基于“M+N”理论的近红外光谱血糖无创测量方法

摘要

In order to improve the accuracy of noninvasive blood glucose measurement,based on the error theory,"M + N" theory illustrates the effect of " M" factor such as non-measurement components and " N" factor such as external disturbance in the spectral analysis,and provides a method to improve measurement accuracy.In this paper,the effects of four kinds of non-measurement components of "M" factor and contact pressure of "N" factor on the predictive value of blood glucose were considered.The single-trial estimation method was used to reduce the error caused by the contact pressure,and four kinds of non-measurement components such as triglyceride and cholesterol were incorporated into the prediction model.192 groups of valid sample data were obtained in this experiment.The prediction model of blood glucose was built based on partial least squares,the correlation coefficient and the average relative error under the influence of multi factors were 0.929 5 and 0.033,respectively,while the correlation coefficient and the average relative error without considering many factors influence were 0.828 5 and 0.046.The results show that the method based on the "M + N" theory can achieve a better prediction accuracy.%为了提高无创血糖测量的精度,基于“M+N”理论从误差理论的角度阐明了在光谱分析中M种非测量组分(“M”因素)和N种外界干扰因素(“N”因素)对测量精度的影响.考虑了“M”因素中四种非测量组分和“N”因素中接触压力对血糖测量值的影响,建立基于“M+N”理论的血糖人体试验系统.利用单沿法减小“N”因素系统误差中接触压力带来的误差,并将甘油三酯、胆固醇等4种非测量组分纳入预测模型中,系统地考虑其对血糖测量的影响.本次实验获得192组有效样本,采取偏最小二乘法建模预测.考虑多因素影响下的预测值与真值的相关系数为0.929 5,平均预测相对误差为0.033,而不考虑多因素影响下的血糖预测值与真值的相关系数为0.828 5,平均预测相对误差为0.046.表明基于“M+N”理论的近红外血糖无创测量方法可以达到较好的预测精度.