the doctor with changes.
Fig. 1.1 shows PCG signal with heart sound components S1, S2, S3 and S4.
First Heart Sound (S1): The first heart sound (S1) is caused by the closure of the mitral and tricuspid valves at the start of ventricular systole. The mitral component (M1) occurs slightly before the tricuspid component (T1) usually 20-30ms after M1 sound 15. The S1 is a low-pitch sound with longer duration 14. The intensity of the M1 sound is much higher than the T1 sound intensity due to the abrupt rise in left ventricular pressure. The first sound must be evaluated with its quality, intensity, and degree of splitting 13- 18. A decrease in the intensity of S1 is associated with myocardial depression, ventricular septal defect, and acute aortic regurgitation. The splitting of S1 sound is usually 60ms in patients with right bundle branch block (RBBB), or ventricular tachycardia or premature ventricular contraction (PVC) 14.
Second Heart Sound (S2): Usually at the end of the systole, the second heart sound (S2) is caused by the closure of the aortic and pulmonic valves. The S2 is shorter and slightly higher pitched sound. The S2 sound has frequency components in the range 10-400 Hz and the duration in the range 50-150ms 9. The S2 is composed of aortic closure sound (A2) and the pulmonic closure sound (P2). They last for less than 50ms 9. The delay between the closure of the aortic and pulmonic valves results in a split S2 sound. The sound S2 is evaluated based on the presence and degree of respiratory splitting and the relative intensities of A2 and P2. The amplitude and frequencies of A2 sound is slightly higher than the P2 sound 3. The sound interval at split widens on inspiration and narrows on expiration. The time interval between the A2 and P2 components is an indicator of aortic blood pressure 8. The various pathologies related to split S2 are pulmonic stenosis, RBBB, left bundle branch block (LBBB), atrial septal defect and right ventricular failure. In normal cases, the S1-S2 interval (systole) is shorter than the S2-S1 interval (diastole) 30- 40.
Third Heart Sound (S3): During diastolic period after 100-150ms of the S2 the third heart sound (S3) is produced by the sudden deceleration of blood flow within the ventricles. The S3 includes significant low amplitude-frequency components as compared to the first and second heart sounds. The S3 has 30-90 Hz frequency components with sound duration of 70 ± 15ms during diastole period 43, 44. S3 often occurs in patients with impaired myocardial reserve 16. The clinical studies show that the S3 can provide clinical information about hemodynamic and systolic dysfunction, and evaluation of patients with congestive heart failure 41- 48. The auscultation of S3 in adults is abnormal related with heart failure.
Fourth Heart Sound (S4): The fourth heart sound (S4) is caused by the contraction of atria resulting in forcing of blood into the distended ventricle. The S4 is a low-pitched sound that occurs before the first heart sound. The S4 sound arises from low-frequency vibrations with frequency of 20 to 30Hz. It is present in patients with diminished left ventricular compliance 13- 18. The clinical studies show that the diastolic heart sounds combining with electrocardiogram (ECG) may improve the non-invasive diagnosis of myocardial ischemia.
Heart Murmurs and Other Pathological Sounds
Murmurs are caused by turbulence flow of blood or vibration in tissues. Valvular dysfunctions results in pathological murmurs. Murmurs may be systolic, diastolic or continuous during every systole and diastole. Heart murmurs are organised into various categories by the timing (early, mid, late, or pan), intensity, duration, pitch (low, medium, or high), quality (blowing, harsh, rumbling or musical), and shape configuration crescendo (increasing intensity), decrescendo (decreasing intensity), crescendo-decrescendo (increasing then immediate decreasing intensity) 6, 14- 18. The pitch and intensity depends on the velocity of blood flow that produces the murmur. The timing of a murmur helps in accurate diagnosis of diseases.
The systolic murmurs can be separated into groups namely the early systolic murmur of acute mitral regurgitation and tricuspid regurgitation; the mid-systolic murmurs of aortic stenosis (AS), pulmonic stenosis (PS), hypertrophic obstructive cardiomyopathy (HOCM) and atrial septal defects (ASD); the late-systolic murmurs of mitral valve prolapse (MVP); and the holosystolic (or pansystolic) murmurs of the mitral regurgitation (MR), tricuspid regurgitation (TR), and ventricular septal defects (VSD). The diastolic murmurs include groups such as aortic and pulmonic regurgitation (early diastolic), and mitral or tricuspid stenosis (mid-late diastolic). The murmur of a patent ductus arteriosus (PDA) and systemic arterio-venous fistulae (AVF) is continuous throughout systole and diastole 11- 17, 49- 68. Fig. 1.2 shows the different types of systolic and diastolic heart murmurs. The click and snap sounds are associated with valves opening of the semilunar valves and the mitral and tricuspid valves. The clicks and snaps are associated with distinctive features of some heart defects.
Clinical PCG Parameters for CVD Diagnosis
In clinical studies, the specific heart sound indexes are measured for evaluating heart functions of subjects, maternal, foetal and infants with various physiological and pathological conditions. The heart sound parameters are: the cardiac contractility change trend (CCCT) (the increase of the S1 amplitude after exercising with respect to the S1 amplitude recorded at rest) 10; the amplitude of S1 29; the ratio of S1 amplitude to S2 amplitude (S1/S2) 10, 28; the ratio of the amplitude of tricuspid sound to the amplitude of the mitral sound (T1/M1); the ratio of S3 amplitude to S2 amplitude (S3/S2); the ratio of diastolic to systolic duration (D/S) 25, 10, 27; S1 localization 7; the duration, energy of instantaneous frequencies (EIFs) and splits of the aortic (A2) and pulmonic (P2) valve components 1, 3, the heart rate (HR) 25, 27; the duration and frequencies of S3 and S4 sounds, and the timing (location), configuration (shape), loudness (intensity), spectral content, duration of murmurs. Although most modern digital stethoscope can amplify, play, display and record heart sound signals in real time, automatic and quantitative measurement of heart sound parameters is very important for accurate and effective diagnosis of various cardiac diseases and disorders.
Challenges of Automated Diagnosis
The major challenge faced during the recording of the heart sounds is that the problem of acoustic noise component hinders the detection of the milder heart sounds. So separation of noise from heart sound becomes important. Separation of noise makes the heart sound robust and suitable for further processing. Pre-processing of heart sounds is used to assess the signal quality in terms of performance metrics such as Signal-to-Noise Ratio (SNR) and Segmental-Signal-to-Noise Ratio (SSNR). Pre-processing removes baseline changes and high frequency noises. Pre-processing can be used to extract relevant features.
Identification of heart sound components in noise free heart sound is important to understand the normality of the heart as well as working of individual valves in the heart. The process of identification of heart sound involves segmentation of the heart sound, prominently normal S1-S2 sounds, murmurs, clicks, and snaps. Segmentation is used to delineate the start and the end of each phase of the heart beat-S1, systole, S2, diastole.
Once the heart sound is segmented, the sound can further be classified as normal or abnormal sound by means of the classification process by identifying features in heart sound. Classification is done by comparing the various features of heart sound with that of the sounds in the reference database. In the process, we map the features for each segmented phase of the beat to the unknown phase or sound or the entire recording of the pathology.
Organisation of the Thesis
Chapter 1 describes the introductory concepts of phonocardiogram signal, their advantages in clinical decision analysis along with the problem statement and proposed solution. Chapter 1 also describes the heart sounds both normal and pathological in detail. The chapter discusses the clinical parameters for CVD diagnosis. The challenges for automated diagnosis is also described here in this chapter.
Chapter 2 describes the literature survey with focus on existing methods of heart sound analysis with respect to Pre-processing, segmentation and Classification of heart sounds. This section also focusses on the database used for heart sound analysis. Gaps and Limitations of the existing methods are also discussed here. The chapter also includes identification of the problem and objectives of the proposed work. There is a special focus on the methodology used in the proposed work.
Chapter 3 discusses the popular heart sound segmentation algorithms namely Homomorphic Filtering Segmentation and Segmentation using Mel-Scaled Wavelet Transform. The chapter proposes a new method of Heart Sound Segmentation using the Event Synchronous Method. There is a focus on the comparison between the proposed and existing methods in literature in terms of the obtained results.
Chapter 4 describes the de-noising procedure for PCG signals using time frequency techniques. There is a special focus on the time frequency block threshold method. A comparative study with wavelet de-noising and Time frequency soft threshold and overlapping group shrinkage algorithm is also described in this chapter.
Chapter 5 describes the extraction of loudness features and unsupervised classification of heart sounds with the state of the art classifiers namely K-Means, Fuzzy C-Means and GMM classifiers.
Chapter 6 describes the conclusion and the future direction of the thesis.