近红外光谱无创血糖测量中背景扣除方法的研究

Research on the Background Correction in the Non-Invasive Sensing of Glucose by Near-Infrared Spectroscopy

在近红外光谱人体血糖无创检测中,由于葡萄糖在体内含量非常低,而光谱中的噪声组成复杂且变化幅度大,从而导致测量的信噪比较低。背景扣除是提高测量信噪比的有效预处理手段之一。文章首先对常用的背景扣除方法进行了理论推导,并进一步考虑不同样品在测量过程中存在的样品本身特性变化,提出了相近背景扣除的方法,即选择与样品的光学特性变化相似的样品作为背景,能更有效地消除样品特性变化和仪器漂移的影响。并在纯吸收和散射介质中分别加入葡萄糖进行实验验证。结果表明,在血浆溶液和Intralipid-2%溶液的葡萄糖实验中,选用与样品特性更接近的样品作为背景时,葡萄糖浓度的预测精度分别提高了25.9%和40.1%。

For the non-invasive blood glucose concentration sensing by the near-infrared spectroscopy, the signal to noise ratio of the optical measurement system is very low. Both the content of glucose in body and the absorption coefficient of glucose in the near-infrared region are quite weak. More over, the structure of spectral noise is complicated and the variation of noise intensity is very large. The background correction is one of the most effective pre-processing methods to improve the signal to noise ratio of the near-infrared optical detection system. In this paper, the theory expression formula of traditional background correction method was induced firstly. Then in order to avoid the influence from the variation of optical characteristics in the sample and the drift in the optical system, the similar background correction method was proposed, that is, the background which has the similar optical characteristics with the sample was chosen as the reference. The in vitro experiments of pure absorption media and scattering media were conducted to validate the effect. The results showed that, for the glucose in the blood plasma solution and Intralipid-2% solution, after the correction of the background which has the similar optical characteristics with the sample, the prediction precision of multivariate model for glucose concentration has been improved by 25.9 % and 40. 1%, respectively.