Abstract

Successful lung sound extraction from lung auscultations requires precise filters. Lung auscultations include heart and lung sounds. Therefore, the filter design should identify heart sound characteristics. These include periodicity, pitch, frequency, amplitude, etc. Auscultation acquisition sensors are inefficient, so these signals have noise. To get a clean auscultation signal, ambient noise, periodic noise, DC offsets, baseline errors, harmonic noise, etc. must be addressed separately. Data cleaning must maintain a high peak-signal-to-noise ratio without removing desirable signal information. Heart sound patterns are extracted from input auscultations using filter banks on these signals. After heart sound extraction, the output sound sample should only contain lung sounds. Still, heart sound harmonics corrupt output lung sound due to similarities between heart and lung sound patterns. This lowers lung-sound extraction quality and automatic lung-based disease identification efficiency levels. This article proposes a 2-level multi-ensemble filtering model with 43 filters that analyze 15 other parameters to denoise and extract lung sound from lung auscultations to eliminate these drawbacks. Combining LMS, NLMS, and RLS is used for denoising process. For heart sound identification, an ensemble of Savitzky-Golay, FIR equiripple, Butterworth, Chebyshev, Elliptic, and wavelet filters is used. This selective combination of filters improves PSNR by 20% compared to sole filter performance, while signal entropy, crest factor, root mean squared error, kurtosis, etc. also improves for different scenarios.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call