Reverberant blind separation of heart and lung sounds using nonnegative matrix factorization and auxiliary function technique. (August 2021)

• Record Type: Journal Article
• Authors: Xie, Yuan; Xie, Kan; Yang, Qiyu; Xie, Shengli
• Is Part Of: Biomedical signal processing and control. Volume 69(2021)
• Publication Date: 2021-08
• Subjects: Blind source separation (BSS) -- Heart sound mixtures -- Lung sound mixtures -- Reverberation environment -- Nonnegative matrix factorization (NMF); Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102899
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Separation of heart and lung sounds from the observated mixtures has become a hot research topic in the prevention and diagnosis of heart and lung diseases. Especially, in real clinical situations, clinicians can perform auscultation by using the bioacoustics knowledge of sounds to diagnose the diseases of heart and lung. Therefore, to have an intimate knowledge of the current health status of patients, it is of great significance to develop intelligent auscultation methods to improve the effectiveness of hearing and assist clinicians in practice. In this paper, a novel blind source separation (BSS) method is proposed based on nonnegative matrix factorization (NMF) and auxiliary function technique for the separation of heart sound mixtures and lung sound mixtures in the reverberation environment. In the algorithm, a denoising pre-processing technique is firstly used to the noise reduction such that eliminating the influence of the additive white Gaussian noise. Moreover, the principal component analysis method is applied to reduce the dimension so that the number of source signals is equal to the number of sensors. Then, NMF is introduced as the source model to capture the spectral structure of each source. Afterwards, an auxiliary function is built as the cost function, the demixing matrix and the NMF model parameters are updated by building the fast and stable update rules. Finally, the separated source scales are restored to obtain the reconstructed sources. Experimental results according to the BSS performance of heart sound mixtures and lung sound mixtures demonstrate that the proposed method is superior to the conventional BSS methods, especially for the highly reverberant environment.