Abstract
The determination of blood glucose is usually performed in the laboratory or using invasive glucose meters. Unusual, fruity, or bad breath odor can be a sign of metabolic disease. The objective is the estimation of the blood glucose by breath analysis. To do that, we designed a three-gas prediction model (alcohol, acetone, and propane). We exhaled in front of the three gas sensors and then recorded the concentrations of the three volatile organic compounds (VOCs) in a database. This database included the measurements of 200 volunteer people. We then systematically studied the correlation between the glucose level obtained by blood analysis and each of these three gases. We also studied the different possible combinations of two and three gases. The best correlation was obtained for the non-diabetic subjects with a correlation coefficient (r = 0.97), for the diabetic subjects (r = 0.35). We implemented these two linear regressions and we used the validation set to estimate the accuracy of the prototype. For the non-diabetic subjects, the relative error of the measurement of the glycemia was equal to 3.93%. For the diabetic subjects, the error = 13.42%. The results have shown that the determination of glycemia by breath analysis can become a serious alternative. (c) 2020 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.