A Noninvasive Glucose Monitor: Preliminary Results in Rabbits

Abstract

Background: Glucose in the aqueous humor appears to correlate with plasma glucose in humans. It is therefore a potential substrate for noninvasive optical glucose monitoring techniques. We wished to determine the potential for using rabbit aqueous humor as a model to develop a noninvasive method of measuring glucose with Raman spectroscopy. Methods: Aqueous humor from 32 rabbit eyes was removed immediately after sacrifice by rapid exsanguination under anesthesia. Raman spectroscopy was performed on the aqueous humor using near infrared (NIR) excitation wavelengths. The Raman spectra from 785 nm and 787.2 nm were subtracted one from the other to eliminate broadband fluorescence. The spectra were then analyzed with linear and nonlinear multivariate analysis and assessed for ability to predict actual aqueous humor glucose concentration. Nine other rabbits were anesthetized with xylazine in order to cause elevation of blood glucose by blocking release of insulin. Blood and aqueous humor glucose were measured at various times after injection of xylazine. Correlation of aqueous humor glucose with simultaneous plasma glucose was assessed. Results: Partial least squares analysis of raw Raman spectra of aqueous humor showed fair correlation with actual glucose concentration (r2 = 0.76). When the fluorescence spectrum was subtracted prior to linear multivariate analysis correlation was good (r2 = 0.90). When back-propagation with an artificial neural network was added to the analysis, correlation was further improved (r2 = 0.98). Aqueous humor glucose concentration exceeded blood glucose concentration at normoglycemic levels. When blood glucose rose above 200 mg/dL, aqueous humor glucose correlated linearly with plasma glucose. Conclusions: Raman spectroscopy can accurately predict glucose concentration in rabbit aqueous humor in vitro. Although rabbit aqueous humor is probably not a good model of human aqueous humor glucose physiology, its reliable correlation with plasma glucose makes it a good model on which to test noninvasive optical sensing techniques.