Joint analyses model for total cholesterol and triglyceride in human serum with near-infrared spectroscopy.
Joint analyses model for total cholesterol and triglyceride in human serum with near-infrared spectroscopy.
复制标题
DOI：
10.1016/j.saa.2016.01.022
复制
发表时间：
2016-04
期刊：
影响因子：
--
通讯作者：
Lijun Yao;Ning Lyu;Jiemei Chen;T. Pan;Jing Yu
中科院分区：
文献类型：
--
作者：
Lijun Yao;Ning Lyu;Jiemei Chen;T. Pan;Jing Yu
小型专用近红外光谱仪的开发具有广阔的应用前景，如联合分析人血清中的总胆固醇（TC）和甘油三酯（TG），以预防和治疗大量人群的高脂血症。合适的波长选择是研制这种光谱仪的关键技术。为此，提出了一种新的波长选择方法，即等距组合偏最小二乘法（EC-PLS），用于近红外分析人血清中TC和TG的波长选择。对标定集和预测集进行了严格的划分，以实现稳定的建模优化。通过应用EC-PLS，建立了一个模型集，该模型集由与最优模型等效的各种模型组成。进一步选取两个指标的联合分析模型，仅50个波长。从建模过程中排除的随机验证样本用于验证所选模型。预测的均方根误差、相关系数和性能偏差比分别为0.197 mmol L−1、0.985和5.6，TG为0.101 mmol L−1、0.992和8.0。高脂血症的敏感性和特异性分别为96.2%和98.0%。这些结果表明预测精度高，模型复杂度低。所提出的波长选择为设计小型专用高脂血症光谱仪提供了有价值的参考。方法框架和优化算法具有通用性，可应用于其他领域。
The development of a small, dedicated near-infrared (NIR) spectrometer has promising potential applications, such as for joint analyses of total cholesterol (TC) and triglyceride (TG) in human serum for preventing and treatinghyperlipidemiaof a large population. The appropriate wavelength selection is a key technology for developing such a spectrometer. For this reason, a novel wavelength selection method, named the equidistant combination partial least squares (EC-PLS), was applied to the wavelength selection for the NIR analyses of TC and TG in human serum. A rigorous process based on the various divisions of calibration and prediction sets was performed to achieve modeling optimization with stability. By applying EC-PLS, a model set was developed, which consists of various models that were equivalent to the optimal model. The joint analyses model of the two indicators was further selected with only 50 wavelengths. The random validation samples excluded from the modeling process were used to validate the selected model. The root-mean-square errors, correlation coefficients and ratio of performance to deviation for the prediction were 0.197 mmol L− 1, 0.985 and 5.6 for TC, and 0.101 mmol L− 1, 0.992 and 8.0 for TG, respectively. The sensitivity and specificity forhyperlipidemiawere 96.2% and 98.0%. These findings indicate high prediction accuracy and low model complexity. The proposed wavelength selection provided valuable references for the designing of a small, dedicated spectrometer forhyperlipidemia. The methodological framework and optimization algorithm are universal, such that they can be applied to other fields.