Design and Algorithms of the Device to predict Blood Glucose Level based on Saliva Sample using Machine Learning

Regular tracking of blood glucose level is an integral part of treatment of Diabetes Mellitus, a chronic disease that occurs either when the pancreas does not produce enough insulin. Commercially available devices are involve pricking and analyzing of blood sample. Observing the pain, cost and chance of infection, non-invasive devices to measure blood sugar level are under research and development for the past decade. The paper proposes a method to develop a portable and economic device that can determine the blood glucose level of the patient by analyzing a saliva sample deposited by patient using spectroscopic methods. The device is constructed considering two theories as basis. The first theory is relation of glucose concentration of a solution to the attenuation observed while performing absorption spectroscopy. The second theory is the existence of relation between salivary glucose level and blood glucose level. Using the first theory, a device was constructed which uses NIR spectroscopy to find concentration of glucose in the given solution. The attenuation and concentration have been correlated using Machine Learning, $\mathrm{R}2=0.96$ . Data of salivary glucose level and blood glucose level was acquired and correlated using machine learning and $\mathrm{R}2=0.87$ . The proposed device requires patient to deposit saliva in a test tube and place it in the device. The device first predicts the glucose concentration of the solution(saliva). The device then uses the correlation between salivary glucose level and blood glucose level to find blood glucose level. The data is displayed and stored in the cloud.