Respiratory sound denoising using Empirical Mode Decomposition, Hurst analysis and Spectral Subtraction. (February 2021)

• Record Type: Journal Article
• Authors: Haider, Nishi Shahnaj
• Is Part Of: Biomedical signal processing and control. Volume 64(2021)
• Publication Date: 2021-02
• Subjects: ALE Adaptive Line Enhancer -- ANN Artificial Neural Network -- ANOVA Analysis of Variance -- ARMA Auto Regressive and Moving Average -- BSE Blind Source Extraction -- BSS Blind Source Separation -- COPD Chronic Obstructive Pulmonary Disease -- CC Cross Correlation -- CT Computed Tomography -- DFT Discrete Fourier Transform -- EMD Empirical Mode Decomposition -- FD Fractal Dimension -- FIR Finite Impulse Response -- FFT Fast Fourier Transform -- FPGA Field Programmable Gate Array -- HS Heart Sound -- ICF Instantaneous Cycle Frequency -- IMF Intrinsic Mode Function -- MRI Magnetic Resonance Imaging -- MSE Mean Squared Error -- MAE Mean Absolute Error -- NMF Nonnegative Matrix Factorization -- PET Positron Emission Tomography -- PSNR Peak Signal to Noise Ratio -- SNR Signal to Noise Ratio -- SDR Signal to Distortion Ratio -- SPECT Single Photon Emission Computed Tomography -- S-G Savitzky-Golay -- STFT Short-Time Fourier Transform -- WT Wavelet Transform; Chronic pulmonary disease -- Denoising -- Empirical Mode Decomposition -- Hurst analysis -- Respiratory sound -- Spectral subtraction; 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.2020.102313
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: The study introduced the use of Hurst analysis to sort out the problem of intrinsic mode functions selection in empirical mode decomposition method, due to the presence of heavy noise issues in captured lung sounds. Apart with this, the study have used spectral subtraction to bring more improvement in the lung sound denoising outcome. It is a novel approach to denoise lung sound using empirical mode decomposition, Hurst analysis and spectral subtraction. To test the efficacy of the proposed algorithm on the large data set, downloaded from various web sources. The data set used in study includes 60 respiratory sounds, i.e., 30 COPD lung sound and 30 healthy lung sound. The study validated the denoising operation on both normal as well as adventitious lung sound. The study have achieved a higher denoising performance with the signal to noise ratio of 27.32 dB and the peak signal to noise ratio of 43.23 dB. This achieved denoising result is considerably higher in comparison to the available studies. Abstract: Chronic pulmonary diseases, specifically Chronic Obstructive Pulmonary Disease (COPD), is in third position for causing deaths all over the globe. Misdiagnosis and higher health care cost is the reason behind the heavy loss of life every year. To detect such diseases, computerized respiratory sound based diagnosis is one of the non-invasive, economical, convenient and harmless procedures, which could be one of the solutions to this acute problem. But, this diagnostic method is often affected due to noise issues. This paper presents a new method to denoise the respiratory sound using empirical mode decomposition (EMD), Hurst analysis and spectral subtraction. Using this algorithm, the highest signal to noise ratio (SNR) acquired is 27.32 dB and the peak signal to noise ratio (PSNR) is 43.23 dB. The proposed denoising algorithm could be a significant approach for assisting clinicians to make clear interpretations from the respiratory sound. The future work will be based on the elimination of heart sound noises from the respiratory sound signal.