光子学报, 2012, 41 (6): 684, 网络出版: 2012-06-19
血红蛋白对胆红素光学检测的影响分析
Influence of Hemoglobin on Noninvasive Optical Bilirubin Sensing
血红蛋白 胆红素 光谱检测 偏最小二乘建模 Hemoglobin Bilirubin Spectral measurement Partial least squares regression modeling
摘要
针对实验得到的光谱数据,应用偏最小二乘回归方法进行多变量建模,结果发现:血红蛋白的加入使建立模型的主成分数从一个变为三个,而对模型贡献最大的波长从459 nm变为432 nm和433 nm,说明偏最小二乘模型的原始光谱数据变量引进了血红蛋白的吸收噪音.结合偏最小二乘建模的原理与评价指标分析表明:较纯胆红素溶液而言,混合溶液模型预测均方根误差增大了24.90%,相关系数减小了0.29%,加入血红蛋白后的模型对胆红素浓度预测能力降低.进一步分析血红蛋白和胆红素及其混合溶液吸收光谱发现:在检测的紫外可见光波段两者吸收光谱发生重叠,两者混合溶液在408 nm处产生一个新的吸收峰,且相比同浓度血红蛋白吸收峰而言408 nm处吸收峰发生了减色效应,这是由于血红蛋白与胆红素混合会发生某些弱的键位结合引起的,新的吸收峰改变了建模所用的主成分数,降低了模型的预测能力.结果说明在胆红素经皮测量中必须采取措施减少血红蛋白的影响,才能提高胆红素经皮检测的准确度.
Abstract
The spectral data were analyzed by using the Partial Least Squares Regression. The result shows that the addition of hemoglobin changes the number of principle component from one into three, while the largest contribution wavelength changes from 459 nm to 432 nm and 433 nm. This indicates that absorption noise of hemoglobin was introduced in the spectral data used for PLS model. Furthermore, compared with pure bilirubin solution, the Root Mean Squared Error of Prediction(RMSEP) of the model for mixed solution increases 24.90%, and the correlation coefficient of the model for mixed solution reduces 0.29%. It means that the addition of hemoglobin decreases the prediction ability of the model. Meantime, a new absorption peak at 408 nm appears, which seems there is a hypochromic effect, due to some weak key combinations between hemoglobin and bilirubin. And the new absorption peak changes the number of principal components and reduces the predictive ability of the model. Overall, the influence of hemoglobin should be reduced in order to improve the transcutaneous detection accuracy of bilirubin.
蒋景英, 龚启亮, 徐可欣. 血红蛋白对胆红素光学检测的影响分析[J]. 光子学报, 2012, 41(6): 684. JIANG Jingying, GONG Qiliang, XU Kexin. Influence of Hemoglobin on Noninvasive Optical Bilirubin Sensing[J]. ACTA PHOTONICA SINICA, 2012, 41(6): 684.