Study on the Wavelength Selection Based on VIP Analysis in Noninvasive Measurement of Blood Components
HE Wen-qin1, 2, YAN Wen-juan3, HE Guo-quan3, YANG Zeng-bao3, TAN Yong3, LI Gang1, 2, LIN Ling1, 2*
1. State Key Laboratory of Precision Measurement Technology and Instruments, Tianjin University, Tianjin 300072, China 2. Tianjin Key Laboratory of Biomedical Detecting Techniques & Instruments, Tianjin University, Tianjin 300072, China 3. School of Electronic Information Engineering, Yangtze Normal University, Chongqing 408100, China
摘要： 自变量筛选是定量光谱分析领域的研究热点，简便且高效的自变量筛选方法不但可以降低分析计算量，提高分析精度，同时还可以减轻对仪器光谱分辨能力的依赖，降低分析成本。波长筛选也是光谱法无创血液成分检测研究的重要环节。动态光谱理论为血液无创检测提供了极佳的思路，但长期局限于使用宽带光源和高分辨率的光谱仪器，分析中需要大量波长限制了动态光谱法的进一步发展。为了去除冗余信息，使检测走向低成本化和集成化，提出了基于变量投影重要性(variable importance in projection，VIP)分析的波长筛选方法。通过分析PLS模型中各维自变量对因变量的解释能力，从而剔除重要性较低的变量保留解释能力强的波长。以232例受试者的临床实验数据为基础，以血红蛋白含量为分析对象，经投影重要性分析后将波长数由586降至64，波长筛选后血红蛋白预测模型的测试集平均相对误差(MREP)为1.82%，使用了极少的波长便可得到满意的结果;结合Bootstrap方法对模型进行显著性检验后验证了波长变量的解释能力。首次指出了使用动态光谱法检测血红蛋白的敏感波长带。基于投影重要性分析的波长筛选迈出了动态光谱走向实用的重要一步，为实现低成本在线分析打下了基础，同时也为其他领域的光谱分析提供了重要的参考和新的思路。
关键词：波长筛选;动态光谱;变量投影重要性;无创血液成分检测;光电容积脉搏波
Abstract：Selection of independent variable is a hotspot in the field of quantitative spectral analysis. Efficient and easy-to-use method for wavelength selection can not only reduce the computation and improve the accuracy of analysis, but also can reduce dependency on the spectral resolution of instruments and cut cost. Wavelength selection is also an important part of the research about noninvasive measurement of blood components by spectrum technology. Dynamic Spectrum theory provides excellent ideas for researchers, but only broadband light source and high-resolution spectrograph were used in correlational studies for a long time. The large number of wavelengths needed for analysis limits the further development of Dynamic Spectrum method. In order to remove redundancy information and make the devices low-cost and integrated, the method of Wavelength selection on the basis of Variable Importance in Projection (VIP) analysis was proposed. Variables with less importance were removed and wavelengths with great explanation power were retained after VIP analysis the number of wavelengths was reduced from 586 to 64. PLS model with 64 wavelengths get a satisfactory predict result that the MREP is 1.82%. The significance test through Bootstrap method validate the selected wavelengths’ explanation power. Moreover, the study pointed out the sensitive wavelengths in Dynamic Spectrum method for the first time. Study also took the first step toward the practical application of Dynamic Spectrum and laid the foundation of low-cost on-line analysis, and it also provided valuable references and new ideas for spectral analysis in other areas.
Key words：Wavelength selection;Noninvasive measurement of blood components;Variable importance in the Project (VIP);Dynamic spectrum;photoelectric plethysmography
贺文钦1, 2，严文娟3，贺国权3，杨增宝3，谭 勇3，李 刚1, 2，林 凌1, 2*. 无创血液成分检测中基于VIP分析的波长筛选[J]. 光谱学与光谱分析, 2016, 36(04): 1080-1084.
HE Wen-qin1, 2, YAN Wen-juan3, HE Guo-quan3, YANG Zeng-bao3, TAN Yong3, LI Gang1, 2, LIN Ling1, 2*. Study on the Wavelength Selection Based on VIP Analysis in Noninvasive Measurement of Blood Components. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2016, 36(04): 1080-1084.
[1] Kraitl J, Klinger D, Fricke D, et al. Advancement in Sensing Technology. Springer Berlin Heidelberg, 2013. 237. [2] Wang H, Li G, Zhao Z, et al. Transactions of the Institute of Measurement and Control, 2011. 0142331211411965. [3] CHEN Xing-dan(陈星旦). Optics and Precision Engineering(光学精密工程), 2008, 16(5): 759. [4] Li W, Lin L, Li G. Analytical Methods, 2014, 6(4): 1082. [5] ZHANG Zheng, FENG Guo-shuang(张 政, 冯国双). Modern Preventive Medicine(现代预防医学), 2012, 39(22)：5813. [6] Bergdahl M, Bergdahl J. Journal of Dental Research, 2000, 79(9): 1652. [7] Oussama A, Elabadi F, Platikanov S, et al. Journal of the American Oil Chemists’ Society, 2012, 89(10): 1807. [8] ZENG Tao, JU Cun-yong, CAI Ti-jiu, et al(曾 涛, 琚存勇, 蔡体久，等). Journal of Beijing Forestry University(北京林业大学学报), 2010, 32(6): 37. [9] LIU Guo-hai, CHENG Jin-xiang, DING Yu-han, et al(刘国海, 程锦翔, 丁昱函, 等). Computers and Applied Chemistry(计算机与应用化学), 2013, 11: 12. [10] Broderick G, Craddock R C, Whistler T, et al. Pharmacogenomics，2006, 7(3): 407. [11] Sara Bergante, Gianni Facciotto, Gianfranco Minotta. Central European Journal of Biology，2010, 5(4)：522. [12] Araújo M C U, Saldanha T C B, Galvo R K H, et al. Chemometrics and Intelligent Laboratory Systems, 2001, 57(2): 65. [13] Lee K R, Lin X, Park D C, et al. Proteomics, 2003, 3(9): 1680. [14] Zhou Mei, Lin Ling, Wang Mengjun, et al. Analytical Methods, 2013, 5：4660. [15] JIA Ping, ZHANG Bao-ju, ZHANG Zhi-yong, et al(贾 萍, 张宝菊, 张志勇, 等). The Journal of Infrared and Millimeter Waves(红外与毫米波学报), 2010, 29(2): 132. [16] WANG Hui-wen, WU Zai-bin, MENG Jie(王惠文, 吴载斌, 孟 洁). Partial Least-Squares Regression—Linear and Nonlinear Methods(偏最小二乘回归的线性与非线性方法). Beijing: National Defense Industry Press(北京：国防工业出版社), 2006. [17] Efron B. The Annals of Statistics, 1979, 71(1). 1. [18] Gosselin R, Rodrigue D, Duchesne C. Chemometrics and Intelligent Laboratory Systems, 2010, 100(1): 12.
