A robust method for heart sounds classification based on empirical mode decomposition adaptive reconstruction.

Abstract Background: Heart sound signal analysis is an important method for noninvasive diagnosis of cardiovascular disease. In addition, an effective heart sound classification is an important prerequisite for heart sound analysis. Therefore, the classification of heart sound is of significant value. Methods: In view of ambient noise and pathologic features of heart sounds are the main unavoidable interference in heart sounds classification. This paper presents a new method for heart sounds correction and classification based on empirical mode decomposition (EMD) adaptive reconstruction. Firstly, since the noise reduction effect of EMD depends on the selection of intrinsic mode function (IMF), the signal-to-noise ratio (SNR) as an evaluation index of noise reduction effect was taken, and the selection of IMF was optimized based on it. Secondly, in order to reduce the false screening of peak points, the traditional single threshold method was improved by introducing a second threshold. Afterwards, the peak points were screened by the two-step threshold method. Finally, a heart sound correction algorithm for correcting the false components was proposed. Based on this, the heart sound signals were classified by the time domain characteristics. Results: The experimental results show that the localization has a detection sensitivity (Se) of 97.32%, a correct detection rate (P) of 99.19% and algorithm evaluation score (F1) of 0.982. Moreover, the Se, P and F1 of component classification reach 94.93%, 96.27% and 0.956, respectively. Conclusion: The test results show that the heart sound correction algorithm can effectively distinguish the difference between heart sounds and murmurs, correct the errors caused by pathologic conditions, and accurately locate and classify heart sounds.