| Summary: | Diabetes mellitus is a common chronic disease with more than 8% of the world
population suffered from it. Millions of diabetics have to undergo painful and grueling
invasive blood glucose testing several times a day to monitor their blood glucose level.
In order to solve this problem, scientists all over the world have spent a great effort on
the research of non-invasive blood glucose measurement and proposed a variety of
methods, but none of them has been proven to be reliable in clinical practice.
This dissertation attempts to develop a microwave-based non-invasive blood glucose
measurement method with Machine Learning technique, in order to increase the
accuracy of blood glucose level estimation.
In this work, by creating a human earlobe biological model and simulating it in CST
microwave studio, the dielectric behavior of blood with the variation of glucose
concentration is investigated in a high frequency range. Applying broadband sweep to
the model, a feasible operating region is found at the frequency range of 60-62 GHz.
The result of the simulation is validated by an in-vitro experiment conducted on
artificial blood plasma in the electromagnetic environment, and the data obtained is
used for training Machine Learning models. By comparing the performances of the
models, the SVM classifier is selected to be the best solution in this case. The work in
this dissertation has the potential to be used in blood glucose monitoring to reduce
their pain and warn them of upcoming life-threatening conditions.
|
|---|