Lung Sound Denoising with Adaptive Noise Cancellation of Heart Sounds on a Raspberry Pi-Powered Stethoscope

Lung auscultation offers a quick and non-invasive diagnosis of respiratory conditions. However, background and heart noises may influence lung sound clarity and may reduce diagnosis accuracy. Heart noises are particularly challenging to separate from lung sounds due to overlapping frequencies. Therefore, this research proposes a method of reducing heart sounds through an LMS adaptive noise cancellation filter with Hilbert Transform and peak thresholding to estimate a reference heart sound signal, within a Raspberry Pi 4B-based digital stethoscope. The lung sound denoising algorithm starts with a Type 1 Chebyshev Bandpass Filter for background noise reduction, followed by wavelet threshold denoising using SureShrink adaptive thresholding, for finer noise reduction. The LMS adaptive noise cancellation filter was then applied to reduce heart sounds and ending with applying the same bandpass filter to smoothen the signal. Performance evaluation of the lung sound denoising algorithm demonstrated significant lung sound clarity improvement, with most of the background noises greatly reduced, achieving an average heart peak reduction of 87.5<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, a good LMS filtering performance with a Pearson correlation coefficient of 98.9% between estimated and replicated reference signals and a large reduction in magnitude within the heart sound frequency range was observed in an FFT. The system's robust signal processing capabilities make it a promising tool for a more accurate respiratory diagnosis and has great future potential to include machine learning to diagnose respiratory diseases.