Results in Physics 13 (2019) 102170

Contents lists available at ScienceDirect

Results in Physics
journal homepage: www.elsevier.com/locate/rinp

Application of factorial experimental design on the optical absorption from T

glucose–insulin samples in mid-infrared spectroscopy
⁎
G. Romo-Cárdenasa, , J. de D. Sánchez-Lópeza, P.A. Luquea, C.M. Gómez-Gutiérreza,
Juan I. Nieto-Hipólitoa, Santiago Camacho-Lópezb, Arturo Jiménez-Cruzc, C. Gastelum-Acostaa,
M. Vázquez-Briseñoa, C. Navarro-Cotaa
a
Facultad de Ingeniería Arquitectura y Diseño, Universidad Autónoma de Baja California, 22860 Ensenada, Mexico
b
Departamento de Óptica, Centro de Investigación Científica y de Educación Superior de Ensenada, 22860 Ensenada, Baja California, Mexico
c
Universidad Autónoma de Baja California, Facultad de Medicina y Psicología, 22427 Tijuana, Mexico

A R T I C LE I N FO A B S T R A C T

Keywords: An important challenge to be addressed in the search for an optically based, non-invasive blood glucose
Non-invasive blood glucose quantification quantification method, is related to the spectral interference between glucose and insulin in the mid-infrared
Glucose - insulin region. This work proposes the use of a factorial experimental design in order to understand the optical over-
Factorial experimental design lapping of these compounds. The results show how in the 1100–950 cm−1 range, glucose has a greater con-
mid-IR spectroscopy
tribution than insulin to the absorption spectrum, without a significant effect from their interaction. These
results and the use of the proposed technique could bring in interesting insights for biomedical applications.

Introduction
The search for an effective non-invasive, optically based method for
blood glucose measurement is a current work in progress. Several re-
search groups have reported significant advances in the development of
a technique for such purpose, particularly in the mid-IR region [1–3].
Where an important challenge to be addressed is related with the
overlapping in the absorption spectra for glucose and insulin, which has
been recently reported [4]. This is an important issue, since both
compounds belong to the glucose regulation mechanism and their
concentration in blood changes at rates of similar magnitude [5]. In-
sulin presence in blood samples, even at low physiological concentra-
tions, doesn’t allow a direct glucose quantification [4]. Therefore, it is
highly relevant to study the involvement of each compound in the
spectroscopic analysis, aiming for both glucose quantification and the
selection of a suitable signal processing and analysis techniques as re-
quired to address the spectral overlapping problem. An important
question to be answer, concerns the magnitude of the contribution to
the absorption spectrum from each compound, at similar proportions as
they are found in blood.
A statistical technique that has been reported to effectively evaluate
and compare design configurations and component effects is known as
factorial experimental design [6–8]. In this methodology, a process or
system can be represented as a model built from a combination of
inputs, from where a response or output is analyzed. The effect of each
factor or component is studied with the use of levels [7]. For k factors,
at 2 levels for each, 2k experimental readings are required [6]. From
where, a statistical analysis of the output allows to build models based
in the inputs, from where the magnitude of the effect of the inputs can
be analyzed and described. This work uses this methodology and the
configuration above described (2k), in order to understand the con-
tribution from insulin and glucose to the absorption spectrum in the
mid- infrared region.
Methods
For the factorial experimental design, two levels of concentration,
Low and High, were defined for the inputs, considering physiological
significance from the glucose regulation mechanism. Based on litera-
ture, for glucose (Pisa, Solution DX-10®), low concentration (Lg) was
defined at 80 mg/dL and 180 mg/dL for high concentration (Hg) [9].
For insulin (Lilly, Humulin N®), low concentration was defined at
30 μU/mL (Li) and 130 μU/mL for high concentration (Hi) [5]. The
combination of these levels resulted in the following order of samples:
Lg-Li, Lg-Hi, Hg-Li, Hg-Hi. The absorption spectra were obtained with the
use of a FTIR spectrometer (PerkinElmer, Spectrum Two), with a re-
solution of 4 cm−1. Readings from a total of 5 trials of each sample were
acquired and used for the study. In order to process the spectral signal

⁎
Corresponding author.
E-mail address: romo.gerardo@uabc.edu.mx (G. Romo-Cárdenas).

https://doi.org/10.1016/j.rinp.2019.102170
Received 29 October 2018; Received in revised form 2 March 2019; Accepted 2 March 2019
Available online 04 March 2019
2211-3797/ © 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/BY-NC-ND/4.0/).
G. Romo-Cárdenas, et al. Results in Physics 13 (2019) 102170

3 Table 1
x10
Analysis of Variance for Absorption.
1
LgLi DF Sum of Squares Mean Square F Value P Value
LgHi −13 −13
Insulin 1 7.61e 7.61e 35.74 1.93e−5
HgLi
Absorption (A.U.)

Glucose 1 2.44e−12 2.44e−12 114.50 1.06e−8

0 HgHi Interaction 1 4.61e−14 4.61e−14 2.16 0.1605
Model 3 3.24e−12 1.08e−12 50.80 2.12e−8
Error 16 3.40e−13 2.13e−14

-1
Table 2
Regression model for Absorption.
Estimate Std.Err. t value Pr(> |t|)
-2
−6 −8
Intercept 2.55e 3.262e 78.18 2.0e−16
Insulin 1.95e−7 3.262e−8 5.979 1.93e−5
1140 1120 1100 1080 1060 1040 1020 1000 980 960 Glucose 3.49e−7 3.262e−8 10.70 1.06e−8
-1 Interaction 4.8e−8 3.262e−8 1.472 0.16
Wavenumber (cm ) AdjR2 = 0.887 F = 50.8 p-val = 2.12e−8
Fig. 1. Mid-IR absorption spectra from glucose and insulin samples.

in Table 1. Where given that only glucose and insulin have a p-value
-6
x10 smaller than 0.05, they have a significant effect over the absorption,
2 which is not the case for the interaction. Meaning that the combination
of compounds has no effect in the absorption. Table 2 shows the results
of the regression model. The second column of this table shows the
1 estimate of the coefficients, which represents the involvement of each
Second derivative (A.U.)

compound in the outcome. This result is key in order to understand the

0 noticeable that glucose has a greater contribution to the absorption
-1
-2
-3
1100 1080 1060 1040 1020 1000
-1
Wavenumber (cm )
Fig. 2. Second derivative spectra of glucose and insulin samples.
and to obtain a quantification for the optical absorption from the
sample, derivative spectroscopy was used [10]. The technique requires
smoothing of the acquired absorption spectrum, which was made ap-
plying the Nyquist-Savitzky-Golay algorithm (NSG) [11]. Followed by
the calculation of the second derivative of the spectrum. After executing
the previously mentioned processes on the acquired signal, a relation
between sample constitution and optical absorption, can be obtained by
computing the substraction between the values of the lowest minimum
and highest maximum from the derived spectrum [10]. An analysis of
variance (ANOVA) and a regression model of the absorption in terms of
the concentration of the compounds, were used to study the contribu-
tion from both insulin and glucose into the absorption spectrum.
contribution of each substance. Here, comparing these coefficients, is
spectra than insulin. Their combination, so called interaction, has no
considerable involvement in the outcome. This model is capable to
LgLi explain 88% of the variations in the absorption.
LgHi
HgLi Conclusions
HgHi
The use of the factorial experimental design allowed to find that
glucose has a greater influence in the optical absorption than insulin in
combined samples. Also, that the interaction of this substances has no
significant influence in the outcome. This could be explained given that
980
despite of the presence of both compounds, there were no chemical
bonding changes that could lead to changes in the absorption. These
findings were obtained regardless of the known spectral overlapping of
this substances. Furthermore, the magnitude of the contribution of the
substances involved in the experiment, could be used to define che-
mometric and signal processing techniques, aiming for selective quan-
tification. As well as to propose in-vivo and in-vitro experimental sce-
narios of glucose regulation. Similar to this case, the use of the factorial
experimental design could provide new and interesting insights for
optical materials characterization.
Acknowledgments
We acknowledge the financial support from Universidad Autónoma
de Baja California, MyDCI graduate program and CONACYT for the
realization of this project.

Results
As shown in Fig. 1, the absorption spectrum for the samples made
from a combination of glucose and insulin was located in the range of
1100–950 cm−1. The signals were smoothed with the NSG method. Low
and high peaks were identified and subtracted in order to obtain an
absorption parameter, as is shown in Fig. 2. The absorption from all
samples and trials were used to carry out an analysis of variance and to
obtain a linear regression model. The results of the ANOVA are shown
References
[1] Werth A, Liakat S, Dong A, Woods CM, Gmachl CF. Implementation of an in-
tegrating sphere for the enhancement of noninvasive glucose detection using
quantum cascade laser spectroscopy. Appl Phys B 2018;124:75.
[2] Yoshioka K, Kino S, Matsuura Y. Noninvasive measurement of blood glucose level
using mid-infrared quantum cascade lasers. Biomedical Imaging and Sensing
Conference, International Society for Optics and Photonics. 2017. 102511U.
[3] Kasahara R, Kino S, Soyama S, Matsuura Y. Noninvasive glucose monitoring using
mid-infrared absorption spectroscopy based on a few wavenumbers. Biomed Opt
Express 2018;9:289–302.
[4] Romo-Cárdenas G, Sánchez-López JdD, Luque P, Cosío-León M, Nieto-Hipólito JI,

2
G. Romo-Cárdenas, et al.
Vázquez-Briseño M. Insulin overlapping in whole blood FTIR spectroscopy in blood
glucose measurements. Results Phys 2017;7:1221–2.
[5] Aronoff SL, Berkowitz K, Shreiner B, Want L. Glucose metabolism and regulation:
beyond insulin and glucagon. Diabetes Spectrum 2004;17:183–90.
[6] Montgomery DC. Design and analysis of experiments. John wiley & sons; 2017.
[7] Esfe MH, Rostamian H, Shabani-samghabadi A, Arani AAA. Application of three-
level general factorial design approach for thermal conductivity of MgO/water
nanofluids. Appl Therm Eng 2017;127:1194–9.
[8] Regti A, Laamari MR, Stiriba S-E, El Haddad M. Use of response factorial design for
process optimization of basic dye adsorption onto activated carbon derived from
Persea species. Microchem J 2017;130:129–36.
Results in Physics 13 (2019) 102170
[9] Pardo-Buitimea NY, Bacardí-Gascón M, Castañeda-González L, Jiménez-Cruz A.
Glycaemic index and glycaemic load of three traditional Mexican dishes. Int J Food
Sci Nutr 2012;63:114–6.
[10] Shen Y, Davies A, Linfield E, Taday P, Arnone D, Elsey T. Determination of glucose
concentration in whole blood using Fourier-transform infrared spectroscopy. J Biol
Phys 2003;29:129–33.
[11] Romo-Cárdenas G, Avilés-Rodríguez GJ, Sánchez-López JdD, Cosío-León M, Luque
PA, Gómez-Gutiérrez CM, Nieto-Hipólito JI, Vázquez-Briseño M, Navarro-Cota CX.
Nyquist-Shannon theorem application for Savitzky-Golay smoothing window size
parameter determination in bio-optical signals. Results Phys 2018;11:17–22.