Acetone discriminator and concentration estimator for diabetes monitoring in human breath

This study investigates the core of the development of an electronic nose envisioned for non invasive blood glucose monitoring. Tungsten trioxide (WO3), tin dioxide (SnO2) and zinc oxide (ZnO) homemade gas sensors have been exposed to various concentrations of acetone and ethanol in the presence of water vapor in order to create a dataset containing the changes in the resistance of the sensitive layer of sensors during the experiment. For the pattern recognition part of the system, this work reports the extraction of six features: two conventional steady-state features, two transient features and two features extracted from transform domain. The ReliefF algorithm is then used for the selection of the most efficient features. A classification method based on support vector machine (SVM) using a linear Kernel function is employed and to estimate the gas concentration of acetone and ethanol, the same training data in the classification step are used to create a prediction model based on least squares-SVM. A classification accuracy of 100% is achieved and the concentration of acetone and ethanol is estimated using a new method based on the combination of the best features of three sensors with a root mean square error of 0.2236 and 0.6639 respectively.

Automated summary

This study successfully developed an electronic nose for non-invasive blood glucose monitoring, utilizing homemade gas sensors exposed to acetone and ethanol. Through feature extraction, selection, and classification, the system achieved 100% accuracy in estimating gas concentrations with a root mean square error of 0.2236 and 0.6639 for acetone and ethanol, respectively.