在近红外无创血糖测量中, 由葡萄糖引起的信号变化十分微弱, 极易受到人体背景、 测量仪器、 周围环境等变化的影响, 限制了无创血糖检测的精度。 针对这个问题, 提出应用浮动基准位置理论进行背景变异的校正。 但是由于个体的差异性与人体环境的复杂多变性, 浮动基准位置会因人而异。 通过对人体手掌三层皮肤模型进行蒙特卡洛模拟, 发现其在1 000～1 700 nm近红外波段的浮动基准位置基本处于距光源径向距离2 mm附近。 为了提高浮动基准位置理论在不同人体之间的适用性, 提出了一种近浮动基准参考测量方法, 即以径向距离2 mm处作为参考位置进行背景干扰的修正, 并通过仿体实验验证其效果。 选取与人体手掌皮肤模型的浮动基准位置较为接近的2%和3%浓度的Intralipid仿体溶液进行实验。 实验结果表明近浮动基准修正法可以消减光源漂移对测量结果的影响, 提高数据的重复性和稳定性; 同时通过对不同葡萄糖含量的2%和3%浓度的Intralipid溶液进行多次漫反射信号采集并建立葡萄糖浓度预测模型, 发现修正后的回归模型的预测均方根误差(RMSEP)分别降低了3851%~7998%和2972%~5222%, 说明该方法能够比较有效地消除两个测量位置处共同的背景干扰, 提高校正模型的预测精度。 仿体实验验证的结果, 为下一步近浮动基准参考测量方法的在体测量提供了有力的支撑。

In near-infrared non-invasive blood glucose measurement, the signal variation caused by glucose is very weak, vulnerable to human involvement, measuring instruments and environmental changes, limiting the accuracy of non-invasive blood glucose sensing. To address this issue, the group applies the theory of floating reference position to eliminate background interference.However, due to the individual differences and the complexity of human environment, the floating reference positions will vary from person to person.Based on the Monte Carlo simulation of three layer skin model of human palm, the floating reference positions of 1 000~1 700 nm near infrared band were found in the vicinity of 2 mm of the radial distance from the light source.So in order to improve the applicability of the floating reference methods in different persons, this paper presents a near floating reference measuring(N-FRM) method. The radial distance 2 mm was chosen as the reference position for background interference’s correction, then the effect by phantom experiments is validated. Furthermore select the phantom Intralipid solution with 2% and 3% concentration for experiments, whose floating reference positions are close to the skin model of human palm. The results showed that the N-FRM method could reduce the influence of light source drift, improve the repeatability and stability of data. Based on multiple diffuse reflection signal acquisition of 2% and 3% intralipid solutions with different glucose content and the glucose concentration prediction models, and it was found that the regression models' root mean square errors of prediction were reduced by 3851%~7998% and 2972%~5222%. The comparison showed that this method could effectively eliminate the common background changes at two measured positions,improve the prediction accuracy of the calibration models. The results of phantom experiments provide a strong support for the application of N-FRM’s in vivo measurement.