A Weak Signal Extraction Method for Human Blood Glucose Noninvasive Measurement using Near Infrared Spectroscopy

Background interference from optical absorption of matrix components, low spectral selectivity and low spectral sensitivity are the main interference factors for human blood glucose noninvasive measurement using near infrared (NIR) spectroscopy. In order to extract the weak glucose concentration information, a modified uninformative variable elimination (mUVE) method combined with successive projections algorithm (SPA) named as mUVE-SPA, is proposed. mUVE is used to eliminate matrix background and high-frequency noise by wavelet multi-resolution technology. SPA is followed to select variables with minimum colinearity by projection algorithm in a vector space. The proposed method was applied in two NIR spectra data sets (plasma samples experiment in vitro and human blood glucose noninvasive measurement experiment in vivo) respectively. The performance and adaptability of the proposed strategy were discussed. The results indicate that the proposed hybrid method can give an alternative path to extract weak glucose information and yield more parsimonious models with higher precision.