Revisited Otsu Algorithm for Skin Cancer Segmentation
ALAA TOM, JIHAD DABA
Department of Electrical Engineering,
University of Balamand,
LEBANON
Abstract: - Computer-aided technology can be used to perform a quantitative and objective evaluation of pigmented
skin lesions during the clinical assessment procedure. This helps to expedite the procedure. The growing
development of non-invasive techniques can be of significant benefit in the early identification of malignant
melanoma, which can, in turn, help to minimize the necessity for invasive biopsies. The system is primarily
focused on two principal schemes: Establishing an effective lesion border detection method and then creating an
efficient classification scheme. We address two primary areas in this work. First, we study skin lesion detection to
analyze any sign of malignancy for skin cancer diagnosis. This is followed by the system implementation of a
color-based method for all the images from the RGB color space using a revisited OTSU thresholding
segmentation scheme. The results proved to be promising with at least an 80% accuracy detection rate for a wide
range of clinical skin lesion images.
Keywords: - ABCD method, automatic lesion, melanoma, Otsu algorithm, segmentation
Received: April 25, 2022. Revised: January 8, 2023. Accepted: February 5, 2023. Published: March 2, 2023.
1 Introduction
One of the most deadly illnesses and the main cause
of death worldwide is cancer. According to WHO
(World Health Organization) statistics, cancer killed
10 million people globally in 2020, [1], and by 2030,
there would be 13.1 million cancer-related deaths,
according to predictions, [2].
With 1.2 million cases expected in 2020, skin
cancer is the fifth most prevalent malignancy overall,
[2]. It is divided into three types: Skin cancer that is
not melanoma, such as basal cell and squamous cell
cancer, and melanoma skin cancer that affects
melanocytes. Squamous cell cancer and basal cell
cancer are both common but less dangerous.
Compared to non-melanoma skin cancers, melanoma
is less common, but it has a higher risk of spreading
and becoming lethal, [3].
Although skin cancer is a life-threatening disease,
with a poor prognosis, especially for melanoma, it
may be treated if caught early. According to studies,
the cure rate is more than 90% if melanoma is
detected early on, whereas the cure rate is less than
50% if it is detected late, [4]. The largest phenotypic
risk factor for this condition is the presence of a
significant number of melanocytic nevi, especially in
the case of atypical cutaneous syndrome. Other risk
factors include fair skin, a high density of freckles,
and eyes and hair colour. Melanoma is currently
diagnosed mostly by a dermatologist by examining
the skin with his naked eyes or with a derma-scope
during the diagnosing process.
Dermatologists commonly utilize the ABCDE
(Asymmetry─Border─Color─Diameter─Evolution)
criteria to discover signs of skin cancer during this
stage, [5]. Asymmetry, borders, colors, diameters,
and evolution (or evolving) are all characteristics of
a skin patch, [5]. Dermatologists can determine if a
skin patch is benign or malignant by analyzing the
early indications and applying the ABCDE approach,
[3].
Computer-aided skin cancer detection approaches
have been studied to help dermatologists in
melanoma diagnosis to enhance the diagnostic rate.
The main idea behind computer-aided detection
(CAD) is to detect questionable patches of the skin
using image processing and pattern recognition
techniques. It is commonly utilized as a
dermatologist's recommendation. In this work, we
are mainly interested in the primary approach used in
CAD systems to identify skin cancer.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.7
Alaa Tom, Jihad Daba
E-ISSN: 2224-3402
50
Volume 20, 2023
2 Literature Survey
The detection of skin cancer using CAD systems has
been a topic of intensive research for more than 30
years, [6]. For instance, numerous melanoma
detection techniques have been created and
researched, [7]. A fascinating overview of methods
that have been tested on clinical and dermoscopic
images from 1984 to 2012 was provided by Korotkov
and Garcia in [8]. They arranged the review from the
data acquisition stage to the classification and
diagnosis using ABCDE criteria, [7], as well as
several other methods created in clinical and CAD
systems. Additionally, Maglogiannis and Doukas
reported in [9] a non-exhaustive comparison of the
most cardinal implementations, particularly
characteristics selection like colour and border, and
they offered the most widely used classifiers in the
literature: ANNs (Artificial Neural Networks) and
SVMs (Support Vector Machines).
Abbas et al. in [11] employed modified RACs
(Region-based Active Contours) created by Lankton
et al. in [10] using total variation regularization for
the segmentation part. If the boundary lesion was
visible and had a strong contrast with the surrounding
normal skin, their approach could segment the lesion
border accurately. It did not, however, pick up lesions
that blend seamlessly into the healthy skin around
them.
Another strategy utilized by Ma et al. in [12] is
wavelet decomposition banks for artificial neural
network classifiers for melanoma and non-melanoma
case discrimination. 72 melanoma and 62 benign
tumors from a database were subjected to this
scheme, making a total of 134 skin lesion images.
The drawn results reached a sensitivity of 90% and a
specificity of 83%.
Celebi et al. proposed in [13] an automated fusion
of the thresholding method with a Markov random
field; they examined 90 dermoscopic images with 23
malignant melanoma and 67 benign lesions; they
then compared their findings to the state-of-the-art
methods, expressing the results using exclusive-OR
errors of 9.16 ± 5.21 %.
3 Methodology
Our study is divided into four major stages as
depicted in the block diagram in Fig. 1. The first
stage comprises the pre-processing stage comprising
image enhancement, resizing, and getting rid of all
unwanted details. The second stage is the
segmentation stage consisting of segmenting the skin
lesions from the input image and performing image
processing techniques. The third stage consists of the
feature extraction stage including the ABCD
parameters and some other parameters such as area,
perimeter, and axis. And finally, we employ the TDS
calculation stage which states that the lesion is either
benign, suspicious, or malignant.
Fig. 1: Block diagram of the proposed algorithm
3.1 Preprocessing
The majority of dermoscopy images need some
adjustments when it comes to brightness, contrast,
thick hair, and any unwanted detail such as air
bubbles, etc. For this matter, a median filter is used to
get rid of every detail considered noise.
3.1.1 Data collection
Any image processing technique starts with acquiring
the image, which is the process of selecting an image
from the databases selected before.
3.1.2 Image input
The acquired image is usually in the RGB color
space because all pictures are coded according to the
jpg or jpeg standards.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.7
Alaa Tom, Jihad Daba
E-ISSN: 2224-3402
51
Volume 20, 2023
3.2 Segmentation
The purpose of the segmentation process is to make
work easier or to convert the image into a
representation that is easier to comprehend and
evaluate. Using Otsu's thresholding, a grayscale
image can be converted into a binary image for
clustering-based image segmentation.
Otsu's threshold-based segmentation is employed
as the technique for carrying out the segmentation
work for our method.
3.2.1 Gray scaling
The bulk of jpg and jpeg photos are in the RGB color
space, making it one of the most frequently used
color spaces, [14]. The RGB images are converted
into gray images, meaning that the red, blue, and
green values are averaged to get the intensity of the
grayscale image using the following equation:
Grayscale ,
3
R G B
(1)
where R is the value of the red pixel, G is the value
of the green pixel, and B is the value of the blue
pixel.
3.2.2 Histogram
A critical stage before the thresholding process is the
construction of the histogram of image pixel
distribution. In the histogram analysis, we should
distinguish between the foreground and the
background. The pixel probability distribution of
each level is computed by the following equation:
,
i
i
n
PN
(2)
where Pi is the pixel probability to i, ni is the pixel
with grayscale level i, and N is the total number of
pixels in the image.
3.2.3 Binarization
After gray scaling the image, we should perform
binarization, which is transforming the gray image
into a binary image. The binary image consists of 2-
pixel values: either 0 or 1. The importance of the
binary image is the increase in efficiency and
decrease in load. To achieve this step, there are two
ways: Global binarization and adaptive binarization.
Global binarization consists of one threshold value
for all the images, while adaptive binarization
consists of a changing threshold value for every
pixel, [10].
3.2.4 Image negation
The act of replacing each white pixel (of value 1)
with a black pixel (of value 0), and vice versa, is
known as image negation. This process assures a
better representation of the ROI (Image Region of
Interest), where the ROC is shaded in white.
3.3 Feature Extraction
The ABCD rule is utilized in the process of
distinguishing benign melanocytic tumors from
malignant melanocytic tumors. The asymmetry,
border, color, and diameter are the four factors that
are considered while applying the ABCD formula to
get the TDS (Total Dermoscopy Score) index.
3.3.1 Asymmetry
Asymmetry is a crucial element that plays a
fundamental role in computer-aided diagnosis. The
suspicious mole is evaluated in 0, 1, or 2 axes. In
other words, the lesion is cut in half and the two parts
should match one another, regardless of the axis of
symmetry. Otherwise, there is an asymmetry score
“A” computed as follows:
0: bi-axial symmetry
1: mono-axial symmetry
2: bi-axial asymmetry
3.3.2 Border
The border score “B” ranges from 0 to 8. To compute
the border irregularity, we should find the perimeter
and the area of the acquired image. And then B is
determined using the following equation:
2,
4
ip
Ba
(3)
where p denotes the lesion's perimeter, a denotes the
lesion's area and
is a controlling parameter set at
22.7.
3.3.3 Color
The RGB color space's three components are
distributed among the color parameter “C”. When it
comes to scoring, the value ranges from 1 to 6,
indicating the presence of up to six well-known
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.7
Alaa Tom, Jihad Daba
E-ISSN: 2224-3402
52
Volume 20, 2023
colors, such as black, white, slate blue, brown, and
red.
3.3.4 Diameter
A cancerous mole usually consists of a diameter “D”
wider than 6 mm. If this occurs, then the mole is
most probably a sign of malignancy, and
automatically we set D to a value of 5.
3.4 TDS Calculation
The total dermoscopy score or TDS index is an
essential criterion for melanoma diagnosis. The
ABCD parameters of the skin lesion are the four
factors that go into the calculation of the TDS index.
If the TDS is below 4.75, the lesion is considered to
be benign. If the TDS is above 5.45, this is an index
that the skin growth is malignant. If the TDS is
between 4.75 and 5.45 the skin lesion is suspicious.
The TDS index can be found by the formula
below:
TDS 1.3 0.1 0.5 0.5 .A B C D
(4)
3.4.1 Accuracy Testing
When a testing technique correctly identifies a
fraction of its true positives (TP), as suggested by Eq.
(5) below, it is said to have high sensitivity, which is
a measurement of how well a test can detect disease.
The specificity of a diagnostic test refers to the
percentage of true negatives that it correctly
identifies. It indicates how well the test can
distinguish bad situations (normal conditions).
Accuracy is defined as the percentage of true
outcomes in a population, which may be true positive
(TP) or true negative (TN). It evaluates the accuracy
of a diagnostic test when used to treat a condition and
then summarizes the findings.
Sensitivity, specificity, and accuracy are described
in Eqs. (5), (6), and (7), respectively, in terms of TP
(true positive assessments), TN (true negative
evaluations), FP (false positive evaluations), and FN
(false negative evaluations):
TP
Sensitivity ,
TP+ FN
(5)
which denotes the number of true positive
assessments over the number of all positive
assessments;
TN
Specificity ,
TN + FP
(6)
which denotes the number of true negative
assessments over the number of all negative
assessments;
TP + TN
Accuracy ,
TP + FN + FP + TN
(7)
which denotes the number of correct assessments
over the number of all assessments.
As stated earlier, we took a sample of 30 images
in our study. Eighteen of the selected images were
melanoma cases and the other twelve were non-
melanoma. After testing, 15 images were analyzed as
TP, 6 as TN, 3 as FP, and 2 as FN. Table 1 highlights
the resulting accuracy parameters.
Table 1. Accuracy parameters
Sensitivity
Specificity
Accuracy
88.2%
66.7%
80.1%
4 Results and Discussions
The Otsu threshold was developed in MATLAB for
this work, which involves the segmentation of
dermoscopy images in melanoma. Figure 2 is an
illustration of an Otsu thresholded dermoscopy image
with a resolution of 1021 by 766 pixels.
During the initial step of this research, a
dermoscopy image of melanoma was entered into the
system so that it could be converted from an RGB
image to a grayscale image that had been calculated
earlier. The result of effectively converting an RGB
image into a gray level is shown in Fig. 3.
Fig. 2: Input image
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.7
Alaa Tom, Jihad Daba
E-ISSN: 2224-3402
53
Volume 20, 2023
Fig. 3: Gray-scaled image
The development of a histogram from a grayscale
image is the next step to follow. Creating this
histogram is helpful before moving on to the next
segmentation stage. The histogram of the grayscale
image is depicted in Fig. 4.
Fig. 4: Histogram analysis
Image segmentation with Otsu Thresholding is
applied to the grayscale image after the histogram of
the image has been created. The image shown in Fig.
5 is segmented using Otsu thresholding.
The between-class variance value of the input
image is 1843, while the threshold value of the image
itself is 125.
Fig. 5: Segmented legion
The color analysis of the segmented image using
MATLAB identifies the presence of the known colors
stated in the methodology section. Every color that
has a percentage greater than 5% increases the TDS
score by 1. Table 2 and Fig. 6 display the outcomes
of the color analysis of the segmented image.
Table 2. Segmentation of dermoscopic images
Fig. 6: Estimated color score
5 Conclusion
In this study, we investigated the development of a
skin lesion adaptive segmentation as well as the
classification of skin lesions into malignant and
benign cases using a set of extracted features.
The primary goal was the creation of a customized
adaptive segmentation scheme for skin lesions. The
research findings demonstrated that the revisited
Otsu algorithm-based segmentation strategy
performed very well for the selected database of
clinical skin lesion images and successfully produced
a promising minimum of 80% accuracy detection
rate.
Undoubtedly, the proposed system suffers from
several limiting constraints which are addressed in
the next section.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.7
Alaa Tom, Jihad Daba
E-ISSN: 2224-3402
54
Volume 20, 2023
6 Future Work
Since our presented system has its limitations, we
propose the following recommendations for future
work:
1. The proposed algorithm considered image
characteristics enhancement as an important
system phase, and it was mainly focused on
brightness, contrast, and quality. However, for
better results and image representation, it
would be best to include air bubbles and hair
detection as pre-processing sub-systems to
produce a better reading of the image.
2. The algorithm presented a perfect malignancy
detection rate, while for the benign cases, the
algorithm failed to give an appropriate lesion
for the area of interest. This is a main
limitation of our study and deep learning-
based algorithms need to be developed to
address this shortfall.
3. All images were selected from a specific
database obtained from the website
dermis.net, and as a result, the generalizability
of the results is considered limited. The
database of images needs to be extended to
include other open-source databases of
clinical images.
A more ambitious study would be to consider a
more advanced stochastic model that captures the
biological and epidemiological characteristics of the
ultraviolet induction of skin cancer, [15]. Under such
models, the number of cancerous cells follows a
doubly stochastic marked Poisson point (DSMPP)
process that is randomly distributed in time and space
and whose estimated density (intensity or rate of the
point process) gives an indication for the presence of
skin cancer or its absence, [16], [17], [18], [19], [20],
[21], [22], [23], [24], [25], [26]. Unexpected spinoffs
of classification and estimation algorithms employed
in cellular communication systems, [27], and next-
generation wireless systems, [28], [29], [30] can be
made for skin cancer detection under the proposed
DSMPP models.
7 Appendix A: Small Database of
Segmented Dermoscopic Images
Table 3 summarizes the segmentation of 20
dermoscopic images with the results followed by an
accuracy detection rate.
Table 3. Segmentation of selected dermoscopic
images
Dermoscopic image
Segmentation
Result
TN
TP
TN
FN
TN
FP
TP
TP
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.7
Alaa Tom, Jihad Daba
E-ISSN: 2224-3402
55
Volume 20, 2023
TP
TN
TP
FN
TP
TP
TP
FP
TN
TP
TP
TP
TP
As shown in Table 3, we obtained a sample of 21
images with 12 TPs and 5 TNs, resulting in an
accuracy detection rate of 80%.
References:
[1] S. Jaiswar, M. Kadri, V. Gatty, "Skin cancer
detection using digital image processing,"
International Journal of Scientific Engineering
and Research, vol. 3 no 6, pp. 138-140, June
2015.
[2] M. Chandrahasa, V. Vadigeri, D. Salecha,
"Detection of skin cancer using image
processing techniques," International Journal
of Modern Trends in Engineering and Research
(IJMTER), vol. 3, no. 5, pp. 111-114, May
2016.
[3] M. A. Farooq, M. A. Azhar, "Automatic lesion
detection system (ALDS) for skin cancer
Classification Using SVM and Neural
classifiers," 2016 IEEE 16th International
Conference on Bioinformatics and
Bioengineering, pp. 301-308, 2016.
[4] N. Razmjooy, S. Mousavi, et al, “A computer-
aided diagnosis system for malignant
melanomas,” Neural Computing &
Applications, vol. 23, no. 7/8, pp. 2059-2071,
2013. DOI: 10.1007/s00521-012-1149-1.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.7
Alaa Tom, Jihad Daba
E-ISSN: 2224-3402
56
Volume 20, 2023
[5] D. Thanh, V. B. Prasath, L. Hieu, et al.,
“Melanoma skin cancer detection method
based on adaptive principal curvature, colour
normalisation and feature extraction with the
ABCD rule,” Journal of Digital Imaging, vol.
33, pp. 574–585, 2020.
[6] I. M. Widyantara, A. T. Kusuma, N. M.
Wirastuti, “Preprocessing pada segmentasi
citra paru-paru dan jantung menggunakan
anisotropic diffusion filter, Open Journal
Systems-Teknologi Elektro, vol. 14, no. 2, pp.
6-10, 2015.
[7] K. Korotkov, R. Garcia, “Computerized
analysis of pigmented skin lesions: A review,”
Artificial Intelligence in Medicine, vol. 56, no.
2, pp. 69-90, 2012.
[8] I. Maglogiannis, C. N. Doukas, “Overview of
advanced computer vision systems for skin
lesions characterization,” IEEE Transactions
on Information Technology in Biomedicine,
vol. 13, no. 5, pp. 721-733, 2009.
[9] Z. She, Y. Liu, A. Damatoa, “Combination of
features from skin pattern and abcd analysis for
lesion classification,” Skin Research and
Technology, vol. 13, no. 1, pp. 25-33, 2007.
[10] N. R. Abbasi, H. M. Shaw, D. S. Rigel, R. J.
Friedman, W. H. McCarthy, I. Osman, A. W.
Kopf, D. Polsky, “Early diagnosis of cutaneous
melanoma: Revisiting the abcd criteria,” Jama,
vol. 292, no. 22, pp. 2771-2776, 2004.
[11] Q. Abbas, I. Fondón, M.
Rashid, “Unsupervised skin lesions border
detection via two-dimensional image analysis,”
Computer Methods and Programs in
Biomedicine, vol. 104, no. 3, pp. e1–e15, 2011.
[12] S. Lankton, A. Tannenbaum, “Localizing
region-based active contours,” IEEE
Transactions on Image Processing, vol. 17. no.
11, pp. 2029-2039, 2008.
[13] L. Ma, B. Qin, W. Xu, L. Zhu, “Multi-scale
descriptors for contour irregularity of skin
lesion using wavelet decomposition,” 2010 3rd
International Conference on Biomedical
Engineering and Informatics, vol. 1, IEEE
Press, pp. 414-418, 2010.
[14] S. Sabbaghi, M. Aldeen, R. Garnavi, G.
Varigos, C. Doliantis, J. Nicolopoulos,
“Automated colour identification in
melanocytic lesions,” Engineering in Medicine
and Biology Society (EMBC), 2015 37th
Annual International Conference of the IEEE,
IEEE Press, pp. 3021-3024, 2015.
[15] D. Pearl, “Stochastic model to explain the
biology and epidemiology of the ultraviolet
induction of skin cancer,” Mathematical
Population Dynamics, Proceedings of the
Second International Conference, Lecture
Notes in Pure and Applied Mathematics, vol.
131, CRC Press, 1991.
[16] J. S. Daba, M. R. Bell, “Segmentation of
speckled images using a likelihood random
field model,” Optical Engineering, vol. 47, no.
1, 2008.
[17] J. Dubois, “Scattering statistics of doppler
faded acoustic signals using speckle noise
models,” IEEE International Conference on
Direct and Inverse Problems of
Electromagnetic and Acoustic Wave Theory
(DIPED), Lviv, Ukraine, pp. 185-189, 2003.
[18] J. S. Daba, M. R. Bell, “Estimation of the
surface reflectivity of SAR images based on a
marked Poisson point process model,” IEEE
International Symposium on Signals, Systems,
and Electronics (ISSSE'95), San Francisco, CA,
pp. 183-186, 1995.
[19] J. S. Daba, M. R. Bell, A. Abdi, S. Nader-
Esfahani, “Comments on statistics of the
scattering cross-section of a small number of
random scatterers," IEEE Transactions on
Antennas and Propagation, vol. 48, no. 5, pp.
844-845, 2000.
[20] J. P. Dubois, O. M. Abdul-Latif, “SVM-based
detection of SAR images in partially developed
speckle noise,” IEC, Prague, pp. 321-325,
2005.
[21] J.P. Dubois, O. M. Abdul-Latif, “Detection of
ultrasonic images in the presence of a random
number of scatterers: A statistical learning
approach,” IEC, Prague, pp. 326-329, 2005.
[22] J. Dubois, “Poisson modulated stochastic
model for partially developed multi-look
speckle,” Proceedings of the American
Conference on Applied Mathematics, Harvard
University, Cambridge, MA, USA, pp. 209-
213, 2008.
[23] J. Dubois, “Segmentation of speckled
ultrasound images based on a statistical
model,” EURASIP Proceedings of the 16th
International Biosignal Conference,
Czech Republic, vol. 16, 2002.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.7
Alaa Tom, Jihad Daba
E-ISSN: 2224-3402
57
Volume 20, 2023
[24] J. S. Daba, M. R. Bell, “Statistics of the
scattering cross-section of a collection of
constant amplitude scatterers with random
phase,” ECE Technical Reports, Purdue
University, p. 194, 1994.
[25] J. S. Daba, M. R. Bell, “Object discrimination
and orientation determination in speckled
images,” Optical Engineering, vol. 33, no. 4,
pp, 1287-1303, 1994.
[26] J. Dubois, “Traffic estimation in wireless
networks using filtered doubly stochastic point
processes,” Proceedings of IEEE International
Conference on Electrical, Electronic, and
Computer Engineering, Cairo, Egypt, pp. 116-
119, 2004.
[27] J. Dubois, J. S. Daba, M. Nader, C. El Ferkh,
“GSM position tracking using a Kalman filter,”
International Journal of Electrical, Computer
and Communication Engineering, vol. 6, no. 8,
pp. 867-876, 2012.
[28] J. Dubois, O. M. Abdul-Latif, “Novel diversity
combining in OFDM-based MIMO systems,”
Proceedings of the American Conference on
Applied Mathematics, Harvard University,
Cambridge, MA, USA, pp. 189-194, 2008.
[29] J. Dubois, O. Abdul-Latif, “Improved receiver
diversity processing over SIMO fading
channels,” Proceedings of the IEEE
International Conference on Signal Processing
and Communications, 2007.
[30] O. M. Abdul-Latif, J. Dubois, “Performance of
UWB system in a partially developed fading
channel with CCI,” Proceedings of the 5th
IEEE GCC Communication and Signal
Processing Conference, Kuwait, pp.1-5, 2009.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.7
Alaa Tom, Jihad Daba
E-ISSN: 2224-3402
58
Volume 20, 2023
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
The authors equally contributed in the present
research, at all stages from the formulation of the
problem to the final findings and solution.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
No funding was received for conducting this study.
Conflict of Interest
The authors have no conflicts of interest to declare
that are relevant to the content of this article.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
