Heart Sound Classification Based on Features of FBPSI

Cardiovascular disease triggers abnormal changes in the structure and function of the heart, leading to a decrease in the heart’s contraction and pumping ability, which seriously threatens the life and health of patients. Therefore, we propose a heart sound analysis method with frequency-balanced power spectral intensity characteristics to effectively distinguish normal and pathological heart sound signals. First, the heart sound signals were denoised and normalized to extract their frequency-balanced power spectral intensities. Then, a machine learning model is applied to classify and recognize heart sound signals by combining the cepstrum coefficients of Mel and Gamma frequencies. The experiments are validated using the Yaseen public dataset with the 2016 PhysioNet/CinC Challenge dataset. The ten-fold cross-validation results show that the proposed effective feature vector combined with the K nearest neighbor classifier has an accuracy of up to 98.30% on the five-classified Yaseen dataset and 90% on the two-classified 2016 dataset. It can provide an effective reference for the clinical prevention and diagnosis of cardiovascular diseases.