[Multidimensional Scaling Linear Regression Applied to FTIR Spectral Quantitative Analysis of Clinical Parameters of Human Blood Serum].

In the present paper it was proposed to establish multidimensional scaling linear regression (MDS-MLR) models by combining multidimensional scaling (MDS) with multiple linear regression (MLR), in which MDS owns the ability of dimensionality reduction on spectral variables. Thus the informative spectral wavelengths can be selected for each of the 4 clinical parameters (i. e. glucose, LDL cholesterol, triglycerides and urea) by MDS-MLR method, and the effect of spectral colinearity can be restrained, so that the calibration models can be optimized. Experiments showed that the MDS-MLR models will allow to produce appreciated modeling results when combined with the preprocessing method of moving average (MA). The optimized models were verified by the samples in validation set, and the validating correlation coefficients for each clinical parameter were all upper than 0. 9. These results indicated that FTIR predicted values and the biochemical values of each clinical parameter were highly correlated, which demonstrated that MDS-MLR method has the potential of being applied to the FTIR spectroscopic analysis of human serum. FTIR rapid determination technology combined with the MDS-MLR modeling method is expected to realize the rapid assessment of human health and sub-health level.