Diagnosis of diabetes based on analysis of exhaled air by terahertz spectroscopy and machine learning

Results of studying the exhaled air of patients with diabetes mellitus in comparison with healthy volunteers with the use of broadband terahertz time-domain spectroscopy are presented. Typical spectral subranges in which absorption spectrum profiles of breath tests of the target and control group differ most significantly are revealed: 0.560, 0.738, 0.970, 1.070, 1.140, 1.180, and 1.400 THz. Using the principal component analysis, it is shown that the set of absorption coefficients in these regions allows one to reliably separate the target and control groups. The obtained results are compared with measurements of acetone vapors in the exhaled air of patients with diabetes mellitus and healthy volunteers.