iGLU 4.0: A continuous glucose monitoring and balancing paradigm with physiological parameters

TL;DR

This paper introduces a non-invasive, continuous glucose monitoring system using NIRS technology combined with physiological parameters and a deep neural network to improve diabetes management remotely.

Contribution

It presents a novel non-invasive glucose monitoring paradigm integrating physiological data and decision modeling, enhancing remote diabetes care.

Findings

Achieved MARD of 12.50% and AvgE of 12.10% with DNN model

Coefficient of determination R2 of 0.97 indicating high accuracy

Enables remote monitoring and decision support for diabetes management

Abstract

The conventional method of glucose measurement such as pricking blood from the body is prevalent which brings pain and trauma. Invasive methods of measurement sometimes raise the risk of blood infection to the patient. Sometimes, some of the physiological parameters such as body temperature and systolic blood pressure (SBP) are responsible for blood glucose level fluctuations. Moreover, diabetes for a long duration usually becomes a critical issue. In such situation, patients need to consult diabetologist frequently, which is not possible in normal life. Therefore, it is required to develop non-invasive glucose balancing paradigm, which measures blood glucose without pricking blood along with physiological parameters measurement and decision model. The proposed paradigm helps to doctor, who is even available at remote location. There will not be any need to consult frequently. In the way of optimized non-invasive system design, an NIRS technique with specific wavelengths along with physiological parameters is taken to predict the precise glucose value. The all parameters (glucose, Blood pressure and body temperature), food intake and insulin levels are parts of decision model, which would help to the doctor to take decision related to the further medicine doses and diet plan. The patients would have suggestions according to maintain their blood glucose level. The proposed system demonstrated an accurate model with MARD and AvgE 12.50% and 12.10% respectively using DNN model. Coefficient of determination R2 has been found 0.97.