Online Tissue Discrimination for Transcutaneous Needle Guidance Applications Using Broadband Impedance Spectroscopy

Abstract

This paper reports on a novel system architecture for measuring impedance spectra of a biological tissue close to the tip of a hollow needle. The measurement is performed online using fast broadband chirp signals. The time domain measurement raw data are transformed into the transfer function of the tissue in frequency domain. Correlation technique is used to analyze the characteristic shape of the derived tissue transfer function with respect to known "library functions" for different types of tissue derived in earlier experiments. Based on the resulting correlation coefficients the exact type of tissue is determined. A bipolar coaxial needle is constructed, simulated by finite element method and tested during various in vitro and in vivo experiments. The results show a good spatial resolution of approximately 1.0 mm for a needle with a diameter of 2.0 mm. The correlation coefficients for the three tested tissue types muscle, fat, and blood allow for a clear tissue classification. Best results have been obtained using the characteristic phase diagrams for each tissue. Correlated to the corresponding library transfer function the coefficients are in the range of +0.96 to +0.99 for the matching tissue. In return, the resulting coefficients for correlation with nonmatching tissues are in the range of -0.93 to +0.81.

Full text

494 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 59, NO. 2, FEBRUARY 2012
Online Tissue Discrimination for Transcutaneous
Needle Guidance Applications Using Broadband
Impedance Spectroscopy
Dennis Trebbels*, Felix Fellhauer, Michael Jugl, Gerd Haimerl, Mart Min, and Roland Zengerle
Abstract—This paper reports on a novel system architecture for
measuring impedance spectra of a biological tissue close to the tip
of a hollow needle. The measurement is performed online using fast
broadband chirp signals. The time domain measurement raw data
are transformed into the transfer function of the tissue in frequency
domain. Correlation technique is used to analyze the characteris-
tic shape of the derived tissue transfer function with respect to
known “library functions” for different types of tissue derived in
earlier experiments. Based on the resulting correlation coefficients
the exact type of tissue is determined. A bipolar coaxial needle is
constructed, simulated by finite element method and tested during
various in vitro and in vivo experiments. The results show a good
spatial resolution of approximately 1.0mm for a needle with a di-
ameter of 2.0 mm. The correlation coefficients for the three tested
tissue types muscle, fat, and blood allow for a clear tissue classi-
fication. Best results have been obtained using the characteristic
phase diagrams for each tissue. Correlated to the corresponding li-
brary transfer function the coefficients are in the range of +0.96 to
+0.99 for the matching tissue. In return, the resulting coefficients
for correlation with nonmatching tissues are in the range of −0.93
to +0.81.
Index Terms—Broadband impedance spectroscopy, cannula
guidance, chirp-signals, needle guidance, tissue classification.
I. INTRODUCTION
NEEDLE guidance during transcutaneous surgical inter-
ventions is a common challenge in many medical applica-
tions such as fine needle biopsies [1], regional anaesthesia [2],
drug delivery [2], catheter insertion [3], vessel puncture [4] and
Manuscript received March 8, 2011; revised June 9, 2011 and August 25,
2011; accepted October 3, 2011. Date of publication November 8, 2011; date
of current version January 20, 2012. Asterisk indicates corresponding author.
*D. Trebbels, HSG-IMIT, Institut f¨
ur Mikro- und Informationstechnik der
Hahn-Schickard-Gesellschaft e.V., 78052 Villingen-Schwenningen, Germany
(e-mail: dennis.trebbels@hsg-imit.de).
F. Fellhauer, HSG-IMIT, Institut f¨
ur Mikro- und Informationstechnik der
Hahn-Schickard-Gesellschaft e.V., 78052 Villingen-Schwenningen, Germany
(e-mail: felix.fellhauer@hsg-imit.de).
M. Jugl and G. Haimerl are with the HFU, Hochschule Furtwangen University,
78054 Villingen-Schwenningen, Germany (e-mail: michael.jugl@hsg-imit.de;
hai@hs-furtwangen.de).
M. Min is with the Tallinn University of Technology, Faculty of Informa-
tion Technology, Department of Electronics, 19086 Tallinn, Estonia (e-mail:
min@elin.ttu.ee).
R. Zengerle is with the Department of Microsystems Engineering, Labora-
tory for MEMS-Applications, University of Freiburg, 79110 Freiburg, Germany
(e-mail: roland.zengerle@gmx.de).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TBME.2011.2174990
brachytherapy [5]. In all of the listed applications it is impor-
tant to place the needle tip inside a well defined tissue type of
the human body whereas often the exact 3-D orientation of the
needle is of minor importance.
Today there are several approaches to accomplish the exact
needle placement based on 2-D or 3-D imaging systems which
allow for controlling the needle orientation and position and sup-
port the surgeon during the needle insertion procedure. Three
well known systems are magnetic resonance imaging (MRI),
computed tomography (CT) and X-Ray fluoroscopy [6]–[8].
All systems work very well and give good spatial resolution
but are too expensive for widespread use. In addition CT and
X-Ray fluoroscopy expose the patient to ionizing radiation. Fur-
thermore, fluoroscopy provides only a limited contrast for soft-
tissues, which is a crucial point for needle guidance. Alternative
imaging systems are based on ultrasound (US) [9], [10]. There
are several ultrasound systems available in the market, but ma-
jor drawbacks are the limited penetration depth and the limited
field of view. In addition, it is difficult to analyze the exact type
of soft tissue close to the needle tip by using US. The use of
force-feedback systems [11] in combination with imaging based
systems or robots [12] can improve the needle steering process
but the overall system again becomes quite complicated and
expensive. In some actual research projects microsensors are
integrated in the needle tip [13] but deal with the inherent lack
of space given by the thin needle geometry and the low-cost
requirements for disposable medical equipment.
A promising alternative approach for exact online tissue clas-
sification at the needle tip during the insertion process into the
human body is broadband impedance measurement, often re-
ferred to as impedance spectroscopy (IS). This method is based
on the continuous measurement of the characteristic impedance
spectrum of a small volume of tissue around the needle tip.
The measured spectrum is analyzed and the exact type of tissue
can be determined. Kalvoy et al. already presented prelimi-
nary results of their configuration based on a monopolar needle
electrode where the impedance is measured against a reference
electrode [14], [15]. The results look promising but a drawback
of the system is the need for the additional reference electrode,
and due to the current density distribution the achievable spatial
resolution is limited to a spherical volume having a diameter
of approximately three to four times the needle diameter. We
already proposed a coaxial needle design in combination with
time domain reflectometry (TDR) measurement technique [16].
Drawback of this technique is the poor signal-to-noise ratio,
which can be achieved for a broadband measurement.
0018-9294/$26.00 © 2011 IEEE

TREBBELS et al.: ONLINE TISSUE DISCRIMINATION FOR TRANSCUTANEOUS NEEDLE GUIDANCE APPLICATIONS 495
Fig. 1. Detailed view of the constructed coaxial hollow needle. The needle
consists of two concentric stainless steel tubes which are electrically isolated
by an inner layer of Teflon (PTFE). The given dimensions correspond to the
prototype needle, which is used throughout the simulations and experiments.
In this paper, we present a new measurement system for coax-
ial hollow needles where the impedance measurement method
is based on short broadband chirp signals. The paper discusses
the system concept, finite element method simulation results
for the expected spatial resolution, chirp signal properties and
processing, the developed laboratory measurement setup and
experimental details for the conducted in vitro and in vivo ex-
periments. The derived raw data are processed and analyzed by
correlation techniques. The correlation results are presented and
discussed particularly with regard to tissue classification. In ad-
dition, some further experiments are done in order to support the
simulation results, investigating the impact of the speed of the
needle during insertion process and evaluating the robustness of
the classification method with regard to mechanical tolerances
and imperfections of the needle.
II. SYSTEM CONCEPT AND MODEL
Goal of the developed system is to continuously analyze and
classify the exact type of tissue close to the needle tip during the
insertion process of the needle into the human body. In order
to classify the tissue close to the needle tip we continuously
measure the complex electrical impedance of the tissue close to
the needle tip at various frequencies within the range of 5 kHz
to 1 MHz. Within this frequency range the complex transfer
function of the tissue is expected to be characteristic for each
individual type of tissue [17]. The chosen range is a compromise
between overall bandwidth, signal duration, and the resulting
total amount of raw data, which must be processed in real time.
A more detailed description of the signal properties is given in
Section IV-C.
A coaxial hollow needle is constructed as illustrated in Fig. 1.
The inner stainless steel tube serves as electrical conductor for
the measurement signal whereas the outer stainless steel tube
is at ground potential. The two tubes are electrically isolated
against each other by a thin layer of Teflon (PTFE). Current can
only flow from the inner conductor to the outer conductor at
the needle tip while inserted into a conductive media (see also
Fig. 3). When the needle is inserted into biological tissue, the
tissue forms a frequency-dependent electrical load on the needle
Fig. 2. (a) Equivalent electrical schematic of biological tissue located on the
needle tip in series to a model for the electrode polarization. (b) The used
cell model from which the equivalent schematic is derived. (c) The electrode
polarization effect by a double layer structure.
Fig. 3. (a) Current density plot of the needle tip in homogeneous resistive
media. The highest current density and therefore the highest sensitivity are
close to the needle tip. (b) The simulated needle tip and the volume virtually
divided in front of the tip in thin slices with 0.1 mm thickness for further analysis
of the losses and sensitivity. The results are shown in Fig. 4.
tip as illustrated in Fig. 2(a). The presented equivalent circuit
model for the tissue is composed of Rp,R
i,and CM.Fig.3(b)
illustrates a corresponding cell model, where Rirepresents the
resistive behavior of the intracellular volume, Rprepresents the
resistive behavior of the extracellular volume, and CMrepre-
sents the capacitive behavior of the cell membrane [18]. In series
to the tissue model we place a second model, which describes the
frequency depending behavior of electrode polarization effects
(Cpol,R
pol,Z
W) [19]. Electrode polarization occurs whenever
a galvanically coupled metal electrode is in direct contact with
an aqueous electrolyte. The effect can be explained by a rigid
double layer, which is caused by metal ions on the electrode
surface and corresponding counter ions in the electrolyte. Often
this layer is referred to as Helmholtz layer. In addition, there
is a weaker diffusion layer caused by the electrostatic coulomb
forces of the charged metal surface, often referred to as Gouy-
Chapman layer [see Fig. 2(c)]. The equivalent circuit model for
the two layers consists of a common layer capacitance Cpol and
the two elements Rpol and ZW[19]. Rpol represents the resistive
behavior of the charge transfer within the Helmholtz layer and
is almost frequency independent. ZWrepresents the impedance

496 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 59, NO. 2, FEBRUARY 2012
of the Gouy-Chapman layer and is frequency-dependent. Often
this frequency dependency is modeled by a constant phase el-
ement. The resistive properties of the electrolyte bulk material
are already represented by Rpof the tissue model, and therefore
here not included in the double layer model.
For tissue classification,it would be ideal to measure the
broadband impedance of the tissue only and eliminating the
parasitic effects caused by the electrode polarization. However,
this is practically impossible due to the 2-electrode interface
formed by the needle tip. All measurement results will always
be a combination of the tissue transfer function and the elec-
trode polarization transfer function. Calibration and precision
correction of the polarization effects is also practically not ap-
plicable because the polarization itself depends on characteristic
tissue properties such as the conductivity of the bulk material
as well as on measurement signal parameters such as the ap-
plied voltage level and the resulting current density. In addition,
a calibration (e.g., in saline) for each hollow needle means a
significant amount of work. In the ideal case, it would be pos-
sible to measure directly without the need for any calibration
procedure.
Throughout the study presented here in this paper, we decided
to go a straight way and simply measure the resulting overall
electrode impedance as a function of frequency inside differ-
ent types of tissue including all polarization effects. Multiple
complex impedance curves are recorded for each type of tissue.
An averaged curve for each individual tissue type is calculated
out of the recorded data and serves as “library function.” Tissue
classification is done by comparing these “tissue library func-
tions” online with continuously measured impedance curves.
The comparison is made by correlation and the tissue classifica-
tion is based on the resulting correlation coefficients. We accept
that the measured impedance curves do not exactly represent
the specific tissue impedance only. Our focus is on tissue classi-
fication between several types of tissue and not on precise tissue
analysis of one specific single tissue. The results discussed later
in the paper show that a pure classification can be achieved by
this method, and there is no need for an additional calibration
procedure using a reference electrolyte.
III. SIMULATION
A low frequency finite element simulation is done in order to
evaluate the current density distribution and the resulting spatial
sensitivity for a needle with the geometry shown in Fig. 1. The
simulation is done at a frequency of 100 kHz using the software
package “Ansys Multiphysics.” The maximum mesh size is set
to 30 μm, the boundaries of the simulated volume are at ground
potential. The excitation signal amplitude is set to 1 V and the
conductivity of the bulk material is set to 1S/m. However, the
absolute values of the resistive bulk material and the excita-
tion signal amplitude are not relevant, since we calculate the
“relative losses.” This means we calculate the losses within a
small tissue volume close to the needle tip and compare them
to the total losses in the simulated system caused by the mea-
surement signal. Any absolute material properties such as the
specific conductivity cancel out. The resulting “relative losses”
Fig. 4. FEM-Simulation results: Relative losses as a function of the volume
in front of the needle tip. The volume is indicated by the distance to the needle
tip according to the slices shown in Fig. 3.
represent the sensitivity as a function of the needle tip geome-
try only. We do not simulate any frequency-dependent behavior
of the tissue since this has no effect on the spatial sensitivity.
The electrical losses are proportional to the square of the cur-
rent density within a small volume of tissue. Fig. 3(a) shows a
simulated current density pattern for the needle tip inserted into
homogeneous resistive media.
As expected, the highest current density and, therefore, the
highest sensitivity are found close to the needle tip, which serves
as two-electrode-pair. For further detailed analysis of the sensi-
tivity we virtually cut the volume in front of the needle tip into
thin slices with a thickness of 0.1 mm each, as shown in Fig. 3(b).
Fig. 4 presents the simulation result. The graph shows the accu-
mulated relative losses inside a volume in front of the needle tip,
as illustrated in Fig. 3(b). The graph shows that approximately
90% of the total losses are generated within a distance of only
1 mm to the needle tip. In addition, we investigated the effect
of the needle insertion depth. The current density on the outer
needle wall surface is very low but, in total, the needle has a
large surface compared to the needle tip. Fig. 4 contains two
graphs, one for 10 mm insertion depth and the other for 40 mm
insertion depth. Both graphs are almost identical which means
the sensitivity is almost independent of the insertion depth of
the needle. The result looks reasonable since almost all losses
(∼90 %) are generated directly in front of the needle tip.
The simulation result leads to the assumption that the hollow
coaxial needle design yields a very good spatial resolution of
the measurement and is ideally suited for precision needle tip
positioning. In addition to the simulation, we conducted an in
vitro experiment. The obtained results support the simulation
result very well and are described in detail within Sections VI
and VII of this paper.
IV. MEASUREMENT SIGNALS AND PROCESSING
A. Measurement Signals
As already mentioned in Section II, the measurement system
must be capable of measuring the complex impedance spec-
trum of the tissue under investigation at the needle tip. Since
needle insertion is a dynamic process, the measurement has
to be done within a short period of time. In an ideal case the

TREBBELS et al.: ONLINE TISSUE DISCRIMINATION FOR TRANSCUTANEOUS NEEDLE GUIDANCE APPLICATIONS 497
Fig. 5. Normalized linear chirp signal plot in time domain. The presented
sample signal has a duration of 30 μs and its frequency ranges from 5 kHz to
1MHz.
measurement time window is so short that the complex tissue
load on the needle tip remains stable for one complete mea-
surement cycle. However, this requirement cannot be achieved
by using conventional measurement techniques such as apply-
ing several frequencies within one slow sweep. Therefore, we
use short broadband signals. One type of broadband signals is
the so called chirp signal which has already been introduced as
a promising signal for impedance measurement [20]. The sine
wave based chirp signal has a general expression
C(t)=Asin θ(t)=Asin ω(t)dt +θ0(1)
with amplitude A, running phase θ(t), instantaneous angle fre-
quency ω(t) and initial phase angle θ0. A simple linear chirp
has an instantaneous frequency ω(t) =dθ(t)/dt, which changes
linearly during the excitation interval Texc with constant accel-
eration dω(t)/dt =d2θ(t)/dt2=kch. Expressing ω=2πf and
denoting f0as an initial, and ffin as a final frequency, also taking
θ0=0 and marking Tch as duration of the chirp pulse, we obtain
the following expression for the linear chirp excitation:
C1(t)=Asin 2πf0·t+(ffin −f0)·t2
2Tch .(2)
The excitation bandwidth Bexc =ffin –f0remains the same when
the excitation time Texc =Tch changes. Only the chirping rate
kch =(ffin –f0)/Tch and signal energy E=(A2/2)Tch vary to-
gether with Tch.
A linear chirp signal calculated by (2) gives maximum flex-
ibility in scaling the signal to the needs of the application. In
this case amplitude, signal duration and covered bandwidth can
be controlled independently. Figs. 5 and 6 show a linear chirp
signal in time domain and frequency domain. One advantageous
property of such a chirp signal is the flat spectrum in the fre-
quency domain. The amplitude of all contained frequency bins
remains stable over the entire frequency range (Fig. 6).
This property allows for measuring the impedance over the
full bandwidth with stable signal-to-noise ratio and, therefore,
leads to precision transfer functions of the measured complex
needle tip impedance.
B. Signal Processing Concept
Goal of the measurement is to derive the complex transfer
function of the load impedance caused by the tissue under inves-
tigation in the frequency domain. Therefore, the unknown tissue
Fig. 6. Normalized amplitude plot of the power spectral density of the chirp
signal presented in Fig. 5. The signal energy equally distributed within the
bandwidth of the signal and therefore allowing broadband measurements with
a stable signal to noise ratio over the whole frequency range.
Fig. 7. Block schematic of the signal processing chain as implemented in
the developed system. The system under investigation is the load impedance
caused by the unknown tissue located at the needle tip. Stimulus and response
signals are voltage and current signals, respectively. Both signals are sampled
and converted into frequency domain for calculation of the transfer function.
is stimulated by a broadband chirp voltage signal and the result-
ing response current is recorded. Both signals are synchronously
sampled by a two-channel AD-converter and transformed into
frequency domain (Fig. 7). The quotient of both signals is the
complex transfer function.
C. Measurement Signal Properties
As already mentioned before, the chirp signal properties can
independently be adjusted to the specific requirements of the
application. In this case the bandwidth of interest is set to
5 kHz – 1.0 MHz, because here we expect a characteristic
impedance spectrum for each biological tissue [17]. Further-
more, the covered frequency range is a compromise between
the excitation signal bandwidth, excitation signal duration, and
the resulting amount of sampled raw data, which must be pro-
cessed online with a reasonable amount of computing effort.
Key consideration is the required sampling rate of the two syn-
chronous AD-converters shown in Fig. 7. According to the Shan-
non theorem, the minimum sampling rate must be at least twice
the frequency of the highest frequency component in the signal.
Therefore, the maximum frequency component fmax in the chirp
signal dictates the minimum sampling frequency fsample-min of
both AD-converters which is fsample-min =2∗fmax. The result-
ing minimum total number of samples Nmin required to capture
one full chirp excitation signal can be calculated and expresses
as Nmin =Tchirp ∗fsample-min =2∗fmax /fmin. This equation
shows that the number of samples Nmin is an exponential func-
tion and becomes quite large for signals with a broad bandwidth
covering several decades. Therefore we choose a compromise
and use a chirp signal with a duration of 200 μs. This results in
a frequency resolution δf=1/Tchirp =1/200 μs=5 kHz and
a minimum number of samples Nmin =400. In addition we
decide to use an oversampling rate of a factor 10, resulting in

498 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 59, NO. 2, FEBRUARY 2012
Fig. 8. Block schematic of the laboratory setup. The unknown impedance Z
and the current measurement resistor are connected in series and the AWG ex-
cites the measurement signal. In addition, the AWG also generates an adjustable
trigger signal for the oscilloscope that allows a customized sampling rate.
4000 samples. The next matching power of 2 is 4096 samples.
The sampling frequency and thus the trigger frequency for the
AD-converter is, therefore, adjusted to 20.48MHz for capturing
a chirp with fmax =1 MHz. Our developed software running
on a laptop computer can perform a complex 4096 point FFT
within 2.4 ms and allows for fast online signal processing.
Another critical parameter is the excitation signal amplitude.
Extremely low amplitudes in the mV or even sub-mV range
cause difficulties associated with noise and the dynamic range
of AD-converters. In contrast very high amplitudes in the range
of several volts may significantly cause measurement errors due
to the resulting high current densities and nonlinear behavior
or even damage of the tissue under investigation. Based on our
experience, we choose the chirp signal amplitude to be 100 mV
as a compromise and use this value within all experiments.
V. LABORATORY MEASUREMENT SETUP
For experiments, a laboratory measurement setup is devel-
oped using state of the art equipment. Fig. 8 shows a block
schematic of the implemented setup. The chirp signal is gener-
ated by a programmable arbitrary waveform generator (AWG,
Tektronix AFG3252) which drives the coaxial cable connected
to the needle via a current measurement resistor. The stimulating
voltage signal and the resulting current signal are synchronously
sampled by a digital storage oscilloscope (DSO, LeCroy
WaveRunner 104 Xi). The AWG generates an additional trig-
ger signal with stable phase relative to the chirp measurement
signal and allows for a stable triggering and sampling. Both
oscilloscope input channels have a limited input impedance of
1 MOhm and a relatively high input capacitance of approxi-
mately 20 pF. Especially at higher frequencies the high input ca-
pacitance causes a low input impedance and leads to frequency-
dependent measurement errors. This unwanted effect is avoided
by employing two identical buffer amplifiers connected to the
oscilloscope inputs, as shown in Fig. 8. The amplifiers have an
input capacitance of approximately 1 pF.
The software calculates the chirp signal and controls the sig-
nal generation and sampling process via LAN and USB. In
addition, the software processes the measured raw data and di-
rectly calculates the transfer function. In principle the transfer
function can directly be calculated out of the measured voltage
and current signals, as illustrated in Fig. 7. Drawbacks of this
Fig. 9. Equivalent electrical schematic of the system including relevant par-
asitic components that cause measurement errors. Riis the output resistance
of the generator, Rcis the dc-resistance of the copper wires that connect the
measurement setup with the needle and Ccis the cable capacitance of the used
coaxial cable. (a) Cccan be estimated by an “open” calibration. (b) Rccan be
estimated by a “short” calibration if Ccis known.
direct method are measurement errors caused by inherent para-
sitic effects such as unwanted capacitive and resistive elements
in the setup. Fig. 9 shows an equivalent circuit including major
effects such as the cable capacitance Ccand the connector/cable
resistance Rc. In order to minimize such measurement errors, we
perform a one-time system calibration by measuring the transfer
function of the “open” system (no load at the needle tip) and the
“shortened” system (short circuit at the needle tip). The mea-
sured transfer functions of the open circuit and the shortened
circuit are used to directly correct measurement errors of the
tissue transfer functions in frequency domain. First, the “open”
calibration has to be performed. In an ideal setup, no current
is expected to flow under the “open” condition. In reality, a
frequency-dependent current can be measured. This alternating
current mainly flows through the parallel cable capacitance Cc.
The equivalent schematic for an “open” condition is presented
in Fig. 9(a).
The measured transfer function TF{open}is directly the
transfer function of the capacitive element TF{Cc}.Fig.9(b)
shows the equivalent circuit for a “short” condition on the nee-
dle tip. In parallel to the capacitance Ccthere is two times the
cable resistance Rc. The transfer function TF{Rc}can be ob-
tained by calculating TF{Rc}=TF{short}–TF{Cc}.Nowall
relevant parasitic components are known and the corresponding
transfer functions can directly be taken into account during the
calculation of any measured tissue transfer function. Fig. 10
shows two plots for a laboratory measurement result obtained
by placing the needle tip into a resistive saline bath with a mea-
sured conductivity of 16.2 mS/cm (almost isotonic). In the test
setup we used a coaxial cable of type RG174A with a length of
1.5 m to connect the needle with our measurement setup. In case
of no calibration and no error correction, a signal degradation at
high frequencies is observed. This is mainly caused by the par-
asitic coaxial cable capacitance. Applying the above described
calibration procedure virtually eliminates the parasitic effects
and shows a stable modulus and phase value as expected for a
pure resistive load within this frequency range. Any polariza-
tion effects at low frequencies could not be observed within this
experiment. According to [19] the polarization effect becomes
more visible at frequencies below 1 kHz, which is outside of our
measurement range.

TREBBELS et al.: ONLINE TISSUE DISCRIMINATION FOR TRANSCUTANEOUS NEEDLE GUIDANCE APPLICATIONS 499
Fig. 10. Influence of parasitic components and result of calibration and error
correction illustrated by the result of an experiment obtained in resistive saline
bath.
VI. MATERIALS AND METHODS
A. Overview
Within this research study, we conducted several in vivo and
in vitro experiments. Goal of the in vivo experiments was to
derive measurement data especially in order to investigate the
possibility of tissue classification. Details of the in vivo experi-
ments are presented in Section VI-B in this paper. In addition to
the in vivo experiments we conducted three in vitro experiments
in order to investigate side effects and robustness of the measure-
ment principle as well as verifying the FEM-simulation results
presented in Section III. Details about the in vitro experiments
are given in Section VI-C.
All measurements have been done with the developed labo-
ratory setup described in Section V. The measurement resistor
R(see Fig. 8) is set to 100 Ohm. The excited chirp signal had a
duration of 200 μs, an amplitude of 100 mV, and covered a band-
width from 5 kHz to 1 MHz. The signals have been captured by
a DSO with a sampling rate of 20.48 MHz and an amplitude res-
olution of 8 bits resulting in a total number of 4096 samples per
sampled chirp signal. The sampled waveforms of the stimulus
chirp signal and resulting current signal have been stored on file
for later analysis.
B. In vivo Experiments
All in vivo experiments have been conducted on anesthetized
pigs. For our study, we measured on three pigs. On each pig the
three-tissue types muscle, fat, and blood have been measured
on two locations with a prototype needle having the dimensions
presented in Fig. 1. A picture of the used prototype needle is
shown in Fig. 11(c). First measurement location was the neck of
the pig, and second measurement location was the hip of the pig.
A veterinarian carefully cut the skin of the pig and dissected the
named three-tissue samples. The samples were prepared in such
a way that the measurement needle could be inserted into either
plain muscle, fat, or arterial blood tissue without any visible
inhomogeneity. Placement of the needles was done manually.
The in vivo experiments have been done in accordance with
animal ethic standards approved by the government of Baden-
W¨
urttemberg, Germany. Institutional and national guides for the
care and use of laboratory animals were followed.
C. In vitro Experiments
Goal of the first in vitro experiment was to verify the simula-
tion result for the expected spatial resolution of the tissue clas-
sification on the needle tip. Therefore, we inserted the needle
Fig. 11. Picture of three different hollow coaxial prototype needles. Needle C
is used throughout the experiments, the dimensions are given in Fig. 1. Needles
A and B are used during additional in vitro experiments. Picture 11D illustrates
the setup used for in vitro experiments using well prepared tissue with good
visible and sharp tissue boundaries between muscle and fat.
into well prepared tissue [see Fig. 11(d)] using a programmable
linear actuator. We started with the needle tip embedded inside
homogeneous muscle tissue. In steps of 0.1 mm we moved the
needle toward the fat tissue. On each step we measured the
impedance spectrum of the tissue load. We stopped the actuator
when the needle tip was fully embedded inside fat tissue.
The second in vitro experiment was conducted with the same
setup, but the needle was moved within homogeneous muscle
tissue only. We continuously moved the needle with various
speeds from 0 to 14 mm∗s−1and continuously measured dur-
ing the movement. Goal was to evaluate if a moving needle
shows different results compared to a non-moving needle due
to disturbance effects of the double layer on the electrodes.
The third in vitro experiment was done with three different
types of prototype needles with various dimensions, as shown
in Fig. 11(a) and (b). All three needles were placed into homo-
geneous muscle and fat tissue and the impedance spectra have
been recorded. The results are compared to each other and give
a first overview about the impact of mechanical parameters on
the tissue classification.
VII. MEASUREMENT RESULTS
A. In vivo Experiments
During the in vivo experiments the three different tissue types
muscle, fat, and arterial blood have been measured. The sampled
raw data of the stimulus voltage signal and the response current
signal have been processed by a custom analysis software ac-
cording to the scheme presented in Fig. 7. The output of the
signal processing software is a Bode diagram for each measure-
ment. Since the measured impedance is complex there are two
plots for each measurement—the modulus plot and the phase
plot as a function of frequency. Below are the obtained results
for the three tissues. We conducted the measurement on three
pigs with two samples per pig resulting in a total of six measure-
ments per tissue. However, for muscle tissue we only show five
measurements since we had to exclude one extreme outlier. The
outlier was caused by accidentally scratching a blood vessel and
almost reflected the blood curve only.
B. In vitro Experiments
The first in vitro experiment was done in order to verify
the simulation result presented in Fig. 4. The experiment was

500 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 59, NO. 2, FEBRUARY 2012
Fig. 12. Bode diagrams for blood measured using a 200 μs chirp signal. The
left diagram shows the modulus and the right diagram shows the phase.
Fig. 13. Bode diagrams for fat measured using a 200 μs chirp signal. The left
diagram shows the modulus and the right diagram shows the phase.
Fig. 14. Bode diagrams for muscle measured using a 200 μs chirp signal. The
left diagram shows the modulus and the right diagram shows the phase.
Fig. 15. In vitro experiment for evaluating the spatial measurement resolution.
Three remarkable needle positions are shown in A, B, and C. The corresponding
positions are also marked with A, B, and C in the phase diagram on the right
side. The plotted phase curve shows the phase angle at a frequency of 100kHz.
conducted as described in Section VI-C. Measurements were
obtained for the needle transition from muscle to fat tissue.
Fig. 15(a)–(c) illustrates the needle position as also indicated
in the associated graph on the right side. The graph shows the
measured phase as a function of the penetration depth for a fixed
frequency of 100 kHz. We chose 100kHz because according to
Figs. 13 and 14, at this frequency we expect a significant change
in the signal depending on which tissue we have in front of the
needle tip.
The second in vitro experiment was done in order to verify
the impact of speed when the needle is moving inside the tissue
during the measurement cycle. The following graph shows the
measured amplitude and phase values at a frequency of 100 kHz
for muscle and fat tissue as a function of speed. The maximum
speed which could be achieved with the given linear actuator
was 14 mm∗s−1which already represents a quite fast insertion
process. During conventional surgical operations the needle in-
sertion speed is usually lower.
The results obtained in the third in vitro experiment are
presented later within Section IX because the data evaluation
is also based on the tissue classification scheme discussed in
Section IX. There the effect of the needle shape is discussed.
VIII. DISCUSSION OF MEASUREMENT RESULTS
A. In vivo Experiments
The presented diagrams in Figs. 12–14 show the measured
complex impedance formed by the tissues on the needle tip. The
data include characteristic tissue properties as well as charac-
teristic electrode polarization effects which cannot be separated
from each other. However, on the first view, the three tissues
seem to have individual characteristic shapes of the plotted
curves. Especially the phase curves show significant differences
in the shape and it looks attractive to classify the different tis-
sue types based on their individual spectra. The classification
approach is done using correlation and is described in detail
within Section IX.
It is remarkable that all measured curves for blood show
an almost perfect match to each other whereas especially the
impedance curves for fat and the phase curves for muscle show
some variations between the lines. According to our measure-
ment experience this is mainly caused by the “quality” of the
contact between the tissue and the needle. Blood is liquid and
always has an ideal contact to the needle tip whereas the con-
tact of fat and muscle tissue depends on the pressure which is
generated by the needle. A high pressure ensures a good contact
whereas a low pressure may significantly affect the measured
values. However, the pressure is given by the mechanical tis-
sue properties, the needle geometry, and the insertion speed.
In real applications, this may vary and cannot be optimized or
stabilized. For tissue classification, this circumstance should be
included in the tests and a resulting classification scheme should
be robust enough to correctly identify the different tissues de-
spite the unknown “quality” of the contact to the needle. In
addition to the variations caused by imperfections of the con-
tact, there is the possibility that small inhomogeneities in the
tested tissues caused variations as well.
B. In vitro Experiments
The measurement results for the first in vitro experiment are
presented in Fig. 15. The graph shows the measured phase
angle as a function of the insertion depth at a frequency of
100 kHz. The measurement values were picked out of the mea-
sured impedance spectrum, which has been recorded for each
measurement position, as described in Section VI-C. The graph
clearly shows that the measured phase value for muscle remains
stable until the boundary layer of muscle and fat is reached with

TREBBELS et al.: ONLINE TISSUE DISCRIMINATION FOR TRANSCUTANEOUS NEEDLE GUIDANCE APPLICATIONS 501
Fig. 16. Modulus and phase angle of the impedance as a function of needle
insertion speed for muscle and fat tissue. The diagram represents the values for
100 kHz obtained from the measured impedance spectra.
the center of the needle tip, indicated with position B. Shortly
after reaching position B, the signal drops to the value measured
for fat where it remains stable again. It is remarkable that there
is a “spatial delay” between position B and the falling signal
edge, which is approximately 0.5 to 1.0 mm. This is caused by
the higher electrical conductivity of the muscle tissue compared
to the fat tissue (according to Fig. 16 approximately by a fac-
tor of two). As long as the needle tip still has contact to the
muscle tissue with higher conductivity, the fat tissue is partially
“shortened.” This limits the steepness of the falling edge as
well. However, according to the simulation result presented in
Section III, the sensitivity of the system should be within 1mm
of the needle tip. The obtained measurement results support
the simulation results very well. The falling signal edge occurs
within 1 mm and the location of the falling edge is within 1mm
to the position B. In summary, the FEM simulation and the ex-
perimental results lead to the conclusion that the sensitivity is
approximately within 1 mm to the needle tip for a needle with
an outer diameter of 2 mm and therefore yields a very good spa-
tial resolution which is ideal for precision needle tip placement
inside a specific target tissue.
The second in vitro experiment shows the amplitude and
phase values for muscle and fat tissue at 100kHz as a func-
tion of speed of the moving needle. The results are presented in
Fig. 16. The speed was varied from 0 to 14mm∗s−1. Obviously,
the measured values remain quite stable, which indicates that
moving the needle within the tested speed regions and within
the tested tissues does not significantly affect the measurements.
IX. TISSUE CLASSIFICATION
A. Classification Method
Goal of the studies is to implement an online tissue classifi-
cation system which enables to identify the type of tissue on the
needle tip in real-time during the insertion process. Key compo-
nent of such a system is a method that allows for identification
of a specific tissue based on the characteristic properties of its
spectrum. We use traditional correlation technique for “compar-
Fig. 17. Averaged curves from n=5 measured transfer functions for muscle
and n=6 measured transfer functions for fat and blood tissue. All tissue types
show an individual characteristic shape.
Fig. 18. The table shows the average correlation coefficientsfor the impedance
modulus curves for each type of tissue correlated to each other averaged transfer
function. The standard deviation is enclosed in brackets.
Fig. 19. The table shows the average correlation coefficients for the phase
curves for each type of tissue correlated to each other averaged transfer function.
The standard deviation is enclosed in brackets.
ing” the shape of a measured spectrum with the shape of several
known tissues contained in a “tissue transfer function library.”
Within our research, we obtained the “library functions” for
each tissue by simply averaging the measured curves presented
in Figs. 12–14. The averaged result for each tissue is presented
below in Fig. 17.
The obtained measurement data (modulus and phase dia-
grams) for each single measurement within each type of tissue
are correlated to all three library functions presented in Fig. 17.
The resulting correlation coefficients for impedance are pre-
sented in Fig. 18, the results for phase in Fig. 19. The tables
show the obtained average value and the standard deviation
enclosed in brackets. A correlation factor of 1 means both cor-
related curves are identical and, therefore, indicate a perfect
match between the measured transfer function and the library
transfer function. A coefficient of 0 shows that there are no sim-
ilar properties and a coefficient of −1 indicates a perfect match
associated with a negative relationship to the original curve.
In order to derive a robust tissue classification scheme, it is
desirable to obtain a correlation coefficient close to 1 for the

502 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 59, NO. 2, FEBRUARY 2012
matching curves (e.g., measured blood curve correlated with li-
brary blood curve) and a coefficient as close to zero as possible
for non-matching curves (e.g., measured blood curve correlated
with library muscle curve). Second important factor for a ro-
bust classification scheme is a tight standard deviation of all
correlated values. This ensures proper classification even in the
case that two average correlation factors for different tissues are
relatively close to each other (e.g., blood to library blood =0.9
and blood to library muscle =0.8).
The data presented in Figs. 18 and 19 show that a tissue clas-
sification of the three tested tissues can be done without any
trouble based on the experimental dataset. Classification can ei-
ther be done on modulus or phase curves only whereas the phase
curves show slightly better correlation results for classification.
The standard deviation is at least by a factor of approximately
30 smaller than the smallest absolute difference in correlation
coefficients (e.g., fat impedance correlated with library fat ver-
sus fat impedance curve correlated with library blood yields an
average difference in the correlation coefficient of 0.992 – 0.907
=0.085 whereas the standard deviation is 0.0027 for fat).
Another interesting point that has to be investigated is the
effect of mechanical variations. Any variation in the needle
geometry, variation in size, and mechanical tolerances as well
as imperfections of the metal surface have impact on the ef-
fective electrode surface and thus affect the electrode transfer
impedance to the tissue as well as the resulting current den-
sity inside the tissue. Since it is difficult to quantitatively cover
the impact of all possible tolerances and variations, we simply
manufactured two more needles with the same outer diameter of
2 mm but different inner conducting tubes and different PTFE
insulation thicknesses. The two additional needles are shown in
Fig. 11(a) and (b). Needle A has an inner diameter of 0.4 mm
and 0.4 mm PTFE insulation, needle B has an inner diameter
of 0.8 mm and 0.2 mm PTFE insulation. All three needles are
used within the same in vitro experiment and it is measured ten
times inside muscle and fat tissue with each needle. The derived
transfer functions of all three needles have been correlated to
“library functions” obtained by only averaging the values from
the original needle C. That means we compare the measurement
results from needles A and B to needle C using the correlation
technique despite the needles have significantly different di-
mensions. The resulting correlation coefficients are presented
in Table I. Tissue classification is still possible, but the average
coefficients become closer to each other compared to the values
shown in Figs. 18 and 19. That means less “buffer” for a defi-
nite tissue classification. However, it shows that the correlation
method has good potential to be robust enough against mechan-
ical imperfections caused by manufacturing of a needle with a
well defined geometry.
B. Limitations
So far we only investigated the potential for classifying be-
tween homogeneous tissue types. Measuring a mix of tissue on
the needle tip will cause the correlation method to fail as it is
implemented right now. In further research, the classification al-
gorithm has to be optimized. By combining modulus and phase
TAB LE I
CORRELATION COEFFICIENTS FOR NEEDLES AAND B
data within the classification, there might be a way for a robust
classification even for mixed tissue volumes containing at least
two unknown tissue types. The system has to be further tested
with more different tissue types as well as with human tissues.
X. CONCLUSION
Throughout this paper, we presented a novel system approach
for online broadband impedance measurement and online tis-
sue classification on the tip of a hollow needle. The devel-
oped system architecture shows good potential for building up
a stand-alone needle guidance system or it could be integrated
into existing needle steering systems. Here it can potentially
improve the weakness of such systems that they often can-
not perfectly classify between different types of soft tissue. The
presented system architecture successfully uses short broadband
chirp signals with well chosen parameters and a following effec-
tive signal processing scheme for obtaining transfer functions
of the load on the needle tip within a very short measurement
time as required by an online system. Tissue classification is
done by simple correlation of the measured complex transfer
functions. The measured functions are compared to known li-
brary functions and based on the correlation coefficients the type
of tissue is identified. The in vivo experiments show excellent
results and allow for a clear tissue identification using either
the modulus or the phase curve of the transfer function whereas
the results obtained by analyzing the phase curve are slightly
better. Additional in vitro experiments indicate that the system
has good potential to be robust against mechanical tolerances of
the needle tip electrodes as well as to be almost immune against
dynamic needle movement within the tested speed range up to
14 mm∗s−1. An FEM simulation and a corresponding in vitro
experiment show an excellent spatial resolution of the sensitive
needle tip, approximately 90% of the signal is derived within
1 mm to the needle tip for a needle with an outer diameter of
2 mm. This is a better resolution than existing prototype systems
based on monopolar electrodes can achieve. According to [14],
a spherical shaped monopolar electrode tip measures 70% of
the total signal within a relatively large volume having a 3.3
times greater diameter than the needle. Clear advantage of the
proposed coaxial needle design is in the close proximity of the
two electrodes at the needle tip and the resulting well defined
current path.

TREBBELS et al.: ONLINE TISSUE DISCRIMINATION FOR TRANSCUTANEOUS NEEDLE GUIDANCE APPLICATIONS 503
REFERENCES
[1] D. Popovic and M. Okoniewski, “Precision open-ended coaxial probe for
dielectric spectroscopy of breast tissue,” in Proc. IEEE Antennas Propag.
Soc. Int. Symp. 2002, vol. 1, pp. 815–818.
[2] M. Vosseler, M. Jugl, and R. Zengerle, “A smart interface for reliable
intradermal injection and infusion of high and low viscosity solutions,”
Pharm. Res., vol. 28, pp. 647–661, 2011.
[3] Y. Thakur, J. S. Bax, D. W. Holdsworth, and M. Drangova, “Design and
performance evaluation of a remote catheter navigation system,” IEEE
Trans. Biomed. Eng., vol. 56, no. 7, pp. 1901–1908, Jul. 2009.
[4] H. Saito and T. Togawa, “Detection of puncturing vessel wall for automatic
blood sampling,” in Proc. 1st Joint BMES/EMBS Conf., 1999,vol.2,
Atlanta, GA, Oct.1999, p. 866.
[5] R. Alterovitz, J. Pouliot, R. Taschereau, I. Hsu, and K. Goldberg, “Sim-
ulating needle insertion and radioactive seed implantation for prostate
brachytherapy,” Stud. Health Tech. Informat., vol. 94, pp. 19–25, 2003.
[6] R. Viard, N. Betrouni, J. Rousseau, S. Mordon, O. Ernst, and S. Maouche,
“Needle positioning in interventional MRI procedure: Real time optical
localisation and accordance with the roadmap,” in Proc. 29th Annu.IEEE
Int. Conf. Eng. Med. Biol. Soc. (EMBS) 2007, Lyon, France, Aug. 22–26,,
pp. 2748–2751.
[7] G. Fichtinger, A. Deguet, K. Masamune, E. Balogh, G. S. Fischer, H.
Mathieu, R. H. Taylor, S. J. Zinreich, and L. M. Fayad, “Image overlay
guidance for needle insertion in CT scanner,” IEEE Trans. Biomed. Eng.,
vol. 52, no. 8, pp. 1415–1424, Aug. 2005.
[8] B. Bascle, N. Navab, M. Loser, B. Geiger, and R. Taylor, “Needle place-
ment under X-ray fluoroscopy using perspective invariants,” in Proc..
IEEE Workshop Math. Methods Biomed. Image Anal., 2000, Hilton Head
Island, SC, Jun. 11–12,, pp. 46–53.
[9] R. S. Singh, S. Natarajan, M. Lee, A. E. Dann, B. P. Cox, D. B. Bennett,
E. R. Brown, H. Lee, W. S. Grundfest, and M. O. Culjat, “Development
of an ultrasound imaging system for needle guidance,” in Proc. IEEE Int.
Ultrason. Symp. (IUS) 2009, Rome, Sep. 20–23,, pp. 1852–1855.
[10] S. Cochran, G. A. Corner, K. J. Kirk, D. I. A. Lines, and M. J. Watson, “P5
C-5 design and validation of an ultrasound array optimised for epidural
needle guidance,” in Proc. IEEE Ultrason. Symp., 2007, Oct. 28–31,
pp. 2255–2258.
[11] A. Haddadi, O. Goksel, S. E. Salcudean, and K. Hashtrudi-Zaad, “On the
controllability of dynamic model-based needle insertion in soft tissue,” in
Proc. 2010 IEEE Annu. Int. Conf. Eng. Med. Biol. Soc. (EMBC), Buenos
Aires, Argentina, Aug. 31–Sep. 4, pp. 2287–2291.
[12] A. Majewicz, T. R. Wedlick, K. B. Reed, and A. M. Okamura, “Evaluation
of robotic needle steering in ex vivo tissue,” in Proc. 2010 IEEE Int. Conf.
Robot. Autom. (ICRA), Anchorage, AK, May 3–7, pp. 2068–2073.
[13] W. Yang, A. Bonvilain, T. Alonso, A. Moreau-Gaudry, and S. Basrour,
“Modeling and characterization of an instrumented medical needle in sight
of new microsensor design for its insertion guidance,” in Proc. 2010 IEEE
Annu. Int. Conf. Eng. Med. Biol. Soc. (EMBC), Buenos Aires, Argentina,
Aug. 31–Sep. 4, pp. 6465–6468.
[14] H. Kalvoy, O. G. Martinsen, and S. Grimnes, “Determination of tissue
type surrounding a needle tip by electrical bioimpedance,” in Proc. 30th
IEEE Annu. Int. Conf. Eng. Med. Biol. Soc. (EMBS) 2008, Vancouver,
BC, Aug. 20–25,, pp. 2285–2286.
[15] H. Kalvoy, L. Frich, S. Grimnes, G. Martinsen, P. K. Hol, and A.
Stubbhaug, “Impedance based tissue discrimination for needle guidance,”
Physiol. Meas., vol. 30, pp. 129–140, 2009.
[16] D. Trebbels, M. Jugl, and R. Zengerle, “Real-time cannula navigation
in biological tissue with high temporal and spatial resolution based on
impedance spectroscopy,” in Proc. Annu. IEEE Int. Conf. Eng. Med. Biol.
Soc. (EMBC) 2010, Buenos Aires, Argentina, Aug. 31–Sep. 4, pp. 1886–
1889.
[17] S. Gabriel, R. W. Lau, and C. Gabriel, “The dielectric properties of biolog-
ical tissues: II. Measurements in the frequency range 10 Hz to 20 GHz,”
Phys. Med. Biol., vol. 41, pp. 2251–2269, 1996.
[18] S. Grimnes and O. G. Martinsen, Bioimpedance and Bioelectricity Basics.
Oxford, UK: Academic Press, 2000.
[19] U. Pliquett, D. Frense, M. Sch¨
onfeldt, C. Fratzer, Y. Zhang, B. Cahill,
M. Metzen, A. Barthel, T. Nacke,and D. Beckmann, “Testing miniaturized
electrodes for impedance measurements within the beta-dispersion—A
practical approach,” J. Electr. Bioimpedance, vol. 1, pp. 41–55, 2010.
[20] T. Paavle, M. Min, and T. Parve, “Using of chirp excitation for
bioimpedance estimation: Theoretical aspects and modeling,” in Proc.
11th Int. Biennial Baltic Electron. Conf. (BEC) 2008, Tallinn, Estonia,
Oct. 6–8, pp. 325–328.
Authors’ photographs and biographies not available at the time of
publication.