Prediction of Blood Glucose in Human Tissue Across Days by Noninvasive Near-Infrared Spectroscopy.

BACKGROUND: Near-infrared spectroscopy was used to implement noninvasive glucose measurements from 8 volunteers with type I diabetes. The objective of the study was to evaluate a practical protocol for calibrating the spectroscopic measurement. METHODS: Three trials were conducted for each of 8 subjects that spanned an average of 16 days. Each trial consisted of an oral glucose tolerance test (OGTT) performed over 4 to 5 hours during which spectra were collected continuously from a skin fold on the back of the hand. The spectra were submitted to a calibration procedure based on a combination of direct orthogonal signal correction and hybrid linear analysis. For each subject, the set of calibration spectra consisted of the data from trial 1 for that subject augmented with a set of spectra from the current trial that corresponded to the beginning fasting period of the OGTT. For each prediction day, this updating procedure would require a single reference glucose measurement and the collection of a corresponding set of spectra. RESULTS: Across the 16 trials performed after the initial OGTT, values of the standard error of prediction ranged from 1.12 to 4.11 mM and averaged 2.33 mM. The mean average relative deviation between predicted and reference glucose values ranged from 0.079 to 0.45 and averaged 0.21 (21%). Error grids computed from 6710 predicted concentrations had 96.7% of the values in the A and B regions. CONCLUSIONS: These results represent a clear step forward in the development of a practical calibration protocol for the noninvasive measurement of blood glucose.