选取a, b, c和d四种类型近红外光谱仪, 分别采用透射模式测定清开灵注射液近红外光谱, 以高效液相色谱法测定值作为参考值, 建立不同仪器类型清开灵注射液中黄芩苷偏最小二乘(PLS)和间隔偏最小二乘(iPLS)定量模型, 并计算模型的多变量检测限(MDL)。 四种仪器的PLS模型决定系数(R2)和预测均方差(SEP)分别为0.976 2和230.4 μg·mL-1(a), 0.956 1和246.4 μg·mL-1(b), 0.966 2和264.4 μg·mL-1(c), 0.998 5和71.5 μg·mL-1(d), 其中d型仪器较其他三种类型能获得更好的模型性能。 经iPLS变量筛选后, a和b两种类型仪器得到的iPLS模型R2pre和SEP分别为0.977 1和218.4 μg·mL-1, 0.975 4和219.4 μg·mL-1, 相较其PLS模型预测性能未见明显提高; c和d未筛选出变量。 不同仪器的MDL(Δ0.05, 0.05)均低于250 μg·mL-1, 其中c和d型MDL分别低至58和2.9 μg·mL-1。 表明不同类型仪器定量预测性能和MDL不同。 创新性采用多变量检测限理论探讨了不同类型近红外仪器的检测性能, 这一方法具有可行性。 在实际应用中应根据研究载体的特征选择合适的仪器类型, 以确保定量准确性。

Multivariate detection limits (MDLs) of different types of near-infrared instruments were investigated to guide the selection of device type for TCM NIR analysis. In this paper, near-infrared spectroscopy of Qingkailing injection was performed in transmission mode on four near-infrared spectrometers named a, b, c and d, respectively. High performance liquid chromatography (HPLC) was used as the reference method to determine the content of baicalin in Qingkailing injection. Partial least squares (PLS) and interval partial least squares (iPLS) quantitative models of baicalin in Qingkailing injection were established and MDLs of quantitative models based on different types of instruments were calculated. The determination coefficient of prediction(R2pre)and the root mean square errors of prediction (SEP) of PLS models of four different near-infrared spectrometers are 0.976 2 and 230.4 μg·mL-1 (a), 0.956 1 and 246.4 μg·mL-1 (b), 0.966 2 and 264.4 μg·mL-1 (c), 0.998 5 and 71.5 μg·mL-1 (d). And the model of instrument d shows a better prediction performance than the other three types. There are no remarkable superiorities in predictability in iPLS models of instruments a and b after variable selection, since the R2pre and SEP values for instruments a and b are 0.977 1 and 218.4 μg·mL-1, and 0.975 4 and 219.4 μg·mL-1, respectively. Models c and d show no results of variable selection. MDLs (Δ0.05, 0.05) of different instruments are all less than 250 μg·mL-1, and the MDLs of instruments c and d reach to 58 and 2.9 μg·mL-1 respectively. The results reveal that the predictability of models and corresponding MDLs are different for different detection equipments. This paper innovatively used the theory of MDL to investigate the detection performance of different types of NIR instruments. The results demonstrated the feasibility of the approach. And it is expected that in actual applications, choosing the right type of instrument should be based on the characteristics of the study carrier to ensure quantitative accuracy.