Review of Heart Sound Analyses from Phonocardiogram Records

Abstract

Cardiovascular diseases are considered as one of the most common causes of death worldwide. Well-beings of people in the risk groups are monitored by  various state-of-the-art tools in clinics and home-care units. Phonocardiograph is one of the them which captures sounds
coming from the heart and gives high-quality graphical records (i.e., Phonocardiogram, PCG)
of them for examination of pathologies. PCG records have been studied and interpreted in order
to localize heart sound segments and classify abnormalities for decades. Moreover, there have
been competitions for heart sound classification and researchers have developed successful
solutions based on signal processing and machine learning approaches. Main steps of those
studies are grouped as preprocessing, segmentation, feature extraction and classification. In this
study we present a survey of proposed methods and used datasets. The features used in the
literature are listed as time, frequency and time-frequency domains. Performances of different
studies are presented and compared. From this perspective, it is concluded that there is still
room for automated heart sound analysis. Larger open access PCG databases are required for
testing state-of-the-art machine learning methods.