Chance correlation in non-invasive glucose measurement using near-infrared spectroscopy

Although the non-invasive glucose measurement technique based on near-infrared (NIR) spectroscopy has been an active research area for over twenty years, a reliable monitoring method has not been established yet. The key problem is that the spectral variations due to glucose concentration are extremely small compared to that from other biological components. In addition, there are also some ambiguous time-dependent physiological processes, which make the explanation of the model more difficult, especially in the universal calibration. Therefore, in order to produce a model that is related to the actual spectral variation of glucose, reproducible measurements and clinical validation experiments that improve the selectivity and signal to noise ratio of glucose measurement are needed. In this paper, chance correlation in spectroscopy analysis is investigated, which is one of the obstacles to achieving successful NIR spectroscopy analysis, especially in in vivo measurement. The reasons for chance correlation in the in vitro and in vivo experiments are analysed. Methods to avoid it are suggested accordingly and verified with the in vitro experiments. We also investigate the chance correlation for the in vivo NIR diffuse reflectance spectroscopy monitoring blood. Results show that there is significant signal variation after glucose is taken, and the potential chance correlation factors including the instrument-related and physiology-related variations during the in vivo experiments do not contribute to the multivariate model for glucose concentration.