Biography
Enrollment Date: 2013
Graduation Date：2014
Degree：M.S.
Defense Date：2014.05.26
Advisors：Hanjun Jiang
Department：Institute of Microelectronics，Tsinghua University
Title of Dissertation/Thesis：Denoising Strategy of Phonocardiogram Based on Novel Computerised Quality Assessment
Abstract:
Auscultation using a stethoscope is used to listen to the internal sounds of the body, including the heart, which is a key indicator of the body health condition . With the development of electronic stethoscopes, heart sounds can be recorded electronically and then analyzed on a computer. These recordings are called phonocardiograms (PCG). All
electronic stethoscopes however are relatively sensitive to noise and the recorded signals need to be denoised before any further processing.The denoising of phonocardiograms is a tough challenge with only little real indepth prior research. One major hinder is the lack of an objective method to evaluate and compare the denoising algorithms. To rate the algorithms, the quality of the signal before and after the application of the denoising algorithm needs to be compared. Almost all prior research was based on high quality heart signals with artificial additivenoise which was usually assumed to be Gaussian, and used the signal-to-noise ratio (SNR) as the measure of quality. This procedure is maintained because of the difficulty to tell apart the signal and the noise in a normal recording. It is insufficient however, since the noise in phonocardiograms originates from multiple, mostly unknown sources, which is extremely unpredictable and can’t be imitated with simple models. A method to evaluate the algorithms without having to add artificial noise is wanted. This entails the requirement of a new procedure to qualify the signals without the need for prior knowledge on the noise and signal individually.To tackle this problem, a novel method has been proposed to measure the quality of heart signals. Three novel methods are described and evaluated that make it possible to automatically assess a quality factor for these signals. The first method is based on the fundamental frequency of the envelope of the heart signal. This frequency corresponds to the heart rate of the signal. The second builds upon this fundamental and its harmonics. The last method exploits the typical characteristics of the correlation of a heart signal. The automatic signal qualification offers a lot of possibilities. First of all it provides the opportunity to evaluate algorithms in a more realistic context. Furthermore it can for example also give instant feedback to a user on the quality of the recording and on the confidence level of all derived calculations with affordable computation complexity. In addition, a general applicable strategy has been proposed to improve the quality of the recorded heart signals as much as possible. Popular preprocessing methods such as a bandpass filter and outlier removal and real denoising algorithms such as wavelet denoising and Savitzky-Golay are re-evaluated with the use of the proposed novel quality factor. Whereas the first three methods proof to be able to improve the signal, the effectiveness of Savitzky-Golay is invalidated. Eventually, a fixed composite denoising strategy is suggested, making use of newly designed outlier removal algorithm, a linear-phase FIR bandpass filter, and wavelet denoising using Coiflet waves. This strategy yields good results for regular adult heart sounds and is also applicable for fetal heart sounds. It could be implemented in wearable electrical stethoscopes for long-term heart monitoring purposes.