Empirical Mode Decomposition, S-method and Steepest Gradient-based Reconstruction Algorithm for Denoising Lung Sound

Abstract

For increased identification of pulmonary disorder, the lung sound-based system must be free of all intervening disturbances. The heart sound is major noise, which complicates the lung sound signal processing and influences the pathological decisions. This paper investigates empirical mode decomposition (EMD) and the S-method steepest gradient-based reconstruction algorithm for cancellation of heart sound noise from lung sound. The method was evaluated using four statistical evaluation metrics: the root mean square error (RMSE), the signal to noise ratio (SNR), the normalized root mean square error (NRMSE), and the percentage of correlation coefficient (percent CC). The maximum SNR of 39.65 dB, the lowest RMSE of 0.00051, the lowest NRMSE value of 0.00022, and the highest percent CC of 98.45 was obtained. Hence the algorithm successfully eliminates the heart sound noise from lung sounds for accurate detection of lung disorder. This work will be enhanced by variable mode decomposition (VMD) approach in future.