Real time segmentation of heart sounds

[摘要] ENGLISH ABSTRACT: The poor state of the healthcare system in South Africa has resulted in unacceptablehigh levels of infant mortality. Congenital heart disease is one ofthe main contributions to these high rates of mortality, with the cost of treatmentand the availability of specialists being the driving factors. Computeraided auscultation is a technological solution to assist with the diagnosis ofthe disease. In its current form, computer aided auscultation is unsuitable forcontinuous patient monitoring.The aim of this thesis is to develop an algorithm that will allow the existingmethods of computer aided auscultation to work in real time so they canbe used in patient monitoring. Existing methods of identifying the first andsecond heart sound are limited to offline processing. The algorithm developedin this thesis uses the correlation of the time-frequency coefficients of individualheart sounds to generate a feature vector for each heart sound that can beused to separate the sounds into different groups. To test the performanceof the algorithm, 230 heart sounds from normal patients were first manuallysegmented and then processed with the algorithm. The noise sensitivity of thealgorithm was also tested using generated heart sounds. Finally, the real timecapability of the algorithm was tested.The testing against sounds for normal patients resulted in a 84.2 % accuracyand an 84.4% hit rate. The synthetic testing showed the system startsto perform badly with a signal to noise ratio lower than -10db. The real timetesting of the system showed that the algorithm is fast enough to be used ina real time environment. This thesis concludes that proposed algorithm issuitable for the detection of the first and second heart sounds in real time.

[发布机构] Stellenbosch University