A machine-learning strategy to evaluate the use of FTIR spectra of saliva for a good control of type 2 diabetes

Abstract

The World Health Organization has declared that diabetes is one of the four leading causes of death attributable to non-communicable diseases. Currently, many devices allow monitoring blood glucose levels for diabetes control based mainly on blood tests. In this paper, we propose a novel methodology based on the analysis of the Fourier Transform Infrared (FTIR) spectra of saliva using machine learning techniques to characterize controlled and uncontrolled diabetic patients, clustering patients in groups of a low, medium, and high glucose levels, and finally performing the point estimation of a glucose value. After analyzing the obtained results with Support Vector Machines (SVM), Artificial Neural Networks (ANN), and Linear Regression (LR), we found that using ANN, it is possible to carry out the characterizations mentioned above efficiently since it allowed us to identify correctly the 540 spectra that make up our database studying the region 4000-2000 cm^-1.

Keywords: Artificial neural networks; Diabetes; FTIR spectroscopy; Non-invasive; Saliva.