【作者】 王怡珊；

【导师】 王平；

【作者基本信息】 浙江大学 ， 生物医学工程， 2012， 硕士

【摘要】 通过检测呼出气体进行疾病诊断的研究已经有将近40年的历史了,近几年有许多研究者将呼出气体中挥发性有机化合物(VOCs)的检测应用于肺癌诊断,但对于肺癌的特征性VOCs成分及其产生机制还没有形成统一的结论。另外,也有许多研究者检测了呼出气体冷凝物(EBC)中肺癌的标志物,如癌胚抗原(CEA)等。呼吸检测是一种快速,无创,新颖的检测手段,具有广阔的应用前景。本论文分析了呼出气体中挥发性肺癌标志物和呼出气体冷凝物中非挥发性肺癌标志物,并用自行研制的电子鼻来检测呼出气体中肺癌特征性VOCs。主要内容包括以下几个方面：本论文进一步分析了前期於锦等人从采集的85例肺癌患者,70例肺部良性疾病患者以及88例健康人呼吸气体样本的质谱数据中提取出的41种内源性VOCs。分析了每一种VOCs的ROC曲线,并根据它们各自的ROC曲线下面积和统计学差异p值,选出了25种在肺癌组和对照组有统计学差异的VOCs,作为肺癌特征性标志物。然后又采用线性判别式分析,建立最佳肺癌诊断模型,该最佳模型的敏感度和特异性分别达到了95.29%和96.20%。我们实验室自行研制了两台电子鼻用来检测呼出气体中的VOCs,一台基于金属氧化物半导体(MOS)传感器,一台基于声表面波(SAW)传感器。本论文开发了两套软件,一套是与基于MOS传感器的CN e-NoseⅡ呼吸检测电子鼻配套的检测分析软件,该软件主要完成对仪器的控制和对传感器数据的处理分析。另一套软件是MOS-SAW复合传感器肺癌诊断软件,该软件实现的是对MOS传感器和SAW传感器的数据进行分析并建立肺癌诊断模型。本论文采用这两台电子鼻分析了42例健康人和47例肺癌患者呼出气体样本,从数据处理后的传感器响应曲线中提取出138个特征值。同时分析了这些特征值各自的ROC曲线,根据ROC曲线下面积提取出53个对区分肺癌组和健康组有统计学意义的特征值作为模型的自变量。最后采用主成分分析(PCA)、线性判别式分析(LDA)、人工神经网络(ANN)以及偏最小二乘回归分析(PLS)这四种模式识别算法建立了以下六种模型：LDA模型、ANN模型、PLS模型、PCA-LDA模型、PCA-ANN模型以及PCA-PLS模型。最终发现PCA-ANN模型具有最高的特异性和敏感度,分别为90.48%和93.62%,并具有较高的建模效率。本论文还采集分析了肺癌患者的EBC样本,对EBC中的癌胚抗原(CEA),神经元特异性烯醇化酶(NSE)以及鳞状细胞癌抗原(SCC)这三种蛋白质的含量进行了检测分析。虽然EBC中肺癌标志物近年来已被很多研究人员研究,但对这三种标志物的研究还很少见,虽然这些标志物是常见的血清肺癌标志物,但很少有人研究其在EBC中的浓度。本论文检测了EBC中这三种标志物,发现CEA和SCC的检出率达到30%左右,本论文还分析这三种标志物与肺癌病理类型之间的关系。更多还原

【Abstract】 The research on disease diagnosis through exhaled breath detecting has been almost 40 years. In the recent years, the detection of Volatile Organic Compounds (VOCs) has been applied in lung cancer diagnosis by many researchers. But the researchers don’t have an agreement on the VOCs biomarkers for lung cancer and the production mechanism of these VOCs biomarkers. In addition, the biomarkers for lung cancer in Exhaled Breath Condensate (EBC), like Carcinoma Embryonic Antigen (CEA), have also been studied by many researchers in recent years. The detection of biomarkers in VOCs and EBC is a quick, noninvasive and novel way to diagnose lung cancer. Its application in clinical diagnosis for lung cancer has a vast prospect.The VOCs biomarkers and the nonvolatile biomarkers for lung cancer in exhaled breath were analyzed in this paper. And the e-Nose was applied in this paper to detect the VOCs biomarkers for lung cancer. This paper includes four parts as in the following.The first part is the research on VOCs biomarkers for lung cancer. The exhaled breath of 85 lung cancer patients,70 lung benign disease patients and 88 healthy people were analyzed using GCMS by Jin Yu in the early days, and 41 endogenous VOCs were extracted out from the GCMS data. These data were reanalyzed in this paper. The availablity of every endogenous VOCs was evaluated through their Receiver Operating Characteristic (ROC) curves.25 VOCs biomarkers for lung cancer were selected out according their AUC (Area Under ROC Cureve) and the values of p. The Linear Discriminant Analysis (LDA) was finally employed to build models to discriminate the lung cancer group and the control group. The sensitivity and specificity of the best model are 95.29% and 96.20% respectively.Our institute has designed two e-Noses to detect VOCs in exhaled breath. One is based on Metal Oxide Semiconductor (MOS) sensors and the other is based on Surface Acoustic Wave (SAW) sensors. This paper developed two sets of software. One is for the CN e-NoseⅡbreath detecting e-Nose based on the MOS sensors. This software can control the e-Nose and dispose the data of sensors. The other software is used to build lung cancer diagnosis models based on the data of MOS sensors and SAW sensors.These two e-Noses were used to analyze the exhaled breath of 42 healthy people and 47 lung cancer patients.138 features were selected out from the sensor curves for every sample. The independent variables of the diagnosis models were selected according the AUC of ROC curve of every feature. Finally, four pattern recognition algorithms:Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA), Artificial Neural Network (ANN) and Partial Least Squares (PLS), were employed to build six diagnosis models:LDA model, ANN model, PLS model, PCA-LDA model, PCA-ANN model and PCA-PLS model. The PCA-ANN model performed best with high sensitivity and specificity (93.62% and 90.48% respectively) and high efficiency.The EBC samples were also collected and the concentrations of Carcinoma Embryonic Antigen (CEA), Neuron Specific Enolase (NSE) and Squamous Cell Carcinoma (SCC) in these samples were measured. Although the lung cancer biomarkers in EBC have been researched by many researchers recently, the detection of these three biomarkers in EBC was rarely researched. According to the result, they all existed in EBC and the detectable ratio of CEA and SCC was about 30%. The relationship between these biomarkers and the types of lung cancer pathology was also analysed)更多还原