基于拉曼光谱法疾病诊断的模式识别研究

【作者】 杨红；

【导师】 刘蓉；

【作者基本信息】 陕西师范大学 ， 工程硕士（专业学位）， 2020， 硕士

【摘要】 拉曼光谱是一种非弹性散射光谱,基于拉曼光谱的分子信息检测法具有灵敏度高、检测速度快、抗干扰能力强等优点,并且可以实现对待检测样本的无损定量分析,是当下分子信息检测领域的一大研究热点。本文选择包虫病与慢性肾功能衰竭两种疾病作为诊断对象,分别提取包虫病患者血清拉曼光谱信号与慢性肾功能衰竭患者尿液拉曼光信号,并与对照组(健康人)样本的拉曼光谱信号进行比对,基于不同算法建立了两种疾病的分类诊断模型来研究拉曼光谱在疾病诊断方面的模式识别与数据处理算法。本论文的主要研究内容如下:1.基于健康人与包虫病患者血清拉曼光谱样本数据建立包虫病分类诊断模型,分别使用两种不同的算法建立包虫病诊断模型,并对两种模型的分类诊断性能进行对比。首先对173例健康人和包虫病患者的血清拉曼光谱信号样本数据进行原始数据的分析及预处理(归一化、扣除荧光背景、基线校正等),原始数据分析得到了健康人和包虫病患者的血清拉曼光谱信号的特点及异同点,原始数据预处理得到了信噪比更高的拉曼光谱信号数据。之后分别建立了基于主成分分析-线性判别(PCA-LDA)和偏最小二乘法-线性判别(PLS-LDA)两种算法的多元统计分析模型(分类诊断模型),前者的预测总正确率率仅为69.2308%,而后者的预测总正确率为92.3077%,两者的诊断准确率差距为25%。因此,PLS-LDA分类诊断模型更适用于包虫病的筛查。2.使用健康人和慢性肾功能衰竭(CRF)患者尿液拉曼光谱信号结合模式识别算法进行CRF疾病的诊断筛查,分别使用不同的模式识别算法建立CRF疾病分类诊断模型,探讨了拉曼光谱检测在快速筛查CRF患者中的适用性。首先对92例健康人和CRF患者尿液拉曼光谱样本数据进行对比分析,之后使用主成分分析法(PCA)对样本数据进行特征提取以减小高维光谱数据的维数,最后基于不同的分类器算法分别建立了反向传播神经网络(BP)诊断模型,遗传算法优化支持向量机(GA-SVM)诊断模型、网格搜索优化支持向量机(GS-SVM)诊断模型和粒子群优化支持向量机(PSO-SVM)诊断模型,四种诊断模型的准确率分别为70.77%、80.77%、84.62%和74.62%。其中,最佳快速检测模型PCA-GS-SVM的灵敏度、特异度和准确度分别为83.33%,85.71%和84.62%。本文的实验研究表明,拉曼光谱结合多元算法模型能够有效准确的完成对两种疾病样本的筛查。本文的研究有助于推进拉曼光谱在感染性疾病监测领域的应用研究,同时对拉曼光谱检测法在其他医学疾病诊断、医学检测领域的应用研究具有一定的借鉴意义。更多还原

【Abstract】 Raman spectroscopy is a kind of inelastic scattering spectroscopy.Raman spectroscopybased molecular information detection method has the advantages of high sensitivity,fast detection speed,strong anti-interference ability,etc.,and can realize non-destructive quantitative analysis of samples to be detected.It is a major research hotspot in the field of detection.In this thiesis,two diseases,hydatid disease and chronic renal failure,are selected as the diagnostic objects.The serum Raman spectrum signals of patients with hydatid disease and the urine Raman light signals of patients with chronic renal failure are extracted,and compared with the control group(healthy people)samples.The Raman spectrum signals are compared,and a classification diagnosis model of two diseases is established based on different algorithms to study the pattern recognition and data processing algorithms of Raman spectrum in disease diagnosis.The main research contents of this theiesis are as follows:1.Establish a classification diagnosis model of hydatid disease based on serum Raman spectrum sample data of healthy people and patients with hydatid disease,and use two different algorithms to establish a diagnostic model of hydatid disease,and compare the classification and diagnostic performance of the two models.First,analysis and preprocessing(normalization,subtraction of fluorescence background,baseline correction,etc.)of the original Raman spectrum sample data of serum samples of 173 healthy people and hydatid patients were obtained from the analysis of raw data.The characteristics and similarities and differences of serum Raman spectral signals of patients with worm disease.The raw data was preprocessed to obtain Raman spectral signal data with higher signal-to-noise ratio.Then,a multivariate statistical analysis model(classification diagnosis model)based on the principal component analysis-linear discrimination(PCA-LDA)and partial least squares-linear discrimination(PLSLDA)algorithms was established.Only 69.2308%,and the total prediction accuracy rate of the latter is 92.3077%,and the difference between the two diagnostic accuracy rates is 25%.Therefore,the PLS-LDA classification diagnostic model is more suitable for screening for hydatid disease.2.Urine Raman spectrum signals of healthy people and patients with chronic renal failure(CRF)were combined with pattern recognition algorithms for the diagnosis and screening of CRF diseases.Different pattern recognition algorithms were used to establish a CRF disease classification diagnosis model,and Raman was discussed.Applicability of Spectral Detection in Rapid Screening of CRF Patients.First,the urine Raman spectrum sample data of 92 healthy people and CRF patients were compared and analyzed.Then,the principal component analysis(PCA)was used to extract the features of the sample data to reduce the dimensionality of the high-dimensional spectral data.The classifier algorithm has established back propagation neural network(BP)diagnostic models,genetic algorithm optimized support vector machine(GA-SVM)diagnostic model,grid search optimized support vector machine(GS-SVM)diagnostic model,and particle swarm optimization support.The vector machine(PSO-SVM)diagnostic model has four diagnostic models with accuracy rates of 70.77%,80.77%,84.62%,and 74.62%.Among them,the sensitivity,specificity and accuracy of the best rapid detection model PCA-GS-SVM were 83.33%,85.71%and 84.62%,respectively.The experimental research in this thiesis shows that Raman spectroscopy combined with a multivariate algorithm model can effectively and accurately screen two disease samples.The research in this article is helpful to promote the application research of Raman spectroscopy in the field of infectious disease monitoring,and it has certain reference significance for the application research of Raman spectroscopy in other medical disease diagnosis and medical detection fields.更多还原