Signal processing algorithms for heart sound analysis

It is critical to improve the ability of early diagnosis and confirmation of cardiovascular disease due to its increasing incidence. As one of the body's most critical physiological signals, the heart sound signal contains a large amount of pathological information about the function of various parts of the heart such as the atria, ventricles, major blood vessels, cardiovascular vessels, and various valves. Therefore, heart sound detection is essential to understand the state of the heart, and an irreplaceable clinical value compared with ECG detection. Aiming on improving the ability of early diagnosis and confirmation of cardiovascular disease, the primary objective for this project is developing a robust algorithm for analyzing the heart sound accurately. For the propose of this project, the signals are first pre-processed by down-sampling, filtering and normalization before feature selection and feature extraction. The output coming from the first stage is used to pass through the SVM model to be classified into two categories, normal and abnormal. In this process, ROC curve, accuracy, specificity, and sensitivity are applied to assess the performance of the model. After thousands of tests, five features vector shows the best performance, whose accuracy is relatively high at 92.8%, the corresponding specificity and sensitivity are at 95% and 88.85% respectively. Besides, 4-features vector with the accuracy at 90.5%, the corresponding specificity and sensitivity are at 91.9% and 88.25% respectively also can be chosen in some sense. This shows that the algorithm presented in this report has high accuracy and good prospects.Master of Science (Signal Processing