Abstract:
Blood glucose is the primary energy source of the human body. With the rise in diabetes,
a condition that causes atypical blood glucose levels, the need to monitor and maintain a
healthy glucose level has become paramount, especially for diabetics, as prolonged high
glucose levels can lead to organ degradation and cardiac arrest.
The most popular and reliable way to monitor one’s glucose level is the finger-prick
method; however, it is both intrusive and costly. Patients might need to prick their fingers
multiple times per day, which causes discomfort and without proper hygienic measures,
can cause other diseases. Furthermore, the cost of replacing the lancets and strips could
quickly add up, becoming an extra financial burden.
Our project introduces a novel glucose monitoring solution using NIRs and Machine
Learning techniques to overcome these shortcomings. By utilizing the property of glucose
to absorb light in the Near Infrared (NIR) region, in conjunction with other parameters
that have a high correlation to glucose levels, we have successfully implemented an
alternative to the classic finger-pricking method. This thesis delves into the exploration,
implementation and evaluation of our approach. This encompasses the hardware implementation,
dataset collection, machine learning and the development and deployment of
a mobile application.
The results of this study demonstrate the potential of NIRs and the strategic application
of machine learning methodologies in providing an adaptable solution to blood glucose
monitoring without the need for invasive methods. It is anticipated that the findings and
methodologies presented will help in advancing the nascent field of NIRs applications as
well as promoting more non-invasive alternatives to popular intrusive biomedical methodologies
that are reliable and inexpensive.
