Files and links (1)
Free to read and download, Open Access
Noninvasive glucose monitoring has been the subject of considerable research because of the high number of diabetes patients who must monitor their glucose levels daily by taking blood samples. Among methods being evaluated for possible use in this application, near-infrared (NIR) spectroscopy has received significant attention because of available glucose absorption bands that can be observed in the presence of the large aqueous background found in tissue spectra. The objective of the research presented here is to evaluate the potential for implementing a noninvasive nocturnal hypoglycemic alarm with NIR spectroscopy. Such an alarm would be used by a diabetic to detect potentially dangerous occurrences of hypoglycemia during sleep.
The approach used is to collect spectra continuously from the patient during the sleep period, followed by the application of pattern recognition methods to determine if a spectrum represents a blood glucose level that exceeds a hypoglycemic threshold. A reference spectrum is collected and a conventional finger-stick glucose concentration measurement is made at the start of the sleep period. The ratio is then taken of each subsequent spectrum to the collected reference, forming a differential spectrum corresponding to the signed difference in concentration relative to the reference. The identification of these differential spectra as "alarm" or "non-alarm" is performed with a classification model computed with piecewise linear discriminant analysis.
This methodology is initially tested with in vitro laboratory data that simulated the glucose excursions that occur during sleep. The performance of the hypoglycemic alarm methodology in the presence of varying levels of urea, glyceryl triacetate, and L-lactate as potential spectral interferents is tested. The robustness of the methodology with respect to time is also evaluated.
The thesis further discusses an experimental procedure to prepare tissue phantoms composed of two main proteins that exist in human skin tissue, keratin and collagen. A new methodology is developed to produce varying-thickness films that allowed the simulation of changes in the content of skin tissue proteins present within the optical path of the NIR measurement. The prepared films are incorporated into in vitro laboratory measurements in which varying levels of glucose, urea, keratin, and collagen are introduced in order to provide a test of the hypoglycemic alarm algorithm that simulates the spectral properties of human tissue.
Finally, the hypoglycemic alarm algorithm is tested with in vivo data collected with rat animal models. Data are presented for single-day experiments performed with anesthetized rats, as well as for multiple-day experiments conducted with awake rats. The results obtained from both the in vitro and in vivo studies confirm that if high-quality spectral data are attainable, the alarm methodology can work effectively to identify hypoglycemic events while exhibiting a low rate of false detections.