Recognition Algorithm Of Heart Sound Based On Neural Networks

Heart sound is one of the most important physiology signals. It can provide valuable information concerning the integrity and function of the heart valves and also on the hemodynamics of heart and possesses important theory and application values in the examination of the condition of the cardiovascular system and in the diagnoses of the cardiovascular diseases. Relative to ECG used universally in the clinic, heart sound signal possesses itself characteristics and advantages. Firstly, it is very easy to record heart sound; secondly, it is more sensitive to some cardiovascular diseases than ECG; finally, heart sound is more suitable to analyze the force variability of heart than ECG, and is the foundation to analyze cardiac contractility change trend. Recognition and definition of the principal components of heart sound are the foundation of heart sound analysis and application. Since the complexity of heart sound signal, many recognition algorithms of heart sound, which are put forward to recognize heart sound signal without the reference signal, are complicate and require a lot of computation and have poor accuracy. So in this paper a recognition algorithm of heart sound based on probabilistic neural networks(PNN) is proposed to improve the accuracy of heart sound recognition. Firstly, heart sound signal is pre-processed by using wavelet transform; secondly, heart sound envelope is extracted by using normalized average Shannon energy transform; thirdly, the peaks of S1 and S2 are picked up and detected based on the extracted envelope of heart sound signal; fourthly, the eigenvectors of heart sound signal are defined and are imported to PNN; finally, S1 and S2 are recognized based on PNN. Furthermore, heart sound signals with high heart rate are also recognized and high accuracy is achieved.All of 551 peaks are recognized by using PNN, which come from nine different kinds of heart sound signal, including 45 heart sound samples. The result shows that the accuracy of recognition of normal heart sound is 95.1% and that of heart sound samples, which include split S1, split S2, S3, and S4, is 100%. And the accuracy of recognition of heart sound samples accompanied with systolic murmur and diastolic murmur is 92.1% and 94.7%, respectively. And the accuracy of recognition of heart sound samples including additional sound is 91.3%. After heart sound signal is recognized correctly, the characteristic parameters of heart sound signal are defined. This can be as a basis for further analysis of heart sound.