基于净信号的混合校正方法用于复杂生物样品光谱的定量分析和物理解析

A Hybrid Multivariate Calibration Method based on Net Analyte Preprocessing for Quantitative Analysis and Physical Interpretation of Complex Biological Sample Spectra

对于复杂生物样品的光谱定量分析,单独应用偏最小二乘回归（PLSR）不易获得被测量信息的物理解析。为此建立了一种基于被测量净信号（NAP）的混合校正模型（NAP-PLSR）,并应用于离体和在体实验,进行了葡萄糖含量的近红外光谱定量分析和物理解析研究。实验结果表明,通过NAP-PLSR模型获得的净信号灵敏度曲线易于分辨,能够提取到与葡萄糖分子吸收有关的1 100~1 300nm和1 500~1 800nm两波段信息,提高了模型精度的同时可获得有效的物理解析。

For quantitative analysis of complex biological sample spectra,only using partial least squares regression（PLSR） is not easy to obtain effective physical interpretation. Thus, a hybrid multivariate calibration strategy based on net analyte preprocessing（NAP）, named as NAP-PLSR,was constructed in this paper. This model was applied in two experiments in vitro and in vivo for spectral quantitative analysis and physical interpretation of glucose concentration noninvasive measurement using near infrared（NIR） spectroscopy. The results indicated that under the NAP-PLSR the net sensitivity curve was prone to be discerned. Two spectral bands （1 100-1 300 nm and 1 500-1 800 nm） related of glucose were extracted, which was benefit to improve prediction accuracy, and at meanwhile, the effective physical explanation could be got for complex sample spectral analysis.