中红外无创血糖测量中光程变化对PLSR模型的影响研究  

Effect of Path-Length Variations on PLSR Calibration Model in Noninvasive Measurement of Blood Glucose by Mid-Infrared Spectroscopy

作　　者：王动民[1] 卢启鹏[2] 丁海泉[2] 黄富荣[1] 

机构地区：[1]暨南大学光电工程系光电信息与传感技术广东普通高校重点实验室,广东广州510632 [2]中国科学院长春光学精密机械与物理研究所应用光学国家重点实验室,吉林长春130033

出　　处：《光谱学与光谱分析》2012年第4期930-933,共4页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(61078038;60938002;60878052);中国科学院知识创新工程领域前沿项目(Y00232Q100);应用光学国家重点实验室开放基金项目(O9Q13FQ090);广东省自然科学基金项目(S2011040001850);广东高校优秀青年创新人才培养计划(LYM11026)资助

摘　　要：在测量人体皮肤中红外光谱的过程中,皮肤与ATR晶体之间的接触面积很难保持一致,导致倏逝波与人体皮肤的作用光程会产生差异。以光程变化信息与葡萄糖浓度信息之间的相关性为基础,建立了两个分析葡萄糖含量的偏最小二乘回归(PLSR)模型,RMSECV分别为31.3和4.52mg.dL-1;RMSEP分别为30.3和98.7mg.dL-1。结果表明,当光程变化信息与葡萄糖浓度信息之间偶然相关时,所建PLSR模型的预测精度与稳健性会受到不同程度的影响,随着二者之间相关性的增强,最优模型对葡萄糖信息的识别能力越来越弱。为提高中红外无创血糖测量分析结果的可靠性,避免获得伪优定标模型提供了实验依据。In noninvasive blood glucose measurement, it is difficult to keep the contact area between skin and internal reflectance element uniform while the mid-infrared spectra of human skin are taken, and this would lead to path-length variations. To study the effect of path-length variations on PLSR calibration model, in the present paper, according to the correlation coefficients between path-lengths and glucose concentrations, two PLSR models were achieved respectively and RMSECV were 31.3 and 4.52 mg · dL-1 , RMSEP were 30. 3 and 98. 7 mg · dL 1 for the validation set. The results show that the chance correlations between path-lengths and glucose concentrations will lead to calibration models with different accuracy and robustness. This is useful to improving the reliability of noninvasive measurement of blood glucose by mid infrared spectroscopy.

关 键 词：中红外光谱 无创血糖测量 糖尿病 偏最小二乘回归 

分 类 号：O657.3[理学—分析化学]