Research on Gaussian Mixture Model-based heart sound component partitioning

Heart sound component partitioning plays an important role in analyzing its pattern for diagnosing cardiovascular disease (CVD), especially for CVD with the characteristics of heart sound split. However, the traditional Gaussian regression model (GRM)-based heart sound partition method suffers from time complexity, leading to difficulty in capturing the important information of the heart sound signal and an inability to diagnose CVD online. To efficiently reduce the high complexity in GRM, a novel Density-Weighted Resampling and lustering-based Gaussian Mixture Model (DWRC-GMM) is proposed to substitute the original GRM, including the approaches on how to transform the envelope's probability distribution into representative sample points through density-weighted resampling, and how to adopt Expectation-Maximization (EM) for building DWRC-GMM. The efficiency of DWRC-GMM is evaluated by the goodness of fit, time complexity, and compared to GRM approaches by conducting experiments on heart sound datasets.