This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

Fourier transformation infrared (FT-IR) spectroscopy has been used to measure glucose concentrations in different matrices. The accuracy of the FT-IR technique does not meet the requirements of medical applications, so we have developed a new, efficient and precise method based on attenuated total reflectance coupled with wavelet transformation (ATR-WT-IR). One thousand interferograms, divided into training- and testing-sets, have been recorded from four glucose concentrations using an ATR-IR unit. Signals were subjected to (WT) and neural network (NN) study in order to design correlation algorithm. The Pearson’s Correlation Coefficient (PCC) obtained by judging the predicted- against the real-concentrations was 0.9954, with a mean square error of 8.4e-005. The proposed ATR-WT-IR method shows efficiency in glucose prediction and could possibly to be integrated into a non-invasive monitoring technique.

Diabetes is a chronic disease that results from deficiency in insulin production or the presence of ineffective insulin. This will lead to uncontrolled amounts of glucose in blood. Control of sugar levels in blood is essential for diabetes mellitus patients to prevent enormous hazardous side effects [

Although the cost of glucose monitoring tests seems high, it is still incomparable with the costs of diabetes complications. The development of techniques to monitor glucose in serum is challenging due to the complicated serum matrix, which includes several types of proteins, sugars, lipids, small molecules, as well as varied concentrations of nutrients [

In this context, infrared spectroscopy (IR) emerges as a powerful methodology. Nevertheless, for medical requirements, only reliable methods with high accuracy and precision are acceptable [

Glucose was purchased from Sigma-Aldrich (Ingelheim/Germany). IR spectra were recorded on TENSOR Series ATR FT-IR Spectrometer (Bruker/Germany). Aqueous solutions of 0.05, 0.10, 0.20, and 0.40 M glucose were prepared in Milli-Q water.

Wavelets and wavelet transforms are relatively new powerful tools for hierarchical multi-resolution analysis of signals. They have been widely applied to a diverse fields, including signal processing, physics, astronomy, medical imaging, and image processing. Wavelets have achieved great success in image compression, enhancements, analysis, classifications, and retrieval.

Wavelets

The signal transformation process takes place in a multi level decomposition. At the first level the signal is decomposed into approximation and detailed coefficients, for the second level the approximation coefficients decomposed again into approximation and detailed coefficients, the decomposition process iterated until reach the pre determined levels number.

In this work as an example, four glucose concentrations were measured 250 times each using IR-ATR technique. The resulting interferogram signals were transformed by the WT algorithm. As it is shown in

For further signal transformation analysis, the technique use the detail and approximation coefficients at each level (

The overall architecture of the work is shown in

Iterations for learning stop if the goal of the best PCC is met or the number of iterations (epochs) is reached. The work is implemented into two stages, in the first stage the influence of the wavelet coefficients are examined by training the NN with 100 epochs using the approximation coefficient at level six, after that, adding the detailed coefficient for the next upcoming levels. The second stage is to inspect the influence of the number of epochs in predicting the accuracy of the concentration.

The NN is trained first to predict the influence of the approximation and detailed coefficients in representing the original signal and its relation with the glucose concentration. The correlation coefficient is increased and the MSE is decreased when adding the detailed coefficient at each step. PCC using approximation coefficient for testing data at level six was 0.5230756, when adding the detail coefficient for the same level it became 0.558089, and 0.99575 when adding detail coefficient from level five. The MSE was 1.686e-4 and 1.215e-2 at level five and six respectively (

In this work, wavelet transformation has been implemented for ATR-IR signal transformation. A neural network has been constructed and used for concentration prediction where the correlation coefficient between the two actual and predicted concentrations was 0.9954. Therefore, ATR-IR coupled with wavelet transformation would be a promising tool to increase the accuracy of non-invasive glucose monitoring devices.

We are grateful to the German-Jordanian University for financial support and for Dr. Fayz Idris—Department of Computer Sciences, for his invaluable help and discussions.

WT transformation of IR-ATR signals of a glucose solution.

Three levels in the Wavelet Decomposition of IR-ATR signals.

Architecture of data analysis and treatment.

Mean Square Error (MSE) for different set of coefficients.

The correlation coefficient between the actual and predicted glucose concentration.