Abstract

Several orthogonal signal correction (OSC) methods coupled with partial least-squares (PLS) were applied to near-infrared spectra in a longitudinal study for the prediction of glucose, urea, and triacetin in liquid samples. The OSC methods included Wold’s OSC, Fearn’s OSC, orthogonal projection to latent structures (OPLS), and direct orthogonal signal correction (DOSC). These preprocessing methods are designed to simplify the spectra and remove information that is orthogonal to the analyte. Samples consisted of various concentrations of glucose, urea, and triacetin in the range from 1 to 19 mM. Calibration and prediction data were collected over 1.7 years to test how well each of the OSC methods helped to maintain analyte prediction performance over time. When comparing the OSC methods, it was found that all of the OSC methods improved the predictions in most data sets throughout the study but that DOSC clearly outperformed the others. On the basis of 18 prediction sets collected across 617 days after the collection of the calibration data, the average percent improvement in the standard error of prediction (SEP) when applying DOSC was 4.6%, 83.1%, and 43.0%, respectively, for glucose, urea, and triacetin. These comparisons were made relative to the SEP values obtained with unprocessed spectra.

摘要

在一项纵向研究中，将几种正交信号校正 (OSC) 方法与偏最小二乘法 (PLS) 结合应用于近红外光谱，以预测液体样品中的葡萄糖、尿素和甘油三乙酸酯。OSC 方法包括 Wold 的 OSC、Fearn 的 OSC、对潜在结构的正交投影 (OPLS) 和直接正交信号校正 (DOSC)。这些预处理方法旨在简化光谱并去除与分析物正交的信息。样品由 1 至 19 mM 范围内的各种浓度的葡萄糖、尿素和甘油三乙酸酯组成。校准和预测数据收集了 1.7 年，以测试每种 OSC 方法在长期保持分析物预测性能方面的效果。在比较 OSC 方法时，结果发现，在整个研究过程中，所有 OSC 方法都改进了大多数数据集的预测，但 DOSC 明显优于其他方法。在收集校准数据后 617 天内收集的 18 个预测集的基础上，应用 DOSC 时，葡萄糖的平均预测标准误差 (SEP) 改善百分比分别为 4.6%、83.1% 和 43.0% 、尿素和三醋精。这些比较是相对于未经处理的光谱获得的 SEP 值进行的。用于葡萄糖、尿素和三醋精。这些比较是相对于未经处理的光谱获得的 SEP 值进行的。用于葡萄糖、尿素和三醋精。这些比较是相对于未经处理的光谱获得的 SEP 值进行的。