Robust Heart Sound Segmentation and Detection using biased Cramer-Rao Lower Bound Estimation and Variance Algorithm

Cardiac disorders are the most common infirmity suffered by the human race. Medical research states that worldwide almost 30% of the deaths are caused because of cardiovascular diseases. Even though many procedures such as Echocardiography, Carotid pulse and Electrocardiography are available, the conventional approach followed for detecting abnormalities are carried by hearing to the heart rhythms. This proposed study enables the health care physicians to diagnose pathologies of highly non-stationary heart sounds efficiently by using the second-order statistics of the signal. In this paper, a preprocessing methodology is used for removing noise caused by the lower frequency components and to focus only on the primary components S1 & S2, which are further feature extracted by using principal component analysis (PCA). A proposed variance algorithm (VA) is developed to identify the boundary locations of heart sounds and segment the featured signals into series of cardiac cycles. Further, we developed a modified variance algorithm (MVA) using biased Cramer-Rao lower bound (CRLB) to estimate the heart sound (HS) signals that exhibit minimum variance and extract finer boundary locations of the signal components which helps in identifying S1 & S2 positions. When compared with PCA, VA and Shannon energy (SE) methods for the same dataset, the MVA scheme exhibits very low normalized mean square error (NMSE) in extracting boundary locations by -71.60±0.25 dB and achieves high cardiac cycle segmentation ratio of 97.78±0.98 %. A brief analysis of the results showed that the proposed MVA scheme using biased CRLB exhibits 97.4±1.2% accuracy in identifying S1 and S2 heart sounds.