<title>Sensor system for heart sound biomonitor</title>

Abstract

Heart sounds can be utilised more efficiently by medical doctors when they are displayed visually, rather than through a conventional stethoscope. A system whereby a digital stethoscope interfaces directly to a PC will be described along with signal processing algorithms adopted. The sensor is based on a noise cancellation microphone, with a 450 Hz bandwidth and is sampled at 2250 samples per second with 12-bit resolution. Further to this, we discuss for comparison a piezo-based sensor with a 1 kHz bandwidth. A major problem is that the recording of the heart sound into these devices is subject to unwanted background noise, which can override the heart sound and results in a poor visual representation. This noise originates from various sources such as skin contact with the stethoscope diaphragm, lung sounds (patient breathing), and other surrounding sounds such as speech. Furthermore we demonstrate a solution using 'wavelet denoising.' The wavelet transform is used because of the similarity between the shape of wavelets and the time-domain shape of a heartbeat sound. Thus coding of the waveform into the wavelet domain is achieved with relatively few wavelet coefficients, in contrast to the many Fourier components that would result from conventional decomposition. We show that the background noise can be dramatically reduced by a thresholding operation in the wavelet domain. The principle is that the background noise codes into many small broadband wavelet coefficients that can be removed without significant degradation of the signal of interest. Also explored is the use of a phase space representation of the heart sound to exploit different properties of the signal. ᭧

Automated Pediatric Cardiac Auscultation

“…Fourier discovered that any periodic function could be expressed as an infinite sum of periodic complex exponential functions [45,44]. This property of periodic functions was later generalized to non-periodic functions and then to (both periodic and non-periodic) discrete time functions.…”

“…This method uses the property of wavelets to expose sharp discontinuities in a signal, meaning that noise can be readily revealed and hence can be removed by thresholding certain components of the wavelet decomposition. This is a very powerful concept because signals, with energy concentrated in a small number of wavelet dimensions, will have coefficients that are relatively large compared to any other signal, which has its energy spread over a larger number of wavelet dimensions, present [44]. Therefore applying the thresholding operation to the decomposed coefficients will effectively remove any unwanted signal or noise, even if the instantaneous frequency spectra of the two signals overlap.…”

“…Hall et al [44] show that the background noise in heart sound recordings can be dramatically reduced by a thresholding operation in the wavelet domain. The principle on which wavelet de-noising is based is that the background noise is confined to many small broadband wavelet coefficients that can be removed without significant degradation of the signal of interest.…”

“…Multiple FT's using distinct windowing periods would be required to produce the constant Q decomposition offered by wavelets [44]. …”

Optimal wavelet denoising for smart biomonitor systems

“…15 It may be seen that the signal is much cleaner after being denoised in Figure 6, but this dramatic improvement is only because the signal has much additive noise. The Hubert Transform of a signal does not accentuate the noise in a signal and thus does not show such dramatic improval after denoising as is seen in phase space diagrams shown in Maple et al 5 The phase space diagrams take the derivative of a signal thus accentuating the noisy, high frequency, content. This fact is demonstrated in Figure 8.…”