【作者】 黄富荣；

【导师】 陈星旦；

【作者基本信息】 暨南大学 ， 生物医学信息技术， 2010， 博士

【摘要】 近红外光谱技术以其无试剂、快速、低成本、绿色环保等显著优点,正在成为分析领域的重要技术手段。利用近红外光谱的方法实现血液中胆固醇、甘油三酯含量的无试剂定量分析,对临床生物医学检测具有重要意义。本研究采用近红外光谱(NIR)结合化学计量学方法,建立无需化学试剂、精度适中、直接、快速、同时测定血液胆固醇、甘油三酯的定量分析方法。首先分析混合人群血清样品胆固醇、甘油三酯在全谱区内直接偏最小二乘法(PLS)建模的预测效果；然后采用间隔偏最小二乘法(iPLS)进行特征波长优选,在特征谱区内建立结构简单、鲁棒性强的预测模型,用于临床血脂检查的初筛。再次通过人群分类对血清预测模型优化,建立一个结构简单、预测精度更高的人群分类模型,尤其是病人专用模型。最后对全血样品进行分析,探讨无试剂近红外检测全血样品胆固醇、甘油三酯可行性,为无试剂近红外微创检测血脂积累相关的数据。论文取得的主要成果有：(1)混合人群血清胆固醇、甘油三酯直接PLS建模的研究表明：在全谱区内直接PLS建模的精度和稳定度都不高。胆固醇最优预测模型的Rp、RMSEP分别为：0.899、0.565 mmol/L;甘油三酯最优预测模型的Rp、RMSEP分别为：0.847、0.454 mmol/L。(2)iPLS对混合人群血清样品的胆固醇、甘油三酯建模优化的研究表明：采用iPLS不仅可以找到特征吸收波段,而且可以提高模型的预测能力和精度。胆固醇最优建模波段是1700-1798nm,最优预测模型的Rp、RMSEP分别为0.984、0198 mmol/L。甘油三酯最优建模波段是1654-1746nm,最优预测模型的Rp、RMSEP分别为0.967、0.157 mmol/L。(3)人群分类对模型优化的研究表明：健康人或病人单独建模对健康人或病人的预测要比混合模型对健康人或病人预测的精度高，尤其是病人单独建模对病人的预测精度有较大提高。胆固醇病人模型的Rp从0.78增大为0.97, RMSEP由0.22 mmol/L降低为0.14 mmol/L;甘油三酯病人模型的Rp从0.95提高为0.96预测均方差RMSEP由0.18 mmol/L降低为0.15 mmol/L。(4)全血样品胆固醇、甘油三酯近红外光谱分析表明：无试剂近红外分析全血样品胆固醇、甘油三酯含量是可行性的。胆固醇最优建模波1650-1730nm,最优预测模型的Rp、RMSEP分别为0.792、0.502 mmol/L。甘油三酯最优建模波段是2260-2340nm,最优预测模型Rp、RMSEP分别为0.865、0.284 mmol/L。更多还原

【Abstract】 Near Infrared Spectroscopy analytical technique is reagentless, rapid, green, and can be run at low cost. It has become an important method in clinical analyses field. The success of using near infrared spectroscopy to detect content of cholesterol and triglyceride with quantitative analysis will have a profound affection on the clinical biomedical analyses. In this paper, determination of cholesterol and triglyceride in serum and whole blood have been studied with near infrared transmission spectroscopy and chemometric methods. First, we analyze the prediction effect of the cholesterol and triglyceride model in full-spectrum region by partial least squares(PLS) method directly. Second, we use interval partial least squares (iPLS) to select characteristic wavelength, in order to establish simple, robust and strong prediction models of cholesterol and triglyceride for all groups of people. Again, we study the prediction effect of the cholesterol and triglyceride model in different population by population classification for establishing a simple, single population prediction model, especially the patient-specific model. Finally, we explore feadibility of determination of cholesterol and triglyceride in whole blood by near infrared transmission spectroscopy reagentlessly. It will accumulate useful data for minimally invasive blood lipid detection by near infrared.The main results of this study include:(1) There are lower precision and stability of prediction for cholesterol and triglyceride in serum by direct PLS. The prediction correlation coefficient (RP), the root mean square error of prediction(RMSEP) for cholesterol are 0.899,0.565 mmol/L respectively. RP, RMSEP for triglyceride are 0.847,0.454 mmol/L respectively.(2) Results of optimized model for cholesterol and triglyceride of all serum samples show that interval partial least squares.can improve prediction accuracy. For cholesterol, the prediction effect of the model on 1700-1798nm is the best, and RP, RMSEP are 0.984,0.198 mmol/L respectively. For triglyceride,the prediction effect of the model on 1654-1746nm is the best, and RP, RMSEP are 0.967,0.157 mmol/L respectively.(3) Results of optimized model by population classification show that model of independent groups has higher prediction accuracy than model of all groups, especially in patients independent model prediction accuracy for patients has improved greatly. RP was improved from 0.78 to 0.97, RMSEP was reduced from 0.22 mmol/L to 0.14 mmol/L for the prediction of patients with high cholesterol, RP was improved from 0.95 to 0.96, RMSEP was reduced from 0.18 mmol/L to 0.15 mmol/L for the prediction of patients with high triglyceride.(4) It is possible to determine of cholesterol, triglyceride in whole blood by near infrared spectroscopy reagentlessly. The prediction effect of the model for cholesterol on 1650-1730nm is the best, and RP, RMSEP are 0.792,0.502 mmol/L respectively. The prediction effect of the model for triglyceride on 2260-2340nm is the best, and RP, RMSEP are 0.865,0.284 mmol/L respectively.更多还原