Separation and localisation of heart sound artefacts from respiratory data by adaptive selection of Eigen triples in singular spectrum analysis

Pulmonary auscultation is a vastly using diagnosis method over centuries. Together with the breath sound, there is a possibility of hearing heart sound, since both sounds are originated from the human chest. For the electronic analysis of the breath sound, separation of heart sound (HS) is important. In this paper, the separation of HS is achieved by using a modified Singular Spectrum Analysis (SSA) method, by introducing a provision for adaptive selection of SSA parameters. The advantage is Eigen triple grouping in the reconstruction stage of SSA is adaptive, that reduces the human effort. The performance of the new method is evaluated using synthetically mixed data and the real respiratory data and compared the results with the Advanced Line Enhancer (ALE) method which is an established single channel adaptive method. This method can also be useful for localizing the HS interferences in respiratory data, in some heart sound cancellation technique, where the localization is a fundamental preprocessing step. The comparative results suggest that the proposed method is more suitable for both separation and localization of heart sounds than the original ALE.