Abnormal heart sound recognition based on instantaneous frequency and singular value decomposition

It is urgent to study the aided diagnosis of congenital heart disease (CHD) through heart sound due to the high risk of this disease especially in remote regions. In this study, a method based on instantaneous frequency using empirical mode decomposition (EMD) and singular value decomposition (SVD) was proposed to recognize abnormal heart sound. This method include three steps: Firstly, FIR filters were used to filter heart sound into different frequency bands, and then each band signal was decomposed as Intrinsic Mode Function (IMF) using EMD. Secondly, calculating Instantaneous phase using Hilbert transformation, and features were extracted based on singular value decomposition. Finally, support vector machine (SVM) and neural network (NN) were used to classify features to recognize abnormal heart sound. The experiment results show the 25-250 Hz and 250-1000 Hz has higher recognition rate 85.7% and 84.6% compared to other frequency band. The result confirms that there is a potential in heart sounds for the diagnosis of CHD in different frequency bands, but more clinical data is needed in further to verify the recognition effect of different frequency bands.