Near infrared glucose signature identification by subspace projection method

Nowadays the primary non-invasive technique applied to measuring glucose in blood makes use of the near infrared spectrometer, with the collected data analyzed by the partial least squares (PLS) method. The PLS is a stochastically regressive estimation scheme so that some constraints must be considered prior to its application. Herein, the subspace projection (SP) method, which is widely used in array signal processing, is adopted to identify the near infrared glucose signature. The primary reason for using this technique is that it best fits the signal model of the Fourier transform infrared spectrometry (FTIR). When the desired near infrared signature is well established, the computational speed and accuracy of the SP method can be satisfied. For known and unknown constitution, two distinct methods are proposed and evaluated by using simulation and realistic data. The comparison between the SP and PLS method is also included. Experimental results demonstrate that the SP is a highly efficient method for detecting glucose and analyzing its concentration in blood.