近红外光谱无创血糖检测中的偶然相关及波长选择的研究

发布时间：2019-02-13 04:02

【摘要】：糖尿病是危害人类健康的世界性难题,患病率呈上升趋势,无创血糖检测可以降低患者日常监测血糖浓度的不便和痛苦。无创性血糖检测研究中使用较多的是近红外光谱分析技术,但其光谱组成复杂,血糖信号微弱,波长之间多重相关,检测精度需要提高。基于此,本课题对近红外光谱血糖检测中的偶然相关及波长选择的问题进行了研究。运用近红外光谱分析技术进行血糖检测时,需要借助化学计量学方法建立模型来实现对未知样品的定量分析。在模型建立和验证的过程中通常会引入一定程度的偶然相关,从而影响模型的稳健性。本文从算法原理上分析建模方法产生偶然相关的可能性,并采用随机数模拟光谱数据及参考浓度的研究方式进行验证。从建模波长数的选择和交互验证方法两方面考察了不同建模方法在建模的过程中存在偶然相关的概率水平,并给出了最佳的建模波长数以及最优的交互验证方法,以降低引入的偶然相关。在血糖检测中,人体温度等生理因素的改变,同样会引起近红外光谱信息的改变。本文通过离体实验,研究温度对葡萄糖浓度检测的影响关系并指导如何在实际血糖检测中降低温度引入的偶然相关。近红外区域中光谱重叠现象严重,导致在某些波长处的光谱信息的测量选择性和灵敏度较差,直接影响了模型的预测精度。本文采用遗传算法优选人体血糖光谱的波长点,验证其适用性。遗传算法选择的波长多集中在血糖的吸收峰附近,与物理期望一致。传统遗传算法使PLS模型的RMSEP的值降低了11%,动态遗传算法则降低16%。本文通过对偶然相关及波长选择的研究,提高了校正模型的预测精度,对近红外光谱无创血糖检测有一定的价值。
[Abstract]:Diabetes is a worldwide problem that endangers human health, and the prevalence of diabetes is on the rise. Noninvasive blood glucose detection can reduce the inconvenience and pain of daily monitoring of blood glucose concentration. Near-infrared spectroscopy is widely used in noninvasive blood glucose detection, but its spectral composition is complex, the blood glucose signal is weak, the wavelength is multiple correlation, and the detection accuracy needs to be improved. Based on this, the problem of accidental correlation and wavelength selection in near infrared spectroscopy (NIR) blood glucose detection is studied. When using near infrared spectroscopy (NIR) to detect blood sugar, it is necessary to establish a model by means of chemometrics to realize quantitative analysis of unknown samples. In the process of model establishment and verification, a certain degree of accidental correlation is usually introduced, which affects the robustness of the model. In this paper, the possibility of accidental correlation in modeling method is analyzed based on the principle of algorithm, and the random number simulation spectral data and reference concentration are used to verify it. In this paper, the probability level of accidental correlation between different modeling methods in the process of modeling is investigated from the aspects of the selection of modeling wavelength number and the interactive verification method, and the optimal number of modeling wavelengths and the optimal interactive verification method are given. To reduce the introduction of accidental correlation. In the detection of blood sugar, the change of human body temperature and other physiological factors will also cause the change of near infrared spectrum information. In this paper, the influence of temperature on glucose concentration was studied by in vitro experiments, and how to reduce the accidental correlation of temperature in actual blood glucose detection was studied. The spectral overlap in the near infrared region is serious, which leads to the poor selectivity and sensitivity of the spectral information at some wavelengths, which directly affects the prediction accuracy of the model. In this paper, genetic algorithm is used to select the wavelength points of blood glucose spectrum to verify its applicability. The wavelength selected by genetic algorithm is mainly near the absorption peak of blood glucose, which is consistent with the physical expectation. The traditional genetic algorithm reduces the RMSEP value of PLS model by 11 points and the dynamic genetic algorithm by 16%. In this paper, the prediction accuracy of the calibration model is improved through the study of accidental correlation and wavelength selection, and it has certain value for noninvasive detection of blood glucose by near infrared spectroscopy (NIR).
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：R318.51

【参考文献】