应用光程校正空间法提升近红外无创生化分析模型性能

Improvement of model performance for near-infrared non-invasive biochemical analysis by pathlength correction space method

由于采用血流容积差光谱相减法进行近红外无创生化分析得到的样品光谱光程不定,很难得到高性能的定标模型,本文提出采用光程校正空间法来消除光程差异,提升定标模型性能。首先,介绍了净信号分析的原理和校正空间的获取方法;根据血流容积差光谱相减法的特点,提出了采用光程校正空间获取待测组分净信号的方法。然后,以含葡萄糖的脂肪乳溶液为例,通过光程校正空间获取葡萄糖净信号,建立定标模型。实验结果表明:相比没有处理光程直接建立的定标模型,采用光程校正空间后所建立的模型相关系数(R)从0.978 1提升到0.997 7,定标均方根偏差(RMSEC)从77.52mg/dL下降到25.02mg/dL,交叉验证均方根偏差(RMSECV)从93.01mg/dL下降到68.22mg/dL。结果显示,采用光程校正空间的净信号分析能够有效抑制光程差异对定标模型的影响,为血流容积差光谱相减法的实际应用奠定了基础。

As the spectral pathlength obtained by spectral subtraction approach with different flow blood volumes is uncertain in a near-infrared noninvasive biochemical analysis, it is difficult to obtain a high performance calibration model. Therefore, a pathlength correction space method was proposed to eliminate the uncertain and to improve the performance of calibration. Firstly, the principle of Net Analyte Signal （NAS） and how to obtain a correction space were introduced. Then, a new approach to get NAS was proposed using the pathlength correction space according to the features of spectral space, the correlation coefficient has been improved from 0. 978 1 to 0. 997 7, the Root Mean Square Error of Calibration （RMSEC） from 77.52 mg/dL down to 25.02 mg/dL, and the Root Mean Square Error of Cross Validation （RMSECV） from 93.01 mg/dL down to 68.22 mg/dL. The analysis result verifies that the pathlength correction space lengths and the method can set the stage for proach with different flow blood volumes. can effectively restrain the influence of different path the practical application of the spectral subtraction approach with different flow blood volumes.