Measuring Aqueous Humor Glucose Across Physiological Levels: NIR Raman Spectroscopy, Multivariate Analysis, Artificial Neural Networks, and Bayesian Probabilities

We have elicited a reliable Raman spectral signature for glucose in rabbit aqueous humor across mammalian physiological ranges in a rabbit model stressed by recent myocardial infarction. The technique employs near infrared Raman laser excitation at 785 nm, multivariate analysis, non-linear artificial neural networks and an offset spectra subtraction strategy. Aqueous humor glucose levels ranged from 37 to 323 mg/dL. Data were obtained in 80 uL samples to anticipate the volume constraints imposed by the human and rabbit anterior chamber of the eye. Total sample collection time was 10 seconds with total power delivered to sample of 30 Mw. Spectra generated from the aqueous humor were compared qualitatively to artificial aqueous samples and an excitation offset technique was devised to counteract broadband background noise partially obscuring the glucose signature. Feature extraction and data analysis were accomplished using second order Savitsky-Golay derivatives, linear multivariate analysis (partial least squares fit) and non-linear (artificial neural network) techniques. Predicted glucose levels correlated well with expected glucose concentration (R²= 0.98, n=32).