Phonocardiogram (PCG) Signals Based Classification of Heart Abnormalities

Cardiovascular disease (CVD) is a serious illness that affects over the world. Early detection and prevention of CVD is thought to help reduce CVD mortality rates. There are many advanced technologies available for detecting CVD related symptoms, however they are only available in urban regions. Healthcare institutions in rural areas would only be equipped with minimal diagnostic devices, and the primary investigate tool would be the stethoscopes. Therefore, this study focusses on the use of phonocardiogram (PCG) data as a means of detecting heart abnormalities through machine learning approach. This study would concentrate on the use of phonocardiogram signals to detect abnormalities such as murmur, extrasystole, which are frequently cited in literature as early indicator for numerous CVD related illness. A sixth order Butterworth filter with a frequency range of 25Hz – 900Hz was used to remove undesired signal followed by a zerophased digital filter. The signal was then segmented using average Shannon energy. Both morphological features as well as non-linear features derived from Hilbert-Huang Transform (HHT) was used to analyze the PCG signals. Analysis of Variance (ANOVA) was used to select the statistically significant features for classification. Support Vector Machine (SVM), Ensemble and K Nearest- Neighbor (KNN) classifiers were used to quantify the efficacy of the proposed methods in detecting heart abnormalities. Overall accuracy of 70.2 % was achieved using the proposed features and Ensemble classifier. The experimental outcome shows that the use of PCG data is a promising field for development of non-invasive early screening system for CVD and could potentially help to uplift the general healthcare of rural populations..

I. Introduction

As per the records of World Health Organization (WHO), cardiovascular disease (CVDs) or also known as heart disease is the leading cause of death. It is a dangerous illness and a worldwide invisible killer that claims over 17.8 million people in the year 2017 [1]. In Malaysia, heart disease is the main cause of mortality [2]. According to China’s annual report on cardiovascular health and disease, CVDs mortality in rural and urban areas was 45.50% and 43.16% respectively, in 2016. Nonetheless, the CVDs mortality rate has risen [3]. There is no denying that early detection is critical in preventing this disease as well as reducing the mortality rate. Meanwhile, Electrocardiogram (ECG) and cardiac Magnetic Resonance Imaging (MRI) are commonly used in hospitals and clinics to identify this disease. However, these methods have not been influential due to their high cost and lack of portability. Conversely, previous researchersh has established that a phonocardiogram (PCG) signal can be utilized for this disorder. Phonocardiography is a non-invasive, quick and low-cost diagnostic technique that can be used for detecting and recording heart sound. Heart sounds is generally reorganized into four components, which is labelled as S1 which corresponds to first normal sound; S2 representing second heart sound; S3 which describes third heart sound for abnormalities; and S4 for fourth abnormal heart sound. The two common normal heart sounds are S1 and S2, namely a lub and a dub respectively [3]. In contrast, the S3 and S4 are abnormal heart sound components that use to signal cardiac failure during the diastolic period [4]. Most studies in the fields of PCG signals have only investigated the two primary components, S1 and S2 due to the onset of ventricular contraction and the end of ventricular systole respectively [5].