Study on umbilical artery blood signal analysis and classification

臍帯動脈血流信号解析と分類に関する研究

Study on umbilical artery blood signal analysis and classification

Degree Grantors Yamaguchi University

In China, there are about 800,000 congenital diseases among 20 million newborns, of which nearly 200,000 fetuses have serious defects or diseases. The birth of these sick fetuses brings serious economic burden and social problems to the family and even the society. It is therefore important to carry out early fetal monitoring in order to detect fetal defects and diseases as early as possible. Umbilical artery blood signals contain important information about fetal growth and development, reflecting various problems during pregnancy, such as intrauterine g rowth r etardationetardation（IUGR）, hypoxia and maternal hypertension, which can be determined by umbilical artery blood signals. Therefore, the analysis of umbilical artery blood signals is important for prenatal monitoring and fetal health status diagnosis.
The acoustics pectral parameter method is a conventional technique for analyzing the umbilical artery blood signals and consists of three parameters that serve as clinical diagnostic criteria: resistance index (RI), pulsesatility index (PI) and maximum systolic/end diastolic umbilical flow velocity (S/D). However, these parameters ignore phase properties of the signal, such as phase delay, phase frequency and phase mode, and focus only on the fundamental statistical parameters of blood velocity, s uch as maximum, minimum and mean values. This may lead to clinical misdiagnosis.
Umbilical artery blood signals have complicated structures and nonlinear characteristics in addition to changes in signal amplitude. This paper presents a comprehensive new approach for characteristics parameter extraction and classification of umbilical artery blood signals using fractal theory and Chaos theory in order to handle these complex structures and nonlinear properties of the signal. First, by focusing on the fract al characteristics of umbilical artery blood signals, the fractal dimension (BD) and the correlation dimension (CD) are obtained to verified that BD is positively correlated with the gestational week and CD is effective in discriminating normal from abnormal. Next, we obtain the maximum Lyapunov exponent (MLE) of the chaotic characteristics of umbilical artery blood time series, and verified its effectiveness in distinguishing normal signals from abnormal signals. Finally, a diagnostic model is proposed b y applying particle swarm optimization support vector machine (PSO SVM) to the conventional feature parameters (RI, PI, S/D) and newly obtained parameters (BD, CD, MLE) to classify and diagnose the umbilical blood signals in the four statuses (normal, oligohydramnios, umbilical cord around neck, fetal malposition).
This doctoral dissertation consists of 6 chapters.
Chapter 1 introduces the background and means of umbilical artery blood study as well as reviewing the current re search situation. The outline o f this dissertation is also given.
In chapter 2, the fundamentals of fetal hemodynamics are described. The clinical significance and normal reference values of umbilical artery blood signal parameters are outlined. Details of the umbilical artery signal acquisition equipment, data classification and acquisition process are explained.
In chapter 3，the fractal dimension box counting method (BD) and the correlation dimension (CD) are used to investigate the nonlinear characteristics of the umbilical artery blood signals based on fractal theory. First, the BD of the umbilical artery blood signals is calculated and the fractal characteristics of the signals are analyzed. Results show a positive relationship between the fractal dimension of umbilical artery blood signals and gestational weeks. A bnormal and normal umbilical artery signals are then classified into abnormal group and n ormal group. T h e Grassberg P rocaccia algorithm (GP algorithm) is used to calculate and analyze the CD of the two groups. T he overall CD of normal umbilical artery blood signals is greater than that of abnormal signal s. CD is significantly better at discriminating the normality of the umbilical artery blood signal compared to conventional parameters. Furthermore, t he Hurst exponent of umbilical artery blood signal is calculated and analyzed by Lo method. The results show that umbilical artery blood signal belong s to non sta tionary signal and show obvious “1/f fluctuation” characteristics.
In chapter 4，c haotic phase space diagram method and m aximum L yapunov e xponent (MLE) are used to determine the chaotic characteristics of umbilical artery blood signals from qualitative and quantitative perspectives. The attractor reconstruction of umbilica l artery blood signals is performed in t hree d imension (3D) and t wo d imension (2D) phase space. The results show that the chaotic phase diagram of the time series for abnormal umbilical artery signals show a jumbled “ball of wool” state and the chaotic “shape” appears to converge. Application of the r eceiver o perating c haracteristic (ROC) curve to the obtained maximum Lyapunov exponent (MLE) shows that the rate of discrimination of normality of the umbilical artery blood signal is significantly better than the conventional feature parameters.
In chapter 5，an artificial intelligent classifier is proposed to classify the four states of umbilical artery blood signals (normal, oligohydramnios, umbilical cord around neck and fetal malposition). The support vector machine (SVM) classifying method is constructed based on the conventional parameters, S/D, PI and RI. The particle swarm optimisation support vector machine (PSO SVM) classifier is also constructed using the fractal dimension (BD), correlation dimension ( CD) and maximum L yapunov exponent (MLE) derived in Chapters 3 and 4 as feature parameters. The results of the classification tests show that the PSO SVM classifier is more accurate , confirming the usefulness and effectiveness of the proposed classification method.
In Chapter 6, summary of this dissertation and future work are described.

Creators YU KAIJUN