Dou Y, Qu N, Wang B, Chi Y Z, Ren Y L
College of Science, Tianjin University of Science & Technology, Tianjin 300222, China.
Eur J Pharm Sci. 2007 Nov;32(3):193-9. doi: 10.1016/j.ejps.2007.07.002. Epub 2007 Jul 12.
A method for simultaneous, non-destructive analysis of aspirin and phenacetin in compound aspirin tablets with different concentrations has been developed by principal component artificial neural networks (PC-ANNs) on near-infrared (NIR) spectroscopy. In PC-ANNs models, the spectra data were first analyzed by principal component analysis (PCA). Then the scores of the principal compounds (PCs) were chosen as input nodes for input layer instead of the spectra data. The artificial neural networks (ANNs) models using the spectra data as input nodes were also established, which were compared with the PC-ANNs models. Four different preprocessing methods (first-derivation, second-derivation, standard normal variate (SNV) and multiplicative scatter correction (MSC)) were applied to NIR conventional spectra. The result shows the first-derivative model of PC-ANNs multivariate calibration has the lowest training errors and predicting errors. The concept of the degree of approximation was introduced and performed as the selective criterion of the optimum network parameters.
利用主成分人工神经网络（PC - ANNs）结合近红外（NIR）光谱技术，开发了一种可同时、无损分析不同浓度复方阿司匹林片中阿司匹林和非那西丁的方法。在PC - ANNs模型中，首先通过主成分分析（PCA）对光谱数据进行分析。然后，选择主成分（PCs）的得分作为输入层的输入节点，而非光谱数据。同时还建立了以光谱数据作为输入节点的人工神经网络（ANNs）模型，并与PC - ANNs模型进行比较。对NIR常规光谱应用了四种不同的预处理方法（一阶导数、二阶导数、标准正态变量变换（SNV）和多元散射校正（MSC））。结果表明，PC - ANNs多元校正的一阶导数模型具有最低的训练误差和预测误差。引入了近似度的概念，并将其作为优化网络参数的选择标准。
