【作者】 唐晓；

【导师】 吴志健；

【作者基本信息】 武汉大学 ， 计算机软件与理论， 2013， 博士

【摘要】 近年来,随着转化医学的快速发展,以信息科学为技术手段,为疾病的预测、预防、诊断和治疗提供了帮助,使得基础医学与临床医学得到了较好的结合。转化医学的一个重要内容是计算机辅助诊断,将病人就诊过程中产生的大量不同类型的信息通过计算机信息处理,提供个体化治疗策略,从而提高对疾病的诊断效率。其中,如何选用适当的优化算法根据医学模型和临床检测数据,采用系统辨识方法,获取患者个体的医学参数,是必须的环节,也是核心问题之一。乙型肝炎是影响我国人民健康危害最严重的传染病之一,而传统的医学数据分析对每个患者的病情很难有一个总体上的把握,更不用说预测其病情的发展。随着医学技术的发展,人们尝试构建了乙肝病毒动力学模型(HBV动力学模型),通过对该模型中的关键参数进行辨识,并仿真出患者的各个重要指标,对患者的病情进行预测,并对用药后的效果可以做一个提前预测。这对医生治疗起到了很好的指导作用,且提前预测也可以使病者的病情在最短时间得到有效的治疗。而对该模型进行系统辨识是个非线性系统辨识问题,传统的系统辨识方法对这样的非线性系统辨识问题效果不是很好。目前的系统辨识多采用二次规划等解析算法,可辨识的参数少,收敛慢,对参数的初值依赖大。随着智能控制领域研究的不断发展,非线性程度也就越来越高,一些经典的方法就很难满足需要了。本文研究了小种群粒子群优化算法(SPPSO)在非线性系统辨识中的应用,将其应用在实际的医学系统辨识中。根据HBV动力学模型,利用临床检测的动态数据,为乙肝患者的诊断与治疗提供新的辅助手段,主要工作内容是：将SPPSO算法与其他算法在非线性系统辨识中的效果进行了比较,发现SPPSO在保持同样精度的情况下,具有更快的收敛速度,因此更能满足临床分析的需要。由于乙肝病毒动力学模型是非线性微分方程组,并且受到临床动态监测数据少的制约,一般的优化算法很难得到较好的结果。本文将SPPSO算法运用到求解HBV动力学模型的非线性系统辨识中,实现了HBV动力学模型的快速辨识。SPPSO作为一种全局优化算法,易于实现,且收敛速度快,计算效率高。在处理数据量较大的大规模的种群问题时可大大降低时间和资源的开销,因此在系统辨识特别是高度非线性系统中具有很大的意义。而这类复杂的非线性系统在医学系统中有具典型性,所以将该算法用于求解HBV动力学模型上,有很好的研究价值和实用价值。并且针对药物作用的滞后效应,将时滞参数作为待辨识的参数,采用时滞微分方程的求解算法,利用SPPSO算法实现了乙肝病毒时滞模型的辨识。进一步注意到医学参数的时变特点,利用正交多项式的线性组合近似时变参数,将无穷维问题转化为有限维问题,利用SPPSO算法实现了乙肝病毒时变模型的辨识。本文的研究为HBV动力学模型的参数辨识提供了一种新的方法,该方法可以推广到一般非线性系统的辨识中。并为医学测量提供了一种“软测量”的方法：对于医学上不能直接检测或者检测费用很高的变量,可以选取易于检测的”二次变量”建立起”二次变量”与待检测变量之间的”数学模型”,通过系统辨识,估计待检测的变量。这都是很有实际的研究价值和研究意义的。更多还原

【Abstract】 Translational medicine gets fast development in recent years. It combines the basic medicine and the clinical medicine with information technology for the disease prediction, prevention, diagnosis and treatment of help. An important aspect is the computer aided diagnosis. A large number of different types of information in the clinical provide individualized treatment strategies through the computer information processing, so as to improve the efficiency of the diagnosis of disease. It is a necessary link to estimate the medical parameters according to the medical model and clinical testing data and using the system identification method for individual patients. One of the important problems is to choose a appropriate optimization algorithm.Hepatitis B is one of the most serious infectious diseases that affect the people’s health of the world. However, the traditional medical analysis is very difficult to have a general grasp on each patient’s condition not to mention predict disease’s development. With development of medical technology, people are trying to build a dynamic model of hepatitis B virus (HBV dynamics model). Through the parameter identification the model simulation important indexes of Hepatitis B of patient. It may make a prediction about the progress of disease and pesticide effect. This result provides valuable information for doctors and shortens the time of therapy. But this is a nonlinear systems identification problem. Results of traditional system identification methods for nonlinear system identification problems are bad. Existing algorithms like quadratic programming method can identify parameter’s number is very limited and it has the limitations of stagnation and it is heavily dependent on initial values of the parameters. Current methods can only identify four parameters at most. With the continuous development of the area of intelligent control, the degree of nonlinearity becomes higher and higher.This dissertation studies the small population-based particle swarm optimization algorithm (SPPSO) in nonlinear system identification according to the dynamic model of Hepatitis B and the clinical testing dynamic data and provides a new auxiliary means. The main works are:The effectiveness of the genetic algorithm is compared to SPPSO in nonlinear system identification. SPPSO has been found to have faster convergence speed in keeping the accuracy. So it is more suitable to the need of the clinical analysis. SPPSO is an optimization technique for locating the global optimum. SPPSO is easy to realize, quick convergence and effective. It can greatly reduce the time and resource costs in the processing of large data quantity of large-scale population problem. So, in system identification, especially in highly nonlinear systems is more meaningful. And this kind of complex system is typical in medical system. SPPSO is used in solving hepatitis B virus dynamics (HBV) model. It has good research and practical value.According to the drug effect lag effect, the delay parameter is identified as a parameter which will be estimated, realized the Hepatitis B virus delay model identification using SPPSO and the algorithm which is suitable to solve the delay differential equation.Pay attention to the time-varying characteristics of the medical parameters, the linear combination of the orthogonal polynomial is used to approximate the time varying parameter, which is means that the infinite dimensional problem is transformed into finite dimensional problems. And then the identification of the Hepatitis B virus time-varying model is realized using SPPSO.This dissertation provides a new method of parameter identification for Hepatitis B virus dynamics model. This method could be spread to the general nonlinear system identification. The Research result provides a "soft measurement" method for medical measurement:if the medicine parameter can’t be detected directly or the testing cost is very high, we can choose " the second variables " which is easy to detect, set up the mathematical model between "the secondary variables" and the variables which we wish to detect (in infectious diseases, the general form is a nonlinear differential equations), the variables which we wish to detect can be estimated through the system identification. It has good research and practical value.更多还原