Contribution au développement d’un glucomètre non invasif.

Abstract

In this thesis, we propose a new methodology for the non-invasive detection of blood glucose using an electronic nose based on WO3, SnO2 and ZnO transducers. This approach is an indirect detection of blood glucose by measuring acetone vapors present in human breath. In order to validate this approach, it was necessary to highlight the feasibility and the relevance of this detection methodology by the elaboration of three types of thin films based on WO3, SnO2 and ZnO. During the work carried out, the sensitive layers of WO3, SnO2 and ZnO were deposited by RF magnetrons puttering on silicon substrates with a thickness of 50 nm for each layer. After the deposition, the obtained transducers were tested in the presence of acetone and ethanol. The experimental results prove that our transducers are able to detect concentrations of 1 ppm of acetone and ethanol, under the assumption that ethanol and water vapor are considered as interfering. These transducers can therefore form a sensor array as a central element in the design of a non-invasive glucose sensing principle. Indeed, the use of a sensor array instead of a single sensor during the measurements allows increasing the sensitivity and selectivity. The responses from the different sensors constituting an electronic nose are used to create a database. Then, a multivariate analysis was performed to identify the gases and to estimate their concentrations. First, an extraction of six different features of the signal has been applied in order to obtain the most useful information of the signal, subsequently the ReliefF algorithm is used for the selection of the most significant features. For gas classification, a support vector machine (SVM) based method using a linear kernel function is employed, then to estimate the concentration of acetone and ethanol, a new method based on the combination of the best features of three sensors is proposed to create a least squares SVM (LS-SVM) based prediction model. Classification accuracy of 100% is achieved with a root mean square error for acetone and ethanol concentration estimation of 0.2236 and 0.6639 respectively. From these results, we have demonstrated that the proposed method is a promising approach for non-invasive detection of blood glucose level in human blood.