1320 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 65, NO.

6, JUNE 2018

Ultra-Wideband Millimeter-Wave Dielectric

Characteristics of Freshly Excised Normal and
Malignant Human Skin Tissues
Amir Mirbeik-Sabzevari , Student Member, IEEE, Robin Ashinoff,
and Negar Tavassolian , Member, IEEE

Abstract—Millimeter waves have recently gained atten- nonionizing electromagnetic waves. These methods are based
tion for the evaluation of skin lesions and the detection of on the inherent contrast between the electrical parameters of
skin tumors. Such evaluations heavily rely on the dielectric malignant and normal tissues [3]–[5]. It has been shown that
contrasts existing between normal and malignant skin tis-
sues at millimeter-wave frequencies. However, current stud- cancer changes the water content as well as the biochemistry
ies on the dielectric properties of normal and diseased skin (e.g., metal concentration) of tissues [6]–[8], and hence results
tissues at these frequencies are limited and inconsistent. In in a change in their dielectric properties.
this study, a comprehensive dielectric spectroscopy study Microwave imaging techniques have been well developed
is conducted for the first time to characterize the ultra-
over the past years for the detection of breast cancer, lung
wideband dielectric properties of freshly excised normal
and malignant skin tissues obtained from skin cancer pa- cancer, and brain stroke in the frequency range of 300 MHz–
tients having undergone Mohs micrographic surgeries at 10 GHz [9]–[12]. As the frequency increases to the millimeter-
Hackensack University Medical Center. Measurements are wave regime (30–300 GHz), shorter wavelengths allow for the
conducted using a precision slim-form open-ended coaxial achievement of higher spatial resolutions at the cost of reduced
probe in conjunction with a millimeter-wave vector network penetration depths. This makes millimeter waves less suitable
analyzer over the frequency range of 0.5–50 GHz. A one-
pole Cole–Cole model is fitted to the complex permittivity for imaging tissues placed deep within the body. However, the
dataset of each sample. Statistically considerable contrasts penetration depth of these waves varies from 600 μm to 1.2
are observed between the dielectric properties of malignant mm into the body [13], making them very effective for sensing
and normal skin tissues over the ultra-wideband millimeter- pathological changes in different skin layers or in the outer tis-
wave frequency range considered. sue layers of excised organs. It is those layers (just below the
Index Terms—Contrast study, dielectric measurements, surface of the skin or organs) which contain certain epithelial
millimeter-wave measurements, malignant skin tissue, nor- cells where a majority of primary tumors originate [14], [15].
mal skin tissue, skin cancer. As the frequency further increases into the THz regime, electro-
I. INTRODUCTION magnetic waves barely penetrate the surface of the tissues [16]
and therefore have limited ability to detect early-stage tumors
KIN cancer is the most common and fastest growing of
S all cancer types with approximately 3.5 million new cases
diagnosed each year in the US [1]. As all kinds of skin cancer
which reside in deeper tissue layers.
The success and efficacy of using millimeter waves for detect-
ing skin tumors largely relies on the dielectric contrasts between
can be easily treated and managed if diagnosed at an early stage normal and malignant skin tissues at these frequencies. How-
[2], it is imperative that dermatologists be astute in the early ever, a definitive knowledge of such contrasts does not currently
diagnosis and treatment of suspected malignancies. exist as no consistent studies have been performed on evalu-
Recently, there has been considerable interest in the de- ating the millimeter-wave dielectric properties of normal and
tection and management of different types of cancer using malignant skin tissues. There is only one study that measures
the reflection characteristics of basal cell carcinoma skin tissues
Manuscript received April 10, 2017; revised August 7, 2017; accepted in 15 skin cancer patients at the frequency of 42 GHz using a
August 30, 2017. Date of publication September 6, 2017; date of current customized WR-22 waveguide [17]. Although large differences
version May 18, 2018. This work was supported by the U.S. National
Science Foundation under Grant CAREER 1554402. (Corresponding were observed between the reflection properties of tumorous
author: Negar Tavassolian.) and normal skin, the study was performed qualitatively and in a
A. Mirbeik-Sabzevari is with the Department of Electrical and Com- limited and small scale. In addition, they used a relatively large
puter Engineering, Stevens Institute of Technology.
R. Ashinoff is with the Hackensack University Medical Center and also and flange-posing waveguide probe which does not provide a
with the New York University/Langone Medical Center. firm and consistent contact with the skin tissue. In this work,
N. Tavassolian is with the Department of Electrical and Computer we systematically evaluate the dielectric properties of a large
Engineering, Stevens Institute of Technology, Hoboken, NJ 07030 USA
(e-mail: [email protected]). number of ‘freshly-excised’ normal and malignant skin tissue
Digital Object Identifier 10.1109/TBME.2017.2749371 samples over an ultra-wide frequency range of 0.5–50 GHz.

0018-9294 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

Authorized licensed use limited to: University of Applied Sciences Karlsruhe c/o KIT Library. Downloaded on December 23,2021 at 10:41:29 UTC from IEEE Xplore. Restrictions apply.
 MIRBEIK-SABZEVARI et al.: CHARACTERISTICS OF FRESHLY EXCISED NORMAL AND MALIGNANT OF HUMAN SKIN TISSUES 1321

The normal and cancer tissue samples are obtained from

patients who have undergone Mohs micrographic surgeries at
Hackensack University Medical Center [18]. As will be ex-
plained in Section II, in this procedure the malignant part of each
tissue sample is located accurately using histology processing
during the surgery process. Dielectric spectroscopy measure-
ments are conducted at Stevens Institute of Technology, Hobo-
ken, NJ. A slim-form, broadband, open-ended coaxial probe
from Keysight Technologies [19] is connected via a coaxial ca-
ble to a millimeter-wave (10 MHz–70 GHz) Anritsu vector net-
work analyzer (VNA). This probe is flange-free and can provide
a firm and consistent contact with the skin. It is also hermeti-
cally sealed at the aperture, making it robust against chemical
reactions between the fluid in the tissue and probe conductor,
and leakage of the sample fluid into the probe aperture. It has a
small sensing volume which enables the precise assessment of Fig. 1. (a) A photograph of the measurement setup. (b) A photograph of
an SCC tissue specimen collected during the study. The contour shows
the desired measurement site, and offers a broadband response the tumor part (also colored in red by the surgeon).
of 0.5–50 GHz.
We employed a data-processing technique developed in our
earlier work [20] to extract the unknown permittivity of the surgeries on patients with basal cell carcinoma (BCC) and
skin tissue under test from the corresponding reflection coef- squamous cell carcinoma (SCC), the two most common forms
ficients measured with the Keysight slim-form probe [19]. At of skin cancer. The tissue collection procedure was approved by
the first step, the reflection coefficients measured at the probe’s the Institutional Review Board at Hackensack University Medi-
calibration plane are transformed to the reflection coefficients cal Center. Mohs surgery is a skin cancer treatment performed in
of the desired aperture-plane (the surface of the tissue under the convenience of the office under local anesthesia. The tumors
test). Then a rational optimization problem is solved to convert are excised in stages, and the sections are assessed histologically
the aperture-plane reflection coefficients to dielectric permittiv- during surgery, allowing the Mohs surgeon to remove all the
ity values [21]. A single-pole Cole–Cole model was then fitted cancer cells while sparing as much normal tissue as possible.
to the complex permittivity dataset obtained for each sample During the first stage of the surgery, the Mohs surgeon sur-
over the measurement frequency range (0.5–50 GHz). Statisti- gically removes the visible portion of the tumor along with an
cal analyses were then performed to compare the ultra-wideband additional thin piece of marginal tissue. The surgical margin
millimeter-wave dielectric properties of malignant and normal (also named the free margin or “uninvolved” skin) generally has
skin tissues. A separate analysis was also performed to com- a thickness of less than 1–1.5 mm and is excised from all around
pare patient-to-patient and sample-to-sample (within the same the tumor. The surgeon then divides the excised tissue laterally
patient) variabilities. into two or four sections (depending on the size of the tissue)
Section II describes the procedures for tissue sample collec- using a scalpel. Next, each section is color-coded with color-
tion and acquiring and processing the data. Section III presents ing dyes for orientation verification in later stages (this step is
methods for fitting and organizing the datasets, as well as called section mapping). Each tissue section is then separately
conducting statistical analyses for dielectric-properties compar- embedded in an optimal cutting temperature (OCT) compound
isons. Section IV discusses the results of the study. Section V and subsequently placed in a cryostat for freezing and slicing.
concludes the paper and provides future directions. OCT accelerates the freezing procedure as well as provides a
convenient matrix for cryostat sectioning.
II. EXPERIMENTAL PROTOCOLS The tissue sections are placed in a cryostat and frozen to a
temperature below −30 °C. Four or five thin slices are then cut
Fig. 1(a) shows the measurement setup including the slim-
from the bottom of the tissues. The thickness of each slice is only
form probe and a cancerous tissue sample. A typical tissue
about 5 μm (0.1 of the thickness of a sheet of paper). The slices
specimen collected during the study is shown in Fig. 1(b). The
are then processed and stained for histology. The remaining
tissue samples are placed on a stand, and the probe is held
part of the tissue is excess to clinical need and is collected
firmly using a jig which is connected to the stand to control the
for our measurements. The slicing procedure in the cryostat is
horizontal and vertical positions of the probe. For each mea-
performed rapidly, generally within a few minutes. Therefore,
surement, the jig is moved such that the entire probe aperture
the tissues do not fully freeze and we are able to collect fresh
makes firm contact with the tissue sample. Sample preparation
tissues in most cases. The thicknesses of the collected specimens
and handling protocols and the data collection procedure are
vary between 3–10 mm and the lateral dimensions are between
presented in detail as follows.
5–30 mm. From this point on, we will use the term ‘sample’ to
refer to the excess section of the tissue collected for dielectric
A. Sample Preparation
characterization.
Fresh, thick, excess specimens were collected from Hacken- All samples contained the layers of stratum corneum, epi-
sack University Medical Center [18] after Mohs micrographic dermis, and dermis. The thicker samples contained a layer of

Authorized licensed use limited to: University of Applied Sciences Karlsruhe c/o KIT Library. Downloaded on December 23,2021 at 10:41:29 UTC from IEEE Xplore. Restrictions apply.
 1322 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 65, NO. 6, JUNE 2018

TABLE I TABLE II
DETAILS OF TISSUE COLLECTION CONDITIONS A FULL LIST OF TISSUES MEASURED EX-VIVO

Total number of patients 41 Patient No. Site Histological Type

Total number of samples 101
Patient age 34–89 years 1 Crown of scalp BCC-significant positive
Time from excision to measurement 120–240 minutes BCC-negative
2 Right distal forearm BCC-negative
3 Upper shin BCC-negative
4 Left scalp BCC-nodular positive
BCC- negative
subcutaneous fat as well. Normally, only a portion of the sus- 5 Upper cutaneous lip BCC-infiltrative positive
pected tissue is cancerous and this portion has the same appear- BCC-negative
ance and color as the rest of the mole. Therefore, a histological 6 Lower cheek SCC-invasive positive
SCC-negative
evaluation is performed during surgery to specify the exact lo- 7 Nasal sidewall BCC-negative
cation of the cancerous part. The histological assessment of the 8 Right temple SCC-in situ positive
last slice adjacent to the bottom of each excess specimen is gen- SCC-negative
9 Forehead SCC-in situ positive
eralized to that specimen. In this way, the cancerous and normal SCC-negative
parts of each sample are located. The normal parts are used as 10 Nose BCC-significant positive
control tissue for comparison purposes. BCC-negative
11 Upper back BCC-negative
The collected samples were immediately placed in sealed 12 Cheek BCC-negative
containers and transferred to the measurement site at Stevens 13 Mandible BCC-negative
Institute of Technology within 2–4 hours of excision. Each tis- 14 Nasal tip BCC-nodular positive
BCC-negative
sue section was placed in a separate 4-oz container. The time and 15 Mid forehead SCC-negative
body part of excision as well as the mapping and histological 16 Left forearm BCC-negative
assessment were labeled on each container. The sample contain- 17 Vertex scalp BCC-negative
18 Nasal root BCC-negative
ers were stored in an insulated packet and kept at a temperature 19 Central scalp BCC-negative
of 4 °C using ice gel packs during transportation. 20 Chin SCC-negative
A total of 101 samples from 41 patients (skin types I–III), 21 Superior Scalp SCC-in situ negative
22 Base of neck SCC-negative
including males and females with an average age of 61 were 23 Proximal calf BCC-negative
collected. Table I summarizes the details of the tissue sam- 24 Left neck BCC-positive
ples collected. The body parts from which the specimens were BCC-negative
25 Post mid parietal scalp SCC-in situ positive
excised are summarized in Table II, column 2. Their final diag- SCC-negative
nosis, presented in column 3, was made by the surgeon accord- 26 Nasal tip BCC-nodular positive
ing to histology evaluations and tumor margin assessments of BCC-negative
27 Right scalp BCC-negative
their adjacent horizontal 5–μm slices cut for histology. As men- 28 Bridge of nose SCC-negative
tioned before, 2–4 samples were collected from each excised 29 Lower leg latva SCC-negative
tissue. Generally, one sample is negative or cancer-free. Also, 30 Mid sternum SCC-positive
SCC-negative
nearly all the samples marked as cancer contain normal parts as 31 Left medial to left eyebrow BCC-positive
well, therefore the number of measurements was greater than BCC-negative
the number of samples. The tissue samples diagnosed as ‘BCC’ 32 Medial ala BCC-positive
BCC-negative
or ‘SCC’ skin cancer are further categorized in terms of their 33 Left forehead SCC-positive
histological types in Table II. SCC-negative
34 Cutaneous lip SCC-positive
SCC-negative
B. Data Collection 35 Infra orbital BCC-negative
36 Temple BCC-negative
Reflection measurements were performed under a protocol 37 Left chin BCC-positive
approved by the Institutional Review Board at Stevens Insti- BCC-negative
38 Left jaw BCC-nodular positive
tute of Technology [22]. Once the tissues were delivered to the BCC-negative
Stevens site, they were mounted on a piece of wipe for measure- 39 Right medial cheek BCC-nodular positive
ments. All measurements were performed at a controlled room BCC-negative
40 Right ear SCC-positive
temperature of 22 °C. SCC-negative
The sensing depth of the probe is approximately 1–3 mm 41 Temporal scalp BCC-negative
[19], meaning that only of a small tissue volume will be
interrogated by the probe. The reflection coefficients at the cal-
ibration port were recorded via the VNA at 142 frequency spots
from 500 MHz to 50 GHz (a frequency step of approximately tissue was identified and marked by the Mohs surgeon during
358 MHz), and then converted into complex dielectric per- the surgery process. To ensure accuracy of measurements, re-
mittivities using the procedure described in our previous work flectometry experiments were conducted on several areas across
[20]. As explained before, the exact cancer site location in each each of the cancerous and normal regions, followed by averag-

Authorized licensed use limited to: University of Applied Sciences Karlsruhe c/o KIT Library. Downloaded on December 23,2021 at 10:41:29 UTC from IEEE Xplore. Restrictions apply.
 MIRBEIK-SABZEVARI et al.: CHARACTERISTICS OF FRESHLY EXCISED NORMAL AND MALIGNANT OF HUMAN SKIN TISSUES 1323

ing the resulting dielectric permittivities. The averaging process

results in 152 separate datasets from over 500 measurements.

III. METHODS
A. Data Fitting
A one-pole Cole–Cole model [23] is applied to all dielectric
parameters obtained from measurements. Cole–Cole models are
normally used to represent physics-based, frequency-dependent
dielectric properties of biological tissues over wide frequency
bands [24]. It has been shown that a one-pole Cole–Cole model
is sufficient to fit human tissue characteristics [25]. The follow-
ing representation is accordingly considered for each of the 152
datasets: Fig. 2. Histology distribution of the 146 datasets measured for this
study.
εr σs
εr (ω) = εr (ω) − jεr (ω) = εr o (ω) + +
1 + ( jωτ ) 1−α jωε0
(1) Group 4 contains all datasets obtained from malignant SCC
where ω is the angular frequency, εr (ω) and εr (ω) are the real tissues. Fig. 2 shows a histogram of the histology distributions
and imaginary parts of the frequency-dependent dielectric per- of the four groups. We separate negative (normal) BCC tissues
mittivity, and εr o , εr , τ , σs , and α are the Cole–Cole model from negative (normal) SCC tissues to eliminate any possible
parameters which are determined by minimizing the following sources of variabilities in the dielectric properties of the
cost function: corresponding samples. As presented in Table II, the collection
 N  εr (ωi )−εr c (ωi )   N  εr ωi )−εrc (ωi )  process (i.e., body location and time of excision) is different
i=1  (εr (ωi ))  + i=1  (εr (ωi ))  for each acquired sample. Therefore, we compare malignant
c= (2) BCC/SCC datasets with normal tissues obtained from either
N
the same sample or an adjacent (bordering) sample. This kind
Here, N = 142 is the number of frequency spots of the mea-
of paired comparison is employed to eliminate any variation
sured data in the frequency range of 0.5–50 GHz (frequency
between the samples in each group.
step of 358 MHz), εr (ωi ) and εr (ωi ) are the experimental val-
Averaged dispersion curves are obtained for each group by
ues measured at the frequency of ωi , and εr c (ωi ) and εrc (ωi )
averaging the permittivity values for all samples in the group
are the values generated by (1). The fitting procedure is ac-
at 142 equally-spaced frequency points, followed by apply-
complished in MATLAB [26] using the Levenberg–Marquardt
ing a one-pole Cole–Cole model. The standard deviations of
algorithm.
the averaged curves at approximately 10, 20, 30, and 40 GHz
The dielectric datasets of the samples that cannot be fitted
are also calculated to specify the extent of the curves in each
with a Cole–Cole model (i.e., the fit error between the model
group.
and measured data across the entire range of frequencies exceeds
1) Comparison Between Dielectric Properties of Normal
a pre-determined threshold of 15%), are excluded from further
and Malignant Skin Tissues: Multivariate analysis of vari-
analysis. The threshold value in our fitting procedure was cho-
ance (MANOVA) [28] is employed to evaluate the dielectric
sen to have scaled Euclidean distances between the measured
properties of the samples. MANOVA is a generalized form of
data and the fitted dielectric curves (averaged over frequency
univariate analysis of variance (ANOVA) and compares the mul-
as specified by (2)) well below those distances considered in
tidimensional (here two-dimensional (2D)) distributions of the
similar studies of biological human tissues [5], [24], [27]. A
responses of the tissue groups rather than one dimension at a
total number of 6 datasets were excluded based on this crite-
time. It therefore provides a more robust analysis when the re-
rion, leaving 146 datasets for statistical analysis. The overall
sponse has several dependent vectors. The 2D responses in our
distribution of the remaining datasets are: 98 normal datasets,
case comprise the real and imaginary parts of dielectric permit-
25 BCC datasets, and 23 SCC datasets. Experimental errors
tivities.
such as imperfect contact of the probe aperture with the tissue
MANOVA as described above is used to compare the dielec-
surface and/or unsatisfactory sizes of tissues under assessment
tric properties of normal and malignant samples at roughly 11
can contribute to the failed fits.
equally-spaced frequency spots across 1–50 GHz. Two sepa-
rate analyses are conducted to compare BCC-normal and SCC-
B. Statistical Analyses normal tissues. As mentioned earlier, the two categories in each
For statistical analyses, four groups of datasets are formed analysis are paired. In other words, for each observation in either
according to the three diagnosis types of the tissues col- of the SCC or BCC cancer groups, there is a normal observation
lected. Group 1 contains all measured datasets obtained from in the corresponding normal groups which is obtained from an
negatively-diagnosed BCC tissues, Group 2 contains all datasets adjacent tissue. This adjacent tissue can be either in the same
obtained from negatively-diagnosed SCC tissues, Group 3 sample, another adjacent sample, or an average of both in cases
contains all datasets obtained from malignant BCC tissues, and where both datasets are available.

Authorized licensed use limited to: University of Applied Sciences Karlsruhe c/o KIT Library. Downloaded on December 23,2021 at 10:41:29 UTC from IEEE Xplore. Restrictions apply.
 1324 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 65, NO. 6, JUNE 2018

Fig. 3. Examples of successful Cole–Cole fits to three representative experimental datasets. (a) Real part, (b) imaginary part of dielectric
permittivity as a function of frequency for a normal (negative BCC) sample; (c) real part, (d) imaginary part of dielectric permittivity as a function of
frequency for a normal (negative SCC) sample; (e) real part, (f) imaginary part of dielectric permittivity as a function of frequency for a BCC cancer
tissue sample; (g) real part, (h) imaginary part of dielectric permittivity as a function of frequency for an SCC cancer tissue sample. o: measured
data, solid line: Cole–Cole fit.

2) Comparison Between Patient-to-Patient and Sample- datasets in each subgroup correspond to the same location on the
to-Sample (Within Patient) Variabilities: Two separate nested body). ANOVA will assess the differences between the means of
analyses of variance (ANOVA) tests are conducted at roughly the subgroups (variability between patients) as well as the spread
11 equally-spaced frequency spots across 1–50 GHz (1 GHz, of the observations around the means (sample-to-sample vari-
5 GHz, 10 GHz, 15 GHz, . . . , 50 GHz) to examine whether ability within patients). In this way, the variations between the
the variability in the dielectric permittivity between patients is samples from different patients are compared with the variations
larger than the variability between samples obtained from each between the samples obtained from the same patient (adjacent
patient. One analysis is conducted on the real parts of the dielec- ones).
tric permittivities and the other on the imaginary parts. Group 1
(negative BCC) is considered for this analysis as it has a larger IV. RESULTS AND DISCUSSION
number of datasets (50 datasets from 23 patients) compared to
A. Data Fitting and Exclusion
the other three groups and therefore more precisely satisfies the
normal distribution assumption required for ANOVA analyses. Figs. 3 and 4 demonstrate the physics-based criterion em-
To apply ANOVA, 23 subgroups are formed with each subgroup ployed to ensure that the dielectric measurements met the
containing the datasets obtained from one patient (therefore all Cole-Cole model consistency requirement. Fig. 3 shows four

Authorized licensed use limited to: University of Applied Sciences Karlsruhe c/o KIT Library. Downloaded on December 23,2021 at 10:41:29 UTC from IEEE Xplore. Restrictions apply.
 MIRBEIK-SABZEVARI et al.: CHARACTERISTICS OF FRESHLY EXCISED NORMAL AND MALIGNANT OF HUMAN SKIN TISSUES 1325

Fig. 4. An example of a poor-quality fit (there were 6 such cases out

of a total of 152 samples). (a) Real part of the dielectric permittivity as a
function of frequency, (b) imaginary part of the dielectric permittivity as Fig. 5. Averaged curves of (a) real and (b) imaginary parts of the
a function of frequency. o: measured data, solid line: Cole–Cole fit. dielectric permittivities of the four dataset groups consisting of negative
BCC (Group 1), negative SCC (Group 2), malignant BCC (Group 3), and
malignant SCC (Group 4) skin tissue samples. Error bars represent the
standard deviation (SD).
examples of successful fits (fit error <15%). In this figure, parts
(a) and (b) show the real and imaginary parts of the dielectric per- TABLE III
mittivity respectively for a normal (negative BCC) sample, parts COLE-COLE PARAMETERS OF THE AVERAGED DIELECTRIC PROPERTIES OF
(c) and (d) show the real and imaginary parts of the dielectric per- THE FOUR TISSUE GROUPS

mittivity respectively for a normal (negative SCC) sample, parts

(e) and (f) show the real and imaginary parts of the dielectric PARAMETER εr o εr τ (ps) σs α
permittivity respectively for a malignant BCC sample, and parts
Group 1 (Normal BCC) 7 39.52 7.96 0.05 0.05
(g) and (h) show the real and imaginary parts of the dielectric Group 2 (Normal SCC) 3 43.89 8.10 0.04 0.15
permittivity respectively for a malignant SCC sample. The mea- Group 3 (Malignant BCC) 6 43.04 7.66 0.05 0.08
Group 4 (Malignant SCC) 10 32.99 3.04 0.01 0.11
sured data are fitted closely with a one-pole Cole–Cole model
for all cases. In contrast, Fig. 4 illustrates an example of a dataset
that could not be fitted with a Cole–Cole model. It should be em-
phasized that out of 152 measurements, only six were excluded which has caused a shift in the relaxation frequency showing its
from further statistical analysis due to not being fitted with a effect as a deviation from the tissue data in 30–50 GHz range.
Cole–Cole model. In a study as large as ours, even with care-
fully following the procedure, it is inevitable that a small number
of measurements would suffer from experimental errors such B. Statistical Results
as unstable contact of the probe aperture with the tissue surface 1) Dielectric Properties of Four Tissue Groups: The av-
and/or unsatisfactory sizes of tissues under assessment. It is seen eraged dielectric properties of the four groups are shown in
in Fig. 4 that the relaxation frequency, which is the frequency Fig. 5. Each curve represents the averaged measured data af-
at which the tissue loss or the imaginary part of the dielectric ter application of a Cole–Cole model, and the variability bars
permittivity is at its maximum [29], is far greater in this tissue around the averages represent the standard deviations. The
compared to other normal tissues. For this tissue the relaxation Cole–Cole parameters obtained for the averaged dielectric prop-
frequency is greater than 50 GHz, whereas for other normal tis- erties of the four tissue groups are listed in Table III. The average
sues with successful Cole–Cole fitting, the relaxation frequency deviations of the dielectric properties of the parametric models
is close to the relaxation frequency of pure water (∼ 20 GHz). were less than 5% from the measured data for both the real
This could be due to an unstable contact of the probe aperture and imaginary parts in all four cases. We would like to empha-
with the tissue. It is shown in [30] that even small amounts of air size that this compact representation of the database is derived
at the probe interface can have large influences on the measure- from carefully managing a large amount of collected data. As
ments due to the low sensing depth of the probe. Therefore, one expected, Groups 1 (negative BCC) and 2 (negative SCC) ex-
possible reason could be the presence of air in the examined area hibit a high similarity in their dispersive dielectric properties.

Authorized licensed use limited to: University of Applied Sciences Karlsruhe c/o KIT Library. Downloaded on December 23,2021 at 10:41:29 UTC from IEEE Xplore. Restrictions apply.
 1326 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 65, NO. 6, JUNE 2018

The small variability observed is due to the different locations (SD = 2.20), while the average imaginary part for SCC cancer
of body from which the datasets are obtained. As mentioned samples is 12.84 (SD = 1.63), which is 16% lower. Similar
before, each group of normal samples included a variety of dif- trends are observed at other frequencies. The largest difference
ferent body locations. If the exact dielectric characteristics are between the average dielectric properties of normal and SCC
needed for each specific location on the body, separate curves cancer samples is approximately 111% (for the real part near
can be obtained followed by another fitting process to produce 50 GHz).
Cole–Cole parameters accordingly. 3) Comparison Between Patient-to-Patient and Sample-
The results obtained here also indicate that over the entire to-Sample (Within Patient) Variabilities: Two ANOVA tests
frequency range of 0.5–50 GHz, malignant BCC (Group 3) were conducted on the 23 subgroups formed from Group 1
has higher real and imaginary parts of dielectric permittivity to evaluate patient-to-patient and sample-to-sample variabili-
compared to normal skin tissue (Group 1). On average, the real ties. Statistically significant differences ( p < 0.05) were ob-
part of the dielectric permittivity is 6% higher near 10 GHz. This served for both the real and imaginary parts of the dielectric
difference exists at higher frequencies as well, with a maximum permittivities at all the 10 frequency spots considered across the
percentage difference of 15% near 28 GHz. The imaginary part 5–50 GHz range. For instance, patient-to-patient and sample-
of the dielectric permittivity is 9% higher, on average, near 10 to-sample standard deviations (SDs) near 40 GHz for the real
GHz for malignant tissues. This difference becomes larger at parts of the dielectric permittivity were 2.85 and 1.16, re-
higher frequencies, with a maximum percentage difference of spectively. For the imaginary part, the SDs were 3.32 and
12% near 50 GHz. 1.85. This means that as expected, there is a greater vari-
SCC tumor tissues (Group 4) also have a higher real part of ability between the patients’ subgroups as compared to the
dielectric permittivity in comparison with their corresponding samples within each patient subgroups. This is expected as
normal skin tissues (Group 2). However, the imaginary part tissues are excised from different body locations for different
for malignant tissues is less than that of normal tissues over patients. However, as explained in Section III-B, this alteration
the entire range of 0.5–50 GHz. For instance, near 10 GHz, does not affect the analysis conducted between normal and ma-
the average real part of the dielectric permittivity is 34.95 for lignant datasets as paired categories were considered for that
normal samples (Group 2), while this value is 41.12 for SCC analysis.
samples (Group 4). The maximum difference between the real
parts of the SCC and normal samples is 111% near 50 GHz. For C. Discussion
the imaginary part, malignant SCC tissues demonstrate lower The results obtained in our study, as demonstrated in Fig. 5,
values over the entire frequency range of 0.5–50 GHz with a indicate that on average, malignant BCC tissues (Group 3)
maximum difference of 49% near 10 GHz. demonstrate a slightly higher dielectric permittivity (both real
2) Comparison Between Normal and Malignant Tissue
and imaginary parts) compared to normal skin tissues (Group
Dielectric Properties: We found statistically significant dif-
1) over the entire frequency range of 0.5–50 GHz. To analyze
ferences between the dielectric properties of normal and BCC and elaborate on these results, we need to understand the di-
cancer samples (p < 0.05) at all the 11 frequency spots consid- electric behavior of the tissues as well as the main components
ered across the 1–50 GHz frequency range. Near 40 GHz, the comprising the skin tissues.
average real part of the dielectric permittivity for normal tissue In [29], three dispersion mechanisms (α, β, γ ) are introduced
samples (Group 1) is 15.74 with a standard deviation (SD) of to characterize the anomalous electric properties of biomaterials.
2.70. At the same frequency, the average real part of the di- The β-dispersion is linked to the cellular structure of biological
electric permittivity for BCC cancer samples (Group 3) is 17.04 materials with low-frequency (up to several MHz) properties
(SD = 1.87), which is 8.5% higher. Similarly, near 40 GHz, the caused by cell membranes [29]. The α-dispersion is attributed
average imaginary part of the dielectric permittivity for normal to the surface admittance which is lateral to the membrane sur-
tissue samples is 14.89 (SD = 3.88), while the average imagi- face. Alpha dispersion can be from below 1 Hz up to 100 kHz
nary part for BCC cancer samples is 16.24 (SD = 1.54), which [29]. The α- and β-dispersion mechanisms fall below the low-
is 9% higher. We see similar trends at other frequencies. The est frequency in the measured data of our study (∼ 500 MHz).
largest difference between the averaged dielectric properties of Hence, they are not considered in our analyses.
normal and BCC cancer tissues is approximately 15% (for the The γ -dispersion is the dielectric dispersion due to the dipolar
real part near 28 GHz). relaxation of water and the capacitive charge of the cellular
Statistically significant differences were also found between membranes in the tissues [29]. The relaxation frequency of the
the dielectric properties of normal and SCC cancer samples γ -dispersion can be roughly estimated from the frequency at
( p < 0.05) at all the 11 frequency spots considered across the which the maximum loss (the imaginary part of the dielectric
1–50 GHz frequency range. Near 40 GHz, the average real part permittivity) is obtained. It is calculated by [29]:
of the dielectric permittivity for normal tissue samples (Group
2) is 15.31 with a standard deviation of 2.59. The average real 1
fc = (3)
part of the dielectric permittivity for SCC cancer tissues (Group 2π τ
4) is 30.28 (SD = 3.17) at the same frequency, which is 97% For the normal and malignant BCC, this frequency is cal-
higher. Similarly, near 40 GHz, the average imaginary part of culated from Equation (3) and Table III to be approximately
the dielectric permittivity for normal tissue samples is 15.27 20 GHz and 21 GHz, respectively. On the other hand, the re-

Authorized licensed use limited to: University of Applied Sciences Karlsruhe c/o KIT Library. Downloaded on December 23,2021 at 10:41:29 UTC from IEEE Xplore. Restrictions apply.
 MIRBEIK-SABZEVARI et al.: CHARACTERISTICS OF FRESHLY EXCISED NORMAL AND MALIGNANT OF HUMAN SKIN TISSUES 1327

laxation frequency of water is 19 ± 1 GHz according to the where εbm is the permittivity of the dry biological structural
parametric model of [29] in the range of tissue temperatures in components (chosen to be 3 to represent an effective permittivity
our study. The relaxation frequency agreements between BCC of protein and lipid molecules [39]), εw is the permittivity of
(normal and malignant) tissues and pure water suggest that the water (taken from the Cole-Cole parametric model at 25 °C
relaxation frequency of the γ -dispersion for the normal and ma- [40]) and  is the volume fraction of the water component in
lignant BCC skin tissues may be largely influenced by that of the tissue. The 10% increase in the real and imaginary parts
pure water, and relatively independent of other skin constituents. of the dielectric permittivity obtained from (4) is consistent
To confirm this hypothesis, we need to understand the main with the results of our study (according to Fig. 5, the real and
constituents of skin tissues as well as the molecular alterations imaginary parts of the dielectric permittivity are 8.5% and 9%
in skin cancer tissues compared to normal tissues. Living or- higher in BCC tissues compared to their corresponding normal
ganisms are primarily composed of water, organic molecules, tissues, respectively). At other frequencies also, the differences
and ions [31]. Water constitutes 70–85% of the total body mass. predicted by (4) agree well with the data obtained in our study.
It constitutes 60–72% of the skin mass [32]. It is known that In contrast to the case of BCC tissues, for malignant SCC tis-
water in biological materials exists in two states: bulk (free) and sues we observed that the real parts of the dielectric permittivity
hydrated (bound) [33]. Hydrated water interacts closely with were larger on average than their corresponding normal ones,
other molecules as well as with bulk water. In the skin tissue, while the imaginary parts were lower than their corresponding
the majority of water molecules seem to be bound only to each normal tissues. The relaxation frequencies of the γ -dispersion
other, similar to the structure of bulk liquid water, rather than for the normal and malignant SCC tissues are calculated from
interacting with other macromolecules [34] and are therefore Equation (3) and Table III to be approximately 20 GHz and
considered to be in the free state. We also note that the dielectric 51 GHz, respectively. The relaxation frequency for the malig-
response of bound water is below 1 GHz and is therefore mostly nant SCC group is far greater than that of water. This indicates
out of the frequency range of interest [35]. that water may not be the main contributor to the dielectric
The remainder of the skin cellular mass is composed of properties of malignant SCC tissues and that they may be more
carbon-containing molecules, which can be divided into four dependent on other skin components.
classes: proteins, nucleic acids, carbohydrates (sugars), and In [41], Raman spectroscopy was employed to differenti-
lipids [29]. Proteins are polymers of amino acids. Proteins con- ate between normal skin and SCC tissues in-vitro. The Raman
stitute 10–25% of the cell mass, the largest contribution after spectra in the normal tissues showed the presence of vibra-
water [32]. DNA and RNA are polymers of nucleotides. DNA tional bands in protein-specific band peaks with higher inten-
and RNA represent 0.3–1% and 0.8–6% of the total cell mass, sities compared to the carcinoma spectra. Another shift region
respectively [32]. Glucose (sugar) level also varies from 0.1–1% in the Raman spectra showed a difference of intensity in the
of the total cell mass [32]. samples of squamous cell carcinoma, which were attributed to
Several Raman spectroscopy studies have been performed to nucleic acids (DNAs). Another Raman spectroscopy study [42]
investigate the molecular structures of different types of skin showed similar changes. In [43] also, variations were observed
cancer. In [36], the nature of water content and structural alter- in the band patterns corresponding to glucose. However, none of
ations in skin malignancy (specifically BCC) were investigated these studies quantified the changes observed in the constituents.
using Raman spectroscopy. An increase of ∼ 15% in the free There is also no study in the literature on the determination of
(bulk) water content was found in malignant BCC skin tissues the water content in malignant SCC tissues.
compared to normal skin tissues. In another study [37], other According to [44], changes in the complex permittivity (both
molecular structure alterations (including proteins and lipids) of real and imaginary values) and a shift in the relaxation frequency
BCC skin cancer tissues were researched. It was seen that the of tissues are expected when the protein and glucose concentra-
Raman intensity of the lipid-specific band peaks were increased tions are increased. These all show that biochemical alterations
due to an increase in the lipid concentration. Also, no significant in malignant SCC tissues, including changes in protein, nucleic
changes were observed in the intensities of protein and nucleic acid, and glucose concentrations play the most important role
acids bands. Both increases in water and lipid concentrations in their dielectric properties. Further studies are required to in-
result in higher dielectric permittivies (both real and imaginary vestigate the water content of SCC tissues as well as quantify
parts) [35], [38]. the effects of each contributor on the millimeter-wave dielectric
These changes are in full agreement with our findings. To characteristics of malignant SCC tissues.
demonstrate this, we assume a binary mixture of water and As also mentioned in Section I, there has been no prior study
other skin components with a concentration of 60% for water to in the literature which consistently evaluates the millimeter-
represent the skin tissue [32]. A 15% increase in the free water wave dielectric properties of normal and malignant skin tissues.
content in this mixture (corresponding to the BCC cancer tissue In the only available study, in-vivo measurements of reflection
[36]) will result in approximately 10% increase in both real and coefficients of normal and malignant BCC tissues were per-
imaginary parts of the dielectric permittivity at 40 GHz per the formed on the faces of 15 skin cancer patients at 42 GHz [17].
Maxwell–Wagner equation [39]: Our results agree with the results of [17]. The magnitude of
the reflection coefficients from BCC samples are lower on av-
(2εbm + εw ) − 2φ(εbm − εw ) erage than the normal ones although the differences in our case
εmix = εbm (4)
(2εbm + εw ) + φ(εbm − εw ) are less than the ones presented in [17] (due to a difference in

Authorized licensed use limited to: University of Applied Sciences Karlsruhe c/o KIT Library. Downloaded on December 23,2021 at 10:41:29 UTC from IEEE Xplore. Restrictions apply.
 1328 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 65, NO. 6, JUNE 2018

cancer, other biochemical changes including protein structures

and glucose and nucleic acid concentrations affect the dielectric
properties of skin tissues at millimeter-wave frequencies.
The wideband dielectric properties reported in this study can
serve as benchmarks in the design of realistic skin phantoms
with dispersive dielectric characteristics at millimeter-wave fre-
quencies. In our future studies, we will customize tissue mim-
icking materials for excellent agreement with the Cole–Cole
curves of normal and malignant skin tissues obtained from this
study over the entire 0.5–50 GHz frequency range.

ACKNOWLEDGMENT
The authors would like to thank the Mohs Surgery and Derma-
tology Surgery Unit at Hackensack University Medical Center
for the assistance of the personnel.

REFERENCES
[1] American Cancer Society. (2017). Cancer Facts and Figures.
[Online]. Available: https://old.cancer.org/acs/groups/content/@editorial/
documents/document/acspc-048738.pdf
[2] American Cancer Society. (2017). Cancer Survival Rates for Melanoma
Fig. 6. (a) Real and (b) imaginary parts of the dielectric permittivities Skin Cancer, by Stage. [Online]. Available: https://www.cancer.org/
corresponding to negative BCC (Group 1), negative SCC (Group 2), and cancer/melanoma-skin-cancer/detection-diagnosis-staging/survival-
wet skin provided by Gabriel et al. [24]. The literature data agree well rates-for-melanoma-skin-cancer-by-stage.html
(error <10%) with the dielectric properties of our normal BCC and SCC [3] A. P. O’Rourke et al., “Dielectric properties of human normal, malignant
tissues over the entire range of 0.5–50 GHz. and cirrhotic liver tissue: In vivo and ex vivo measurements from 0.5 to
20 GHz using a precision open-ended coaxial probe,” Phys. Med. Biol.,
vol. 52, pp. 4707–4719, 2007.
the measurement setup and tissue conditions). The dielectric [4] P. T. Nguyen et al., “Three-dimensional microwave hyperthermia for
properties of normal BCC and SCC tissues derived in our study breast cancer treatment in a realistic environment using particle swarm
were also compared with the literature values obtained from optimization,” IEEE Trans. Biomed. Eng., vol. 64, no. 6, pp. 1335–1344,
Jun. 2017.
in-vivo human experiments. Dielectric properties provided by [5] M. Lazebnik et al., “A large-scale study of the ultrawideband microwave
Gabriel et al. [24] for wet skin agreed well (error <10%) with dielectric properties of normal, benign and malignant breast tissues ob-
the dielectric properties of our normal BCC and SCC tissues up tained from cancer surgeries,” Phys. Med. Biol., vol. 52, pp. 6093–6115,
2007
to the frequency of 50 GHz as shown in Fig. 6. The data for wet [6] V. Suntzeff and C. Carruthers, “The water content in the epidermis of mice
skin at the 0.5–20-GHz range were obtained from experiments undergoing carcinogenesis by methylcholanthrene,” Cancer Res., vol. 6,
[24], and for the 20–50-GHz range were extrapolated from the pp. 574–577, 1946.
[7] K. G. Stern and R. Willheim, The Biochemistry of Malignant Tumors.
measurement data available for frequencies up to 20 GHz in Brooklyn, NY, USA: Reference Press, 1943.
[24] using the Cole–Cole parametric model. [8] L. Mulay et al., “Trace-metal analysis of cancerous and non-cancerous
human tissues,” J. Nat. Cancer Inst., vol. 47, pp. 1–13, 1971.
V. CONCLUSION AND FUTURE DIRECTIONS [9] E. Fear et al., “Enhancing breast tumor detection with near-field imaging,”
IEEE Microw. Mag., vol. 3, no. 1, pp. 48–56, Mar. 2002.
This paper reports on the ultra-wideband (0.5–50 GHz) di- [10] E. Porter et al., “An early clinical study of time-domain microwave radar
for breast health monitoring,” IEEE Trans. Biomed. Eng., vol. 63, no. 3,
electric properties of 101 freshly-excised normal and malignant pp. 530–539, Mar. 2016.
skin tissue samples (resulting in 152 datasets) obtained from skin [11] S. Semenov, “Microwave tomography: Review of the progress towards
cancer patients having undergone Mohs micrographic surgeries clinical applications,” Philos. Trans. A Math. Phys. Eng. Sci., vol. 367,
pp. 3021–3042, 2009.
at Hackensack University Medical Center. A one-pole Cole– [12] A. Zamani et al., “Lung cancer detection using frequency-domain mi-
Cole model is fitted to the complex permittivity datasets of each crowave imaging,” Electron. Lett., vol. 51, pp. 740–741, 2015.
sample, followed by applying a physics-based criterion to ex- [13] S. Alekseev et al., “Millimeter wave dosimetry of human skin,” Bioelec-
tromagnetics, vol. 29, pp. 65–70, 2007.
clude the datasets that cannot be fitted by the model from further [14] J. Rosai and L. Ackerman, Surgical Pathology. New York, NY, USA:
analysis. Mosby, 2004.
A multivariate analysis of variance (MANOVA) is employed [15] A. Uong and L. Zon, “Melanocytes in development and cancer,” J. Cell.
Physiol., vol. 222, pp. 38–41, 2010.
to evaluate the effects of malignancy on the dielectric properties [16] C. S. Joseph et al., “Continuous wave terahertz transmission imaging of
of the samples. Our analyses indicate that statistically significant nonmelanoma skin cancers,” Lasers Surg. Med., vol. 43, pp. 457–462,
differences exist between the dielectric properties of both BCC 2011.
[17] A. Taeb et al., “Millimetre-wave waveguide reflectometers for early de-
and SCC skin cancer tissues and their corresponding normal tection of skin cancer,” IET Microw. Antenna, vol. 7, pp. 1182–1186,
skin tissues and the difference is larger for the case of SCC 2013.
skin cancer. We also provide evidence and discussions into the [18] Hackensack Medical Center. (2017). Dermatologic Surgery Unit.
[Online]. Available: http://www.hackensackumc.org/our-services/surgic
obtained results. We suggest that in addition to a change in al-services/plastic-reconstructive-surgery/mohs-surgery-dermatology-
water content, which is the main cause for the case of BCC skin surgery

Authorized licensed use limited to: University of Applied Sciences Karlsruhe c/o KIT Library. Downloaded on December 23,2021 at 10:41:29 UTC from IEEE Xplore. Restrictions apply.
 MIRBEIK-SABZEVARI et al.: CHARACTERISTICS OF FRESHLY EXCISED NORMAL AND MALIGNANT OF HUMAN SKIN TISSUES 1329

[19] Keysight Technologies, Inc. (2017). N1501A Dielectric Probe Kit. [31] B. Alberts et al., Molecular Biology of the Cell. New York, NY, USA:
[Online]. Available: http://www.keysight.com/en/pd-2492144-pn- Garland, 2002.
N1501A/dielectric-probe-kit?cc=US&lc=eng [32] F. A. Duck, Physical Properties of Tissue. Bath, U.K.: Academic, 1992.
[20] A. Mirbeik-Sabzevari and N. Tavassolian, “Characterization and valida- [33] Z. Wang et al., “Hydration of fat-free body mass: Review and critique
tion of the slim-form open-ended coaxial probe for the dielectric char- of a classic body-composition constant,” Amer. J. Clin. Nutr., vol. 69,
acterization of biological tissues at millimeter-wave frequencies,” IEEE pp. 833–841, 1992.
Microw. Compon. Lett., to be published. [34] O. F. Nielsen et al., “Optical devices and diagnostic in materials science,”
[21] S. S. Stuchly et al., “A new aperture admittance model for open-ended Proc. SPIE, vol. 4098, pp. 160–168, 2000.
waveguides,” IEEE Trans. Microw. Theory Technol., vol. 42, pp. 192–198, [35] K. R. Foster and H. P. Schwan, “Dielectric properties of tissues and bi-
1994. ological materials: A critical review.” Crit. Rev. Biomed. Eng., vol. 17,
[22] Stevens Institute of Technology. (2017). The Office of Sponsored Pro- pp. 25–104, 1989.
grams (OSP). [Online]. Available: https://www.stevens.edu/research- [36] M. Gniadecka et al., “Water content and structure in malignant and benign
entrepreneurship/faculty-research/sponsored-programs skin tumours,” J. Mol. Struct., vol. 661, pp. 405–410, 2003.
[23] K. S. Cole and R. H. Cole, “Dispersion and absorption in dielectrics—I. [37] M. Gniadecka et al., “Melanoma diagnosis by Raman spectroscopy and
Alternating current characteristics,” J. Chem. Phys., vol. 9, pp. 341–352, neural networks: Structure alterations in proteins and lipids in intact cancer
1941. tissue,” J. Invest. Dermatol., vol. 122, pp. 443–449, 2004.
[24] S. Gabriel et al., “The dielectric properties of biological tissues: III. Para- [38] S. R. Smith and K. R. Foster, “Dielectric properties of low-water-content
metric models for the dielectric spectrum of tissues,” Phys. Med. Biol, tissues,” Phys. Med. Biol., vol. 30, pp. 965–973, 1985.
vol. 41, pp. 2271–2293, 1996. [39] R. Pethig, Dielectric and Electronic Properties of Biological Materials.
[25] M. Lazebnik et al., “Ultra-wideband temperature-dependent dielectric New York, NY, USA: Wiley, 1979.
properties of animal liver tissue in the microwave frequency range,” Phys. [40] W. J. Ellison, “Permittivity of pure water, at standard atmospheric pressure,
Med. Biol., vol. 51, pp. 1941–1955, 2006. over the frequency range 0–25 THz and the Temperature Range 0–100 °C,”
[26] The MathWorks, Inc. MATLAB. 2017. [Online]. Available: https://www. J. Phys. Chem. Ref. Data, vol. 36, pp. 1–8, 2007.
mathworks.com/products/matlab.html [41] R. de M. F. Pereira et al., “Diagnosis of squamous cell carcinoma of human
[27] A. Peyman et al., “Variation in dielectric properties due to pathological skin by Raman spectroscopy,” Proc. SPIE. vol. 5326, pp. 106–112, 2004.
changes in human liver,” Bioelectromagnetics, vol. 36, pp. 603–612, 2015. [42] H. Lui et al., “Real-time Raman spectroscopy for in vivo skin cancer
[28] B. S. Everitt and G. Dunn, Applied Multivariate Data Analysis. London, diagnosis,” Cancer Res., vol. 72, pp. 2491–2500, 2012.
U.K.: Arnold, 2000. [43] A. C. Lieber et al., “Raman microspectroscopy for skin cancer detection
[29] H. P. Schwan, “Electrical properties of tissue and cell suspensions,” in in vitro,” J. Biomed. Opt., vol. 13, pp. 0240131–0240138, 2008.
Proc. IEEE Conf. Eng. Med. Biol. Soc., 1994, pp. A70–A71. [44] M. Zhadobov et al., “Complex permittivity of representative biological
[30] P. M. Meaney et al., “A coaxial dielectric probe technique for distin- solutions in the 2–67 GHz range,” Bioelectromagnetics, vol. 33, pp. 346–
guishing tooth enamel from dental resin,” Adv. Biomed. Eng. Med. Phys., 355, 2012.
vol. 3, pp. 8–17, 2015.

Authorized licensed use limited to: University of Applied Sciences Karlsruhe c/o KIT Library. Downloaded on December 23,2021 at 10:41:29 UTC from IEEE Xplore. Restrictions apply.