Localization of Phonocardiogram Signals Using Multi-level Threshold and Support Vector Machine

Cardiovascular abnormalities are one of the major causes of death in the world. The rate of death due to cardiovascular diseases can be reduced by diagnosing heart vessel blockage in early stage. Auscultation is a method in which digital stethoscopes are used to hear and record heart sounds (i.e. phonocardiograms) for the assessment of heart functionality. Correct diagnosis of phonocardiogram signals requires accurate detection of S1 and S2 states in the signal. To locate the S1 and S2 this paper introduces the concept of peak detection and presents an effective strategy employing multiple thresholds which results in improved localization of S1 and S2. In the first stage, phonocardiogram signal is divided into small segments and features are extracted for each small segment. Afterwards, support vector machine (SVM) classifier is trained to automatically detect the states. Physionet 2016 challenge dataset is used for the training and testing of the developed method. The segmentation accuracy with single-threshold method is 81% which improves to 87% with the incorporation of the developed multi-level threshold segmentation method.