Near infrared (NIR) spectroscopy has become a promising technique for blood glucose monitoring. However, an appropriate model of spectral response in humans is yet to be determined because of the reliability problem. In this study, 48 subjects were recruited. The subjects' left forearms were scanned using near infrared spectroscopy to obtain NIR spectra. Simultaneously, a blood sample of glucose was drawn. A new method based on Monte Carlo approach is applied for partial least squares (PLS), named as PLSMC, is proposed. A large numbers of models are built from calibration subsets which are randomly selected from the whole calibration set in order to minimize the noises. It is then determining the mean value over the models with high correlation and small prediction errors. The results show that the method can enhance the stability of PLS model. Also, the performance of the PLSMCshows more accurate prediction results as compared with conventional PLS.
Deep Learning Analysis for Blood Glucose Monitoring Using Near Infrared Spectroscopy
