heart sound reduction from lung sound using ANN combined with DWT and EMD

Lung sound (LS) auscultation is a classic and effective medical diagnosis. However, it is often obscured by overlapping heart sounds (HS). It has been an important research topic to reduce this interference while considering the condition of preserving the contents of LS for pulmonary disease diagnosis applications. This study proposes a nonlinear adaptive noise cancellation method for reducing HSs in LS recordings with the help of artificial neural networks (ANN), discrete wavelet transform (DWT), and empirical mode decomposition (EMD). The model mentioned above is Then compared to a two-step method that utilizes HS localization and time-domain processing which is a fairly fundamental technique. Power spectral density (PSD) Analysis shows that combining time-frequency representation, oscillatory mode decomposition, and nonlinear modeling provides a robust reduction of HS from LSs and improved auscultation. This adaptive data-driven approach enhances respiratory sound diagnosis by suppressing obscuring heart noises in LS recordings without requiring an additional localization step. Furthermore, the spectrogram analysis as a qualitative metric and correlation coefficient as a quantitative metric indicates the LS signals perseverance conclusively.