Optimal wavelet denoising for smart bio-monitor systems

Abstract

Future smart-systems promise many benefits for biomedicaldiagnostics. The ideal is for simple portable systems that displayand interpret information from smart integrated probes or MEMS-baseddevices. In this paper, we will discuss a step towards this visionwith a heart bio-monitor case study. An electronic stethoscope isused to record heart soundsand the problem of extracting noise from the signal is addressed viathe use of wavelets and averaging. In our example of heartbeat analysis,phonocardiograms (PCGs) have many advantages in that they may bereplayed and analysed for spectral and frequency information. Manysources of noise may pollute a PCG including foetal breath sounds ifthe subject is pregnant, lung and breath sounds, environmental noiseand noise from contact between the recording device and the skin.Wavelets can be employed to denoise the PCG. The signal isdecomposed by a discrete wavelet transform. Due to the efficientdecomposition of heart signals, their wavelet coefficients tend to bemuch larger than those due to noise. Thus, coefficients below acertain level are regarded as noise and are thresholded out. Thesignal can then be reconstructed without significant loss ofinformation in the signal. The questions that this study attempts toanswer are which wavelet families, levels of decomposition, andthresholding techniques best remove the noise in a PCG. The use ofaveraging in combination with wavelet denoising is alsoaddressed. Possible applications of the Hilbert Transform toheart sound analysis are discussed.