Distinguishing between murmurs in different heart valves using phonocardiogram analysis

Abstract

Providing non-invasive and continuous heart assessment for children is crucial in the detection of congenital and acquired heart diseases, especially for the countries having high birth rates. In each cardiac cycle, heart sounds are generated due to the opening and closing of the heart valves (aortic, mitral, pulmonary and tricuspid valves). Phonocardiogram (PCG) obtained through auscultation allows the assessment of these heart sounds, as well as any sub-audible sounds or murmurs. Hence, proper analysis of PCG can provide valuable information regarding impairments caused by congenital and acquired heart diseases. In this study, the relationship between heart murmurs, valve locations and different PCG features was investigated: (i) PCG recordings were analyzed using eight different feature groups and the most distinctive feature types in murmur detection were determined, (ii) binary classification models for any given valve location were trained to distinguish between murmur-present and murmur-absent recordings, (iii) for the first time, a multi-class classification framework was generated to distinguish between the murmurs present in different valves. In the binary task, all four models resulted in 100% performance, and in the multi-class task, the accuracy values for all valve locations were above 80%. Overall, it has been shown that the murmurs present in different heart valves indeed show different morphological characteristics and such differences could potentially be leveraged in the design of continuous monitoring systems to assist in clinical decisions. © 2023 IEEE.