A Novel Adaptive Method for Heart Sound Segmentation

The segmentation quality of heart sounds plays a vital role to achieve an accurate diagnosis of cardiovascular diseases and identification of states. However, due to the differences in cardiac structure under different types of heart diseases and physiological conditions, the process of segmenting heart sounds is affected in some extent. Therefore, to solve the problem and improve the generalizability of the common segmentation methods using fixed parameters, this paper proposes a novel short time modified Hilbert transform (STMHT)-based adaptive segmentation method (ASTMHT) to obtain higher segmentation quality by self-adjusting determined segmentation parameters process. The optimal segmentation parameters, when the heart sound cycle converged to the minimum, is adaptively determined to achieve better adaptation for differences in cardiac structure. The effectiveness of ASTMHT was verified through comparative validation analysis in various heart sound datasets (including quiet state, non-quiet state, and the Michigan heart sound dataset). The results showed ASTMHT outperforms the average Shannon energy (ASE) method and the STMHT method on three datasets, especially on the non-quiet state dataset, with segmentation accuracy improvements of 53.95% over the average Shannon energy and 76.57% over the STMHT.