Non-invasive Estimation of Blood Glucose

A feasibility study

Matthew Sidley and Jayanti Venkataraman

Department of Electrical and Microelectronic Engineering
Rochester Institute of Technology,
Rochester, NY, USA
Email: ;[email protected]

Abstract- The present work demonstrates the feasibility of elements are determined in terms of the change in glucose level
estimating blood glucose level using a microstrip antenna and incorporated into input impedance of the equivalent circuit.
strapped on a patient's arm. Some promising success has been The glucose level is estimated iteratively from subsequent
shown by the RIT ETA Lab research group that an antenna's
measurements of input impedance on the same patient. The
resonant frequency can track, in real time, changes in glucose
estimates, represented in the Clarke's Error Grid [3], compare
concentration. However, these changes in antenna response are
unique to an individual. An equivalent circuit model has been
well with the measurements made by a traditional blood
developed to represent the input impedance with changing glucose glucose meter.
levels. From a first set of measurements of input impedance in
real time on a patient, the circuit elements are determined by
II. REAL-TIME MEASUREMENTS OF INPUT
curve fitting, in terms of the change in glucose level and IMPEDANCE AND RESONANT FREQUENCY WITH
incorporated into the expression for input impedance of the CHANGES IN GLUCOSE LEVELS
equivalent circuit. The glucose level is estimated iteratively from
subsequent measurements of the input impedance on the same The technique utilizes a planar dipole to measure
patient. The estimates, represented in the Clarke's Error Grid, input impedance and resonant frequency. The dipole,
compare well with the measurements made by a traditional blood embedded in tissue modeled in HFSS with a frequency
glucose meter.
dependent dielectric permittivity profile [4], has been designed
for l.4GHz. The antenna strapped on the arm of a patient is
Keywords-blood glucose level, non-invasive monitoring,
equivalent circuit, input impedance, resonant frequency connected to a network analyzer automated to make
measurements every 15 seconds, figure 1. The resonant
I. INTRODUCTION frequency vs. time is shown in Figure 2, for a diabetic and a
non-diabetic patient after a eating a meal.
Recently non-invasive glucose concentration
measurement has become a challenging goal. There have been
several approaches proposed each with some degree of success.
These methods include Interstitial Fluid Chemical Analysis,
Breath Chemical Analysis, Infrared Spectroscopy, Optical
Coherence Tomography, Temperature-modulated localized
reference, Raman Spectroscopy, Polarization Change,
Ultrasound, Fluorescence, Thermal Spectroscopy, Ocular Figure 1. Antenna System inserted into Arm Band
Spectroscopy, and Impedance Spectroscopy. None to date have
been accepted by the medical community.
Some promising success has been shown by the RlT
ETA Lab research group that an antenna's resonant frequency
can track, in real time, changes in glucose concentration [1, 2].
However, changes in antenna response would be unique to an
individual depending on various parameters such as the
person's metabolism, body mass index, chemical profile etc. (a) diabetic patient (b) non-diabetic patient
The objective of the present work is to develop a calibration for Figure 2. Resonant Frequency and Blood Glucose Level vs. Time
blood glucose based on an initial measurement and use this to The antenna successfully tracks changes in glucose
estimate the glucose level for subsequent measurements. An level as stated in [1]. The solid line is the glucose level
equivalent R-L-C circuit model has been developed to measured by a glucose meter and the dashed line is the
represent the antenna input impedance with changing glucose resonant frequency of the antenna. It is seen that here is a
levels. An initial set of measurements are made in real time, on significant difference in the range of blood glucose and
a diabetic subject using a planar dipole antenna connected to an resonant frequency between the diabetic and non-diabetic
automated network analyzer. The corresponding glucose level subjects. This shows that a calibration, unique to each
measured by a traditional meter is also recorded. The circuit patient, is necessary.

978-1-4799-3267-2/13/$3l.00 © 2013 IEEE

 III. EQUIVALENT CIRCUIT MODEL
TABLE I. BLOOD GLUCOSE ESTIMATES USING CALIBRATION GRAPHS
Corresponding to each resonant frequency, figure 2,
which in turn corresponds to a glucose level, is input Time i1g Actual Glucose
impedance Zin. An equivalent circuit is proposed (figure 3a) (min) Zin (n) (mg/dL) Glucose, Estimate
where ZA is the antenna impedance and the circuit elements R, G (mg/dL) G+i1g
L and C are determined iteratively so that the total impedance (mg/dL)
of the circuit corresponds to the measured Zin Figures 3 a-c. 0 88 88
11 41.31-j6.40 68 175 156
25 41.29-j7.46 124 210 212
30 41.31-j7.70 134 223 222
36 41.36-j8.13 149 234 237
78 41.37-j8.25 153 228 241
(a) Equivalent Circuitfor
88 41.29-j7.53 127 205 215
measured input impedance
(b) shunt R The glucose level estimated using the calibration compares
well to that measured by a glucose meter. Analyzing the
estimates on a Clarke's Error Grid [3] diagram (figure 4)
against reference measurements, all the values lie in the
acceptable region, A.
Clarke's Error Grid Analysis
400����-- ���,--'--:
c
350

=' 300

(c) Series reactance (d) Shunt susceptance

i� 250-
.0

Figure 3 Equivalent circuit for measured Zin l 200

corresponding to glucose level .�
� 150 D
•
D
S 100
The circuit components in the equivalent circuit can be
estimated using the following curve fitted equations.

.0.8 where Al Reference Concentration [mgfdl]

R = Al i1g = 93120 (1)
.8 where A2 Figure 4. Clarke's Error Grid Analysis of Blood Glucose Estimates
C = A2 i1g.1 = 264377 (2)
45 compared to Reference Values measured by Blood Glucose Meter
L = A3 i1g.1. where A3 = 188.8 (3)

V. CONCLUSION
i1g is the change in blood glucose level with reference to the
first measurement which is obtained by a traditional glucose
The results show that increase in blood glucose level can
meter. From the equivalent circuit in figure 3a, the input
be estimated non-invasively in real time. An R-L-C equivalent
impedance in terms of i1g is as follows.
circuit model has been developed to represent the antenna input
roL U(ZA +R ) -ZAroRC] impedance with changing glucose levels. The circuit elements
Z. (4)
m jmZARL+ZA +R+m2CL(ZA+R ) are expressed in terms of the change in glucose level which is
then estimated iteratively from subsequent measurements of
(5) the input impedance on the same patient. The estimates,
represented in the Clarke's Error Grid, compare well with the
measurements made by a traditional blood glucose meter.
The antenna impedance ZA is known. For a measured Zin at
any given time, the change in glucose level, i1g, can be REFERENCES
[1] Venkataraman J. and Freer B., "Feasibility of Non-Invasive Blood
estimated iteratively from (5).
Glucose Monitoring", Invited paper," Proceedings IEEE Ant and Prop
Society International Symposium, Spokane, WA, July 5-18, 2011
IV. ESTIMATION OF GLUCOSE LEVEL USING CALIBRATION [2] Sidley M. and Venkataraman J., "Feasibility of Blood Glucose
Estimation from Real Time Monitoring", Proc. IEEE Ant and
From a second measurement made on the same patient for Propagation IntI. Sym., Orlando, July 8-12, 2013.
input impedance at discreet time intervals, the corresponding [3] Clarke WL, Cox D, Gonder-Frederick LA ,Carter W, Pohl SL:
change in glucose level is estimated from (5) and the total Evaluating clinical accuracy of systems for self-monitoring of blood
glucose level is estimated by adding this to the first value glucose. Diabetes Care 10:622-628,1987

obtained using a traditional glucose meter at the start of the [4] Gabriel, S, R W Lau, and C Gabriel. "The dielectric properties of
biological tissues: II. Measurements in the frequency range 10 Hz to 20
experiment. The results are shown in Table 1. GHz." Physics in Medicine and Biology 41, no. II (1996): 2251