Research Article  Open Access
Qingbo Li, Qishuo Gao, Guangjun Zhang, "Improved Extended Multiplicative Scatter Correction Algorithm Applied in Blood Glucose Noninvasive Measurement with FTIR Spectroscopy", Journal of Spectroscopy, vol. 2013, Article ID 916351, 5 pages, 2013. https://doi.org/10.1155/2013/916351
Improved Extended Multiplicative Scatter Correction Algorithm Applied in Blood Glucose Noninvasive Measurement with FTIR Spectroscopy
Abstract
In order to improve the predictive accuracy of human blood glucose quantitative analysis model with fourier transform infrared (FTIR) spectroscopy, this paper uses a method named improved extended multiplicative scatter correction (ImEMSC), which can effectively eliminate the scattering effects caused by human body strong scattering. The principal components of the differential spectra are used instead of the pure spectra of the analytes in this algorithm. Calibrate the unwanted physical characteristic through the shape of the curve of principal components, and extract the original glucose concentration information. ImEMSC can efficiently remove most of the pathlength difference and baseline shift influences. Firstly, ImEMSC is used as a preprocessing method, and then partial least squares (PLS) regression method is adopted to establish a quantitative analysis model. In this paper, the result of ImEMSC is compared with those popular scattering correction algorithms of multiplicative scatter correction (MSC) and extended multiplicative scatter correction (EMSC) preprocessing methods. Experimental results show that the prediction accuracy has been greatly improved with ImEMSC method, which is helpful for human noninvasive glucose concentration detection technology.
1. Introduction
Diabetes and its complications have been a heavy burden on the society. According to the International Diabetes Federation (IDF) latest statistics, there are 371 million individuals with diabetes worldwide in 2012 [1]. The control of blood glucose levels relies on blood glucose measurement. The tradition fingerprick way to measure blood glucose level is painful, potentially dangerous, and expensive to operate. In the last decades, many noninvasive methods [2–7] have been studied to measure blood glucose level.
Optical methods have been developed into the most powerful optical techniques of biomedical research and clinical application in noninvasive approaches for glucose monitoring in the last twenty years [8]. This noninvasive glucose measurement eliminates the painful pricking experience, risk of infection, and damage to finger tissue. The optical measurement of blood glucose is based on the light magnitude absorbed by glucose in blood at glucose absorption peaks, but the measurement accuracy is still a barrier due to the weak signal from blood and interference of other blood components [9]. The midinfrared (MIR) spectroscopy method is one of the most promising optical approaches. The absorption of glucose can be less influenced by other substances in midinfrared region, with the narrow absorption peak [10], which makes it more easy to extract the glucose concentration information from the blood spectra. But because of the human body strong scattering effect, nonlinear relationship exists between glucose concentration and absorbance spectra [11, 12]. In order to successfully use FTIR spectroscopy technique in noninvasive blood glucose measurement, the calibration model must provide a stable and predictive capacity. Therefore, it is important to eliminate the human body strong scattering impact and improve the robustness of the model. An improved extended multiplicative scatter correction (ImEMSC) method is used to effectively solve these problems in this paper.
2. Experiment
2.1. Experimental Instrument
An attenuated total reflectance (ATR) accessory linked to a Thermo Nicolet 6700 FTIR spectrometer, produced in the United States, was used. The ATR accessory was made of ZnSe crystal. The FTIR spectrometer is equipped with a liquid nitrogencooled mercury cadmium telluride detector. The scanning range is 400~4000 cm^{−1}, with 16 scan times, a resolution of 4 cm^{−1}, and a gain of 1.
The reagent used is oral glucose powder produced by Peking University Third Hospital in Beijing, China.
2.2. Acquisition of FTIR Spectra
Experiment procedure is described here. A healthy volunteer had been fasting for 8 h before this experiment began. The measurement site is the middle finger of right hand, which was in close contact with the cleaned ATR crystal in the experiment. Then he drank 100 mL water with 75 g glucose within 5 minutes; in succession the FTIR spectra were collected from the finger every 12 minutes after cleaning the finger. At the time of sampling, the measurement position, measurement pressure, and the psychology of the volunteer were kept invariable as far as possible. In the meantime the corresponding blood glucose reference values were measured by the OneTouch Ultra 2 blood glucose meter produced by Johnson and Johnson Company, USA.
The experiment took 3 hours and a total of 17 FTIR spectra were collected, including 11 spectra acquired in the first day with a glucose concentration range of 91.8~140.4 mg/dL and 6 spectra acquired in the second day with a glucose concentration range of 97.2~142.2 mg/dL.
3. Theory
3.1. ImEMSC Algorithm
Stark and Martens developed multiplicative scatter correction (MSC) into the extended multiplicative scatter correction (EMSC) in 1989 [13]. The EMSC method employs the pure spectra of the analytes and interference effects to improve the optical pathlength estimation. Then, it is possible to reduce or eliminate the pathlength difference due to human body strong scattering in the preprocessing stage [14]. However, EMSC cannot be wildly used due to a lack of the pure spectrum of chemical matter. In this paper, an improved EMSC (ImEMSC) has been adopted. For this method, the principal components of the differential spectra are used instead of the pure spectrum of the interested analytes. Consequently, the scattering effects are corrected without any pure spectrum information.
The algorithm principle is as follows [15, 16].
The infrared spectral analysis is based on Lambert Beer's law [17]; under ideal conditions, the absorbance data can be seen as a sum of the contributions from the different chemical constituents with spectra and concentrations :
Actually, there is a certain translation and rotation relationship between the measured spectra and the ideal spectra, taking into account that the scattering coefficients at all wavelengths are not the same; EMSC method expresses the measured spectra as follows: where is a measured spectrum, is ideal spectrum, is identity matrix, is the wavelength, and , , , and are scalar parameters obtained by calibration.
So the equation can be rewritten as follows:
Spectrum (1) may be rewritten as a deviation from a reference spectrum , which could be, for example, the average of a set of empirical spectra as follows:
In (4), represents the deviations in the analyte and interference concentration compared with that of the reference sample.
Define the differential spectrum matrix : where can be processed by principal components analysis (PCA); each load of the components can represent a specific factor. Then revoice as follows: where is the principal component and is the coefficient of the principal components.
Select the numbers of as needed; these principal components not only contain the concentration differential information, but also include the different physical aspects (optical pathlength, the surface state, etc.). Determine the principal components that represent concentration information and physical characteristics through the shape of the curve of principal components. Calibrate the unwanted physical characteristic according to the actual needs and highlight information of the chemical concentration. Replace by and combine (3) and (6); then where .
Let , and let ; construct calibration models using a multivariate linear calibration method such as PLS. The unnecessary principal components are defined as the interference factors and the left are defined as the effective factors. Correction spectrum can be obtained by subtracting the interference factors.
3.2. Software
The scattering correction algorithms and all the calculations were implemented in Matlab 2011 b.
4. Results and Discussion
4.1. Model Selection and Comparison
Two data sets were prepared. One is training set, consisting of 11 spectra acquired in the first day. The other is test set, consisting of the rest 6 spectra acquired in the second day. The wavelength region of 800–2000 cm^{−1} was selected for calibration and predication. The original spectra are shown in Figure 1.
(a)
(b)
In this paper, the raw spectra were corrected by ImEMSC first. The corrected spectra are shown in Figure 2.
(a)
(b)
Figure 2 shows the preprocessed spectra by ImEMSC. All the spectra were normalized to an average estimated baseline level and an average estimated pathlength level. The variability in the spectra was much smaller.
PLS regression was constructed. Prediction results of the test set are shown in Table 1. Figure 3 gives the detailed comparison of the three preprocessing methods: ImEMSC, EMSC, and MSC.

(a)
(b)
(c)
(d)
4.2. Analysis of Experimental Results
The predicted results of PLS regression after preprocessing by three methods are displayed in Table 1. RMSEP denotes the predictive accuracy of the calibration model. denotes the correlation coefficient.
Table 1 shows that the best preprocessing method for scattering correction is ImEMSC. Compared with the results from original data, RMSEP decreases from 9.3 mg/dL to 8.8 mg/dL. The predication accuracy increases by 5.4%. In addition, increases from 0.86 to 0.95. The success of ImEMSC is attributed to its avoiding of the request of matter pure spectrum, which limits the use of EMSC.
Because MSC and EMSC generated the overcorrection phenomenon, the prediction accuracy reduced in the experiment. As far as the methods of EMSC and MSC are concerned, scattering effects are assumed as a shift in the baseline and the average spectrum is used as a reference spectrum to eliminate the shift in the algorithm [18]. When the range of the concentrations of an interesting target is large, sometimes the actual chemical absorption information is corrected as a baseline shift. The spectra with smaller concentrations information are overcorrected, which result in the low predication accuracy.
5. Conclusions
Due to human body strong scattering, the optical pathlength difference and baseline shift exist in blood glucose noninvasive measurement. In order to solve the problems and increase the predication accuracy, an appropriate method must be adopted in the preprocessing stage. ImEMSC takes account of the wavelength effects, and simultaneously the principal components are used instead of a prior knowledge about the analytes information. The scattering effects are absolutely corrected without any pure spectrum information. ImEMSC is promising because it can raise the model prediction capability and robustness in human blood glucose noninvasive measurement using FTIR spectroscopy.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgment
This work is supported by Programs for Changjiang Scholars and Innovative Research Team (PCSIRT) in the University of China (IRT0705).
References
 International Diabetes Federation, The IDF Diabetes Atlas, International Diabetes Federation, 5th edition, 2012.
 S. F. Malin, T. L. Ruchti, T. B. Blank, S. N. Thennadil, and S. L. Monfre, “Noninvasive prediction of glucose by nearinfrared diffuse reflectance spectroscopy,” Clinical Chemistry, vol. 45, no. 9, pp. 1651–1658, 1999. View at: Google Scholar
 S.J. Yeh, C. F. Hanna, and O. S. Khalil, “Monitoring blood glucose changes in cutaneous tissue by temperaturemodulated localized reflectance measurements,” Clinical Chemistry, vol. 49, no. 6, pp. 924–934, 2003. View at: Publisher Site  Google Scholar
 K. Maruo, M. Tsurugi, J. Chin et al., “Noninvasive blood glucose assay using a newly developed nearinfrared system,” IEEE Journal on Selected Topics in Quantum Electronics, vol. 9, no. 2, pp. 322–330, 2003. View at: Publisher Site  Google Scholar
 I. A. Vitkin and R. C. N. Studinski, “Polarization preservation in diffusive scattering from in vivo turbid biological media: effects of tissue optical absorption in the exact backscattering direction,” Optics Communications, vol. 190, no. 1–6, pp. 37–43, 2001. View at: Publisher Site  Google Scholar
 A. M. K. Enejder, T.W. Koo, J. Oh et al., “Blood analysis by Raman spectroscopy,” Optics Letters, vol. 27, no. 22, pp. 2004–2006, 2002. View at: Google Scholar
 R. O. Esenaliev, K. V. Larin, I. V. Larina, and M. Motamedi, “Noninvasive monitoring of glucose concentration with optical coherence tomography,” Optics Letters, vol. 26, no. 13, pp. 992–994, 2001. View at: Google Scholar
 N. S. Oliver, C. Toumazou, A. E. G. Cass, and D. G. Johnston, “Glucose sensors: a review of current and emerging technology,” Diabetic Medicine, vol. 26, no. 3, pp. 197–210, 2009. View at: Publisher Site  Google Scholar
 C.F. So, K.S. Choi, T. K. S. Wong, and J. W. Y. Chung, “Recent advances in noninvasive glucose monitoring,” Medical Devices, vol. 5, pp. 45–52, 2012. View at: Google Scholar
 G. Li, M. Zhou, H.J. Wu, and L. Lin, “The research status and development of noninvasive glucose optical measurements,” Spectroscopy and Spectral Analysis, vol. 30, no. 10, pp. 2744–2747, 2010. View at: Publisher Site  Google Scholar
 K. Kim, J.M. Lee, and I.B. Lee, “A novel multivariate regression approach based on kernel partial least squares with orthogonal signal correction,” Chemometrics and Intelligent Laboratory Systems, vol. 79, no. 12, pp. 22–30, 2005. View at: Publisher Site  Google Scholar
 X.Y. Zhang, Q.B. Li, and G.J. Zhang, “Modified robust continuum regression by net analyte signal to improve prediction performance for data with outliers,” Chemometrics and Intelligent Laboratory Systems, vol. 107, no. 2, pp. 333–342, 2011. View at: Publisher Site  Google Scholar
 C. E. Miller and T. Naes, “A pathlength correction method for nearinfrared spectroscopy,” Applied Spectroscopy, vol. 44, no. 5, pp. 895–898, 1990. View at: Google Scholar
 H. Martens, J. P. Nielsen, and S. B. Engelsen, “Light scattering and light absorbance separated by extended multiplicative signal correction. Application to nearinfrared transmission analysis of powder mixtures,” Analytical Chemistry, vol. 75, no. 3, pp. 394–404, 2003. View at: Publisher Site  Google Scholar
 H. Martens, S. W. Bruun, I. Adt, G. D. Sockalingum, and A. Kohler, “Preprocessing in biochemometrics: correction for pathlength and temperature effects of water in FTIR biospectroscopy by EMSC,” Journal of Chemometrics, vol. 20, no. 8–10, pp. 402–417, 2006. View at: Publisher Site  Google Scholar
 Q.X. Zhang, G.J. Zhang, and Q.B. Li, “A pathlength correction method on biochemical parameter nondestructive measuring of folium,” Spectroscopy and Spectral Analysis, vol. 30, no. 5, pp. 1310–1314, 2010. View at: Publisher Site  Google Scholar
 H. Martens and E. Stark, “Extended multiplicative signal correction and spectral interference subtraction: new preprocessing methods for near infrared spectroscopy,” Journal of Pharmaceutical and Biomedical Analysis, vol. 9, no. 8, pp. 625–635, 1991. View at: Publisher Site  Google Scholar
 Q.X. Zhang, Q.B. Li, and G.J. Zhang, “Scattering impact analysis and correction for leaf biochemical parameter estimation using VisNIR spectroscopy,” Spectroscopy, vol. 26, no. 7, pp. 28–39, 2011. View at: Google Scholar
Copyright
Copyright © 2013 Qingbo Li et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.