Heart Sound Classification Method Based on Complete Empirical Mode Decomposition with Adaptive Noise Permutation Entropy

Abstract

Abstract This paper proposes a method of complete empirical modal decomposition with adaptive noise (CEEMDAN) arrangement entropy as a characteristic vector of heart sound signal and support vector machine (SVM) as a classifier to classify heart sounds. Firstly, PCG is decomposed into a few intrinsic mode functions (IMF) from high frequency to low frequency with CEEMDAN. Secondly, this method uses the correlation coefficient, energy factor and signal-to-noise ratio of the IMF and the original signal to optimize IMFs for Hilbert transform getting instantaneous frequency. Making use of the instantaneous frequency calculates arrangement entropy of each IMF. These arrangement entropy values form a feature vector. Finally, the extracted features were classified by help of support vector machine (SVM) with the mark of normal and abnormal heart sounds. The 100 heart sound samples from the 2016 PhysioNet /CinC Challenge were classified between normal and abnormal, and experimental results show that this method can effectively improve the recognition accuracy. The classification results of support vector machine reach 87%, and are better than Fisher discrimination methods.