【作者】 陶熠昆；

【导师】 孟濬；

【作者基本信息】 浙江大学 ， 控制理论与控制工程， 2010， 硕士

【摘要】 用非线性系统方法与复杂网络对医学数据进行建模,是研究方法从独立法与隔离法到整体法与系统法的重大转变。相比传统的专家经验与线性分析方法,非线性方法对刻画内部关系错综复杂的非线性医学对象方面具有更好的适应性。本文以非线性系统方法为理论基础,以相空间重构为主要技术手段,针对以下两类医学数据:癫痫患者的EEG脑电数据以及胃癌患者的SELDI-TOF血清蛋白质质谱数据进行非线性分析和分类模型建立,并取得了一系列有价值的成果。本文主要做了如下方面工作:(1)、以癫痫患者EEG数据为对象进行非线性系统分析与建模。从工程实现的角度出发,本文建立了一体化处理管道——包括数据筛选、预处理、降维、相空间重构、分类模型建立、计算量优化,使得本文提出的建模方法具有完整性、鲁棒性与实时性。具体而言,本文通过替代数据法对多通道数据进行非线性特征筛选,因子分析法对数据进行降维,进而应用改进的多通道相空间重构技术重构脑电信号奇异吸引子,并通过奇异吸引子关联维数成功建立癫痫期和癫痫间期EEG数据的分类模型。本文同时详细讨论且优化了关联维工程计算的时间复杂度,以满足关联维实时在线计算的需要。(2)、针对当前蛋白质谱生物标记建模方法的不足,本文提出了基于非线性与系统理论的蛋白质质谱数据建模与判别新思路。本文对胃癌患者SELDI-TOF血清蛋白质质谱数据进行建模与判别。以相空间重构为主要方法,对20例正常个体和20例癌症个体的质谱数据分别在两个数据区段进行相空间重构,并以这两个区段的关联维数为输入建立线性分类模型,验证了该模型优于单个蛋白生物标记的分类模型。(3)、在高维复杂医学数据建模领域引入动态复杂网络建模思想。并试着提出了针对超高维的、有连续性的医学数据进行复杂网络建模的一种可行性方法及可能的后续分析思路。更多还原

【Abstract】 The nonlinear and systematic methods, which describe their research target from a holistic and systematic point of view, show better adaptability when modeling the high-dimensional and complex medical data.In this paper, two types of data: EEG data of epilepsies patients and SELDI-TOF serum protein mass spectrometry data of patients with gastric cancer are processed based on the nonlinear theory and the technology of phase space reconstruction. Some valuable results were obtained from the data processing and model building procedure. The major works involved in this paper are as follows:(1) This paper presented the analysis and modeling of a nonlinear system based on the EEG data of epilepsies patients. From the engineering implementation point of view, this paper established a uniform processing channel, including data filtering, preprocessing, dimension reduction, phase space reconstruction, classifier modeling and computational optimization, which make the method complete, robust and real-time.More specifically, in this paper multi-channel data have been filtered with nonlinear characteristics by method of surrogate data. Factor analysis is used for dimension reduction, Improved multi-channel phase space reconstruction technology is used for reconstruction of the EEG strange attractor. Furthermore, the classifier model of EEG data in illness-period and inter-illness-period has been constructed by the correlation dimension of singular attractors. In the mean time a comprehensive discussion and optimization have been finished on the time-complexity of the correlation dimension computation, in order to satisfy the requirement of real time on-line computation.(2) Focusing on the insufficiency of the current spectra] modeling of protein biomarkers, this paper proposed a new method for protein mass spectrometry data modeling and discrimination based on the nonlinear theory and systematic methods. This paper carried out the modeling and classification of the SELDI-TOF serum protein mass spectrometry data of patients with gastric cancer. Using phase space reconstruction, the data from 20 normal samples and 20 cancer samples are reconstructed in 2 specific data ranges, based on which a linear classifier model is established. The result verified the superiority over the classifier model of single protein biomarkers.(3) The concept of dynamic complex network modeling has been introduced to the modeling of high-dimensional, complex medical data. A feasibility study and further work have been proposed according to the complex network modeling of high-dimensional, continuous medical data.更多还原