IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 1, NO. 11, 2007
1
Optical Absorption Glucose Measurements Using
2.3 µm Vertical Cavity Semiconductor Lasers
Sahba Talebi Fard, Student Member, IEEE, Werner Hofmann, Pouria Talebi Fard, G. Böhm, Markus Ortsiefer,
Ezra Kwok, Member, IEEE, Markus-Christian Amann, Fellow, IEEE, Lukas Chrostowski, Member, IEEE
Abstract— Continuous glucose monitoring has been shown
to help diabetes mellitus patients stabilize their glucose levels,
leading to improved patient health. One promising technique
for monitoring blood glucose concentration is to use optical
absorption spectroscopy. This paper proposes the use of thermally
tuneable 2.3 µm vertical cavity semiconductor lasers (VCSELs)
to obtain blood absorption spectra. The Partial Least Squares
(PLS) technique is used to determine the glucose concentration
from the spectra obtained in aqueous glucose solutions.
predicting glucose concentrations [5]. Recently, appropriate
wavelength semiconductor laser sources have been developed.
In this letter, the lasers used are not only the first realized
electrically pumped VCSELs at a 2.3 µm, but also the
longest wavelength achieved so far for any InP-based interband
laser [6]. This paper demonstrates the feasibility of glucose
prediction using 2.3 µm VCSELs.
Index Terms— VCSELs, absorption spectroscopy, glucose monitoring, diabetes mellitus, implantable biomedical devices.
II. V ERTICAL C AVITY S URFACE E MITTING L ASERS
I. I NTRODUCTION
D
IABETES mellitus is a widespread disease, in which
accurate control of blood glucose will improve patient
quality of life and add savings for health care systems. Currently, acceptable blood glucose monitoring requires pricking
of fingers multiple times per day and patient compliance is an
issue. As a result, research efforts are aimed at developing
patient-friendly techniques of monitoring glucose continuously. With chemical bio-sensors having a limited life span
due to bio-fouling, optical sensors are a promising alternative,
being less susceptible to bio-fouling from blood protein adsorption. Proposed optical techniques include polarimetry, Raman spectroscopy, absorption [1] and reflectance spectroscopy
[2]. Near-infrared spectroscopy can be used for analysis of
biofluids such as whole blood, plasma, or serum. These optical
methods have been widely used for non-invasive glucose
monitoring. However, due to challenges of interference, poor
signal strength, and calibration issues; they are not yet accurate
enough for clinical use. These challenges would be reduced
if the optical sensor had direct access to interstitial fluid or
blood plasma, which would be made possible by a small and
low power VCSEL-based implantable sensor.
There are two major wavelength bands that have shown
promising correlation between their absorption spectra and
glucose concentrations, one of which is 2.0-2.5 µm [2], [3],
[4], and this band has been shown to be very effective for
Manuscript received September 18, 2007.
S. Talebi Fard, P. Talebi Fard and L. Chrostowski are with the Department
of Electrical and Computer Engineering, University of British Columbia,
V6T 1Z4 Canada (604) 822-8507, [email protected]. E. Kwok is with the
Biomedical Engineering Program at the University of British Columbia.
W. Hofmann and M.-C. Amann are with the Walter Schottky Institut,
Technische Universitat Munchen, Am Coulombwall 3, D-85748 Garching,
Germany. G. Böhm and M. Ortsiefer are with VERTILAS GmbH, Garching,
D-85748, Germany.
This work was supported by the Canadian Natural Sciences and Engineering Research Council (NSERC) and the German Research Council (DFG).
Fig. 1.
(left) Band structure of the V-shaped quantum well in the 2.3
µm InP-based long-wavelength VCSEL. (right) Reflectivity (dotted line) and
photoluminescence (PL) intensity (solid line) of a VCSEL structure at 300 K.
Vertical Cavity Surface Emitting Lasers (VCSELs) have
received tremendous attention because of their low-cost, small
size, array operation, low power consumption, excellent spectral properties, and a circular beam pattern; these properties
make them attractive for implantable biomedical applications.
For spectroscopy applications, the VCSEL wavelength, can
be scanned by changing its temperature, by changing the bias
current.
Designing long wavelength lasers for glucose sensing in
the 2-2.3 µm combination band has placed demanding requirements on the epitaxial growth, specifically to achieve the
highly strained quantum-wells required for long wavelength
emission [7]. The challenge is that the critical thickness of 5
strained quantum wells embedded in almost lattice matched
barriers is already reached at an emission wavelength of
slightly above 2 µm.
For further increase of the emission wavelength, we have
introduced V-shaped quantum wells (QW), which prevent an
abrupt change from heavily tensile strained barriers to heavily
compressive strained quantum wells. This allows much higher
IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 1, NO. 11, 2007
strain in the QWs, and the V-shape significantly increases the
emission wavelength, as compared to rectangular QWs. Figure
1 shows a V-shaped QW realized with digitally graded indium
content starting with lattice-matched GaInAs at the interface
and ending in the center with pure InAs. The redistribution of
indium content and strain within the QW and the implementation of significant tensile strain in the barriers balances the
overall strain in the active region. A 5 QW structure has been
successfully realized.
Figure 1 shows the reflectivity measurement of the DBR
and the photoluminescence (PL) measurement of the active
region at 300 K. The FWHM of 23 meV at room temperature
is rather small and goes down to 11 meV at 20 K. This active
region is used in a VCSEL structure optimized for emission
near 2.3µm. Fabricated devices can be operated in continuous
wave mode up to 45◦ C with threshold voltages of about 1 V
and output powers of 1 mW at 10◦ C. As presented in Figure
2, an emission wavelength of 2315 nm has been achieved and
it can be tuned over 5.6 nm by increasing the driving current
from 10 to 28 mA. The side-mode suppression ratio is greater
than 20 dB (measurement limited).
2
controller. This is to have a stable base temperature upon
which VCSEL can be thermally tuned by varying its biased
current. The VCSEL is current-biased using a precision current
source (Keithley 2602). A chopper modulates the beam at
a frequency of 250 Hz before passing through the solution.
The beam is detected with a PbS voltage amplified detector
(ThorLabs PDA30G). A lock-in amplifier (SR810) is used to
demodulate the signal from the detector to a DC value, and
the data are collected and analysed with a computer.
MATLAB
Power Supply-Keithley 2602
Chopper Controller
Temperature Controller
Cuvette
Temperature controlled
VCSEL mount
Fig. 3.
Fig. 2. Current dependent optical output spectra of the 2.3 µm LW-VCSEL,
and wavelength shift with respect to bias current, at 300K, with the tuning
slope of 0.3 nm/mA. The thermal tunability is 0.1 nm/◦ C.
Lock-In Amplifier
Chopper
Cuvette holder
PbS detector
Experimental set-up.
Solutions of distilled water containing concentrations of
glucose ranging from 100-1000 mg/dL have been prepared
and stored in sealed cuvettes. The VCSEL, biased with drive
currents from 8 mA to 22 mA, was used to generate absorption
spectra for each solution. The measurements include the absorption spectral characteristics of the water, glucose, cuvette,
with the addition of noise. To subtract the characteristics of
the water and the cuvette from the analysis, absorption spectra
of distilled water are subtracted from the absorption spectra of
the glucose solutions. The data in Figure 4 show the resultant
differential absorption spectra for different concentrations of
glucose.
4
III. O PTICAL G LUCOSE M EASUREMENT E XPERIMENTS
voltage (mV)
Using a laser as the light source instead of a white light
can provide advantages such as higher signal-to-noise ratio
in the absorption spectrum [8], leading to a more accurate
glucose estimation. The two wavelength regions that are commonly used for glucose monitoring are: the first-overtone band
(1560−1850nm), and the combination band (2080−2325nm),
where glucose has numerous absorption bands and water has
relatively lower absorption [2], [3], [4]. The absorption is
stronger in the combination band, and is the most desirable
wavelength region for glucose measurements using absorption
spectroscopy.
For the absorption spectroscopy experiments, a bench-top
apparatus was designed to investigate the feasibility of the
proposed technique, as shown in Figure 3. In this experiment,
low-cost plastic cuvettes with a 2 mm optical path length are
used to hold the solutions of distilled water and glucose. A
VCSEL mount (Newport 700) with a temperature controller
(Newport 3040) is used for the 2.3 µm VCSEL. The base
temperature of the VCSEL is set to 25 ◦ C using temperature
3
2
13.00 mmol/L
22.33 mmol/L
39.21 mmol/L
50.06 mmol/L
1
0
!1
!2
8
10
12
14
16
18
20
22
biased current points (mA)
Fig. 4. Absorption spectra for concentrations of glucose, in distilled water,
subtracted by the absorption spectra of distilled water.
IV. A NALYSIS AND R ESULTS
Glucose has various absorption bands including 1.67, 2.13,
2.27, 2.33 µm, which arise from combination and overtone
molecular vibration with C-H and O-H bonds of glucose
molecule. These features result in a very broad absorption
spectrum spanning the entire 2-2.3 µm combination band.
IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 1, NO. 11, 2007
3
There are overlapping spectral signatures from other components in blood or interstitial fluid. The broad and highly overlapped characteristics of the near infrared absorption bands
dictates the need for a spectrum (or several bands of spectra)
rather than measurements at one or two discrete wavelengths
[9]. Furthermore, due to the correlation of glucose with other
physiological conditions and components of body fluid as
well as lack of a specific one wavelength point reflecting the
glucose concentration, glucose spectra have broad shape with
non-monotonic increase with glucose concentration as would
be expected from normal lineshape absorption spectra.
This non-linear relation between absorption spectra and
glucose concentration suggests the use of multivariate analysis
[8]. Multivariate models can be designed to predict glucose
concentration, eliminating the effects of potential correlation
and confounding factors, including physiological variations
[10]. Multivariate analysis, such as Partial Least Squares
(PLS), is required to find a correlation between the spectra
and glucose concentrations [3], [11]. PLS is used to estimate
the regression coefficients from a sample of data, and build
a model that can be used to make predictions from new
observations, but it cannot deduce any spectra of individual
component in the solution [12].
In this experiment, nine (9) absorption spectra were collected for each concentration of glucose. From these 9 absorption spectra, 5 random spectra were used to compute regression vectors using 4 principle components (PC). This PLS
glucose prediction model was tested using the other 4 spectra
to investigate the accuracy and precision of the model in
predicting glucose concentration in a given solution. Figure 5
shows the result of this analysis. The estimated concentrations
from the PLS modeling over the clinical range of interest are
on average within 30% around the actual values. The results
show that with only one VCSEL, glucose concentrations can
be estimated from the small spectral window of 5-6 nm at a
EDITcenter wavelength of 2.3µm. This result demonstrates
the feasibility of glucose prediction with VCSELs operating
in this small wavelength region.
200
400
600
800
1000
60
1200
1200
1000
50
800
40
600
30
400
20
200
10
0
0
10
20
30
40
50
60
Estimated Concentration (mg/dL)
Estimated Concentration (mmol/L)
Known Concentration (mg/dL)
0
70
0
70
Known Concentration (mmol/L)
Fig. 5. Glucose prediction using the PLS model, derived from the absorption
spectra data.
In this proof-of-concept experiment, the glucose concentrations were higher than the typical clinical range (50-300
mg/dL). There are several methods of improving the results
to be able to predict lower concentrations of glucose. The first
is using multiple VCSELs at different wavelengths, in order
to capture a larger portion of the glucose absorption feature.
Next, VCSELs can be chosen to better match the glucose
absorption bands, leading to a higher degree of correlation
between the spectra and the glucose concentrations. Choosing
cuvettes with a lower optical absorption, and optimizing the
differential lock-in detection mechanism, can further increase
the signal-to-noise ratio. Once this VCSEL-based glucose
measurement design has been refined for the clinical range,
it will be evaluated for detecting glucose in blood serum or
similar body fluid. The ultimate goal is to package the design
as an implantable device.
V. C ONCLUSION
A method of achieving a high-quality epitaxy with highly
strained quantum-wells have been developed. It allows for the
fabrication of long wavelength VCSELs for glucose sensing
in the 2-2.3 µm combination band. A single 2.3 µm VCSEL
was used to demonstrate the feasibility of glucose sensing
in an aqueous solution, using absorption spectroscopy and a
PLS algorithm. This result leads to further improvement in
accurately predicting glucose in the body.
R EFERENCES
[1] J. R. McNichols and L. G. Cote, “Optical glucose sensing in biological
fluids: An overview,” Journal of Biomedical Optics, vol. 5, no. 1, pp.
5–16, 2000.
[2] S. F.Malin, T. L. Ruchiti, T. B. Blank, S. U. Thennadil, and S. L. Monfre,
“Noninvasive prediction of glucose by near-infrared diffuse reflectance
spectroscopy,” Clinical Chemistry, vol. 45, pp. 1651–8, 1999.
[3] K. J. Jeon, I. D. Hwang, S. Hahn, and G. Yoon, “Comparison between
transmittance and reflectance measurements in glucose determination
using near infrared spectroscopy,” Journal of Biomedical Optics, no. 11,
p. 014022, 2006.
[4] A. K. Amerov, J. Chen, G. W. Small, and M. A. Arnold, “Scattering
and absorption effects in the determination of glucose in whole blood by
near-infrared spectroscopy,” Anal. Chem, no. 77, pp. 4587–4594, 2005.
[5] Y. Yu, K. D. Crothall, L. G. Jahn, and M. A. DeStefano, “Laser diode
applications in a continuous blood glucose sensor,” Proceeding of SPIE,
vol. 4996, pp. 268–274, 2003.
[6] M. Ortsiefer, G. Böhm, M. Grau, K. Windhorn, E. Rönneberg,
J. Rosskopf, R. Shau, O. Dier, and M.-C. Amann, “Electrically pumped
room temperature CW VCSELs with 2.3 µm emission wavelength,”
Electron. Lett., vol. 42, pp. 640–641, 2006.
[7] G. Böhm, M. Grau, O. Dier, K. Windhorn, E. Rönneberg, J. Rosskopf,
R. Shau, R. Meyer, M. Ortsiefer, and M.-C. Amann, “Growth of InAscontaining quantum wells for InP-based VCSELs emitting at 2.3 µm,”
J. Crystal Growth, pp. 301–302 and 941–944, 2007.
[8] V. A. Saptari and K. Youcef-Toumi, “Sensitivity analysis of near infrared
glucose absorption signals: towards noninvasive blood glucose sensing,”
Proceeding of SPIE, vol. 4163, 2000.
[9] J. T. Olesberg, M. A. Arnold, C. Mermelstein, J. Schmitz, and J. Wagner,
“Tunable laser diode system for noninvasive blood glucose measurements,” Applied Spectroscopy, vol. 59, no. 12, pp. 1480–1484, 2005.
[10] R. Liu, W. Chen, X. Gu, R. K. Wang, and K. Xu, “Chance correlation
in non-invasive glucose measurement using near-infrared spectroscopy,”
Journal of Physics D: Applied Physics, no. 38, pp. 2675–2681, 2005.
[11] Y. Shen, A. Davies, E. Linfield, P. Taday, D. Arnone, and T. Elsey,
“Determination of glucose concentration in whole blood using fouriertransform infrared spectroscopy,” Journal of Biological Physics, vol. 29,
pp. 129–133, 2003.
[12] S. Hahn and G. Yoon, “Identification of pure component spectra by
independent component analysis in glucose prediction based on midinfrared spectroscopy,” Applied Optics, vol. 45, pp. 8374–8380, 2006.