Development of Blood Glucose Monitoring System using Image Processing and Machine Learning Techniques

Glucose concentration measurement is essential for diagnosis, monitoring and treatment of various medical conditions like diabetes mellitus, hypoglycemia, etc. This paper presents a novel image-processing and machine learning based approach for glucose concentration measurement. Experimentation based on Glucose oxidase - peroxidase (GOD/POD) method has been performed to create the database. Glucose in the sample reacts with the reagent wherein the concentration of glucose is detected using colorimetric principle. Colour intensity thus produced, is proportional to the glucose concentration and varies at different levels. Existing clinical chemistry analyzers use spectrophotometry to estimate the glucose level of the sample. Instead, this developed system uses simplified hardware arrangement and estimates glucose concentration by capturing the image of the sample. After further processing, its Saturation (S) and Luminance (Y) values are extracted from the captured image. Linear regression based machine learning algorithm is used for training the dataset consists of saturation and luminance values of images at different concentration levels. Integration of machine learning provides the benefit of improved accuracy and predictability in determining glucose level. The detection of glucose concentrations in the range of 10–400 mg/dl has been evaluated. The results of the developed system were verified with the currently used spectrophotometry based Trace40 clinical chemistry analyzer. The deviation of the estimated values from the actual values was found to be around 2-3%.