Machine learning approach for non-invasive detection of blood glucose concentration using microwave sensor

Abstract

People suffering from hyperglycaemia or diabetes mellitus are increasing day by day. The only commercial devices available to measure blood glucose levels are based on invasive methods, such as collecting blood samples from an individual and testing it. However, for a person, whose blood glucose levels have to be monitored at a regular interval, the conventional invasive methods are painful, sore and thus not preferred. In order to overcome these problems, non-invasive methods have to replace the conventional forms. The non-invasive methods have not taken a commercial form yet. This project is an attempt to develop a non-invasive glucose measurement method. A non-invasive blood glucose measuring method based on Microwave transmission and then applying Machine-learning technique for the data obtained is proposed for monitoring the patients' blood glucose level. With this method, a non-invasive measurement of the blood glucose determination of the earlobe portion can be realized by analysing the received microwave signals. In this project, the coefficients of the third order Cole-Cole equation are derived to model the dielectric properties of human tissues. 'Particle swarm optimization' technique is used to determine the coefficients for the glucose concentration dependent equations. With these estimated dielectric values of human tissues, Human ear lobe portion is modelled in a simulation setup and tested over a wide range of frequencies to check for the region of linearity. The proposed method is validated by applying it to a solution prepared, which is impersonating the dielectric properties of blood plasma and it is observed that the region of linearity exists from 6 - 8 GHz. The proposed method of detecting blood glucose concentration is very convenient and is harmless to the patients.