Investigation on algorithms for determination of chemical concentration by near infrared Raman spectroscopy

Near infrared (NIR) Raman spectroscopy has been used to quantify chemicals of interest in human body fluids. The measurement accuracy is greatly affected by calibration method. In this paper, we investigate new calibration methods on the determination of chemical concentration based on NIR Raman spectroscopy, in which the intrinsic spectrum of the desired chemical is incorporated in the calibration model. The method of optimal filtering coupled with Partial Least Square (OFPLS) developed by our group and the hybrid linear analysis (HLA) developed by a group at Massachusetts Institute of Technology are tested to predict the ethanol concentration in human urine samples. The results show that measurement accuracies have been much improved with OFPLS and HLA over the commonly used partial least square (PLS) method, especially when the spectrum of chemical of interest is much weaker than the greatly fluctuated background