Some Segmentation Approaches of Breast Magnetic Resonance Imaging
Tumor in Computer Aided Detection Systems: A Review
ALI AL-FARIS
Computer Science Department
Worcester State University
486 Chandler St, Worcester, MA 01602
UNITED STATES OF AMERICA
Abstract: - This article discusses the approaches and procedures that have been applied to MRI breast tumor
segmentation particularly as well as breast segmentation systems in general. The review begins by outlining the
various breast screening methods and contrasting Magnetic Resonance Imaging (MRI) with other methods like
Mammography, X-ray, and Ultrasonography. Next, it is emphasized how important Computer Aided Detection
(CAD) systems are for Breast MRI. Review and comparison of supervised, unsupervised, and semi-supervised
breast MRI tumor segmentation techniques are done. The study concludes with a discussion and
recommendations based on the methods examined.
Key-Words: - Segmentation, Breast Images, MRI, CAD.
Received: May 17, 2022. Revised: December 9, 2022. Accepted: January 11, 2023. Published: February 14, 2023.
1 Introduction
The most frequent malignancy in women worldwide
is breast cancer, [1], [2]. The International Agency
for Research on Cancer (IARC), an
intergovernmental organization affiliated with the
World Health Organization of the United Nations,
projected that 2.1 million new cases of breast cancer
were identified in 2018. The eight malignancies
with the highest global incidence are depicted in
Fig. 1 along with an estimation of the total number
and percentage of newly diagnosed cases. The
biggest cause of death for women worldwide today
is breast cancer, [3]. In 2018, there were about
627,000 breast cancer fatalities reported. Based on
an IARC study [4], [5], and Fig. 2, the total number
of cancer-related fatalities worldwide is depicted.
Techniques for breast screening are crucial for
cancer detection. The chances that a breast cancer
patient would survive are considerably increased by
precise segmentation for suspected tumors utilizing
computer algorithms. To limit the amount of false-
positive results, image processing techniques are
required to aid radiologists in deciphering the
images and segmenting tumor regions [6].
Fig. 1: Based on an IARC investigation, an
estimated number of cancer cases have been
diagnosed worldwide [4].
Fig. 2: Based on an IARC study, the number of
cancer deaths worldwide is estimated to be [4].
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2023.11.1
Ali Al-Faris
E-ISSN: 2415-1521
1
Volume 11, 2023
2 Breast Screening Mechanisms
Different screening procedures are employed for a
more thorough evaluation in addition to self-
checking and physical inspection for probable breast
cancers. The most popular breast screening methods
in medical settings are; Mammography,
ultrasonography, and magnetic resonance imaging
(MRI).
2.1 Mammography
Mass screening programs frequently employ the
non-invasive X-ray method known as
mammography. To obtain the image's details, this
method includes subjecting the breast to a little
quantity of ionizing radiation [7]. Due to its ability
to create an acceptable image of abnormalities and
its capacity to reveal indirect calcifications,
mammography is frequently used as an image
screening modality [8], [9].
Mammography does, however, have several flaws
and restrictions. These flaws can be seen in
recognizing very small tumors, contrast
characteristics, and narrow dynamic range [10]. Fig.
3 displays a few Mammogram image samples [11].
Fig. 3: Mammogram image examples [11].
2.2 Ultrasonography
Another non-invasive screening method that makes
use of sound waves to visualize the breast is
ultrasound. If a mass contains solid or fluid, the
ultrasound image could be helpful [7], [12].
One benefit of ultrasonography is that it may find
cancers that mammography may not be able to
identify as solid or liquid. Additionally, the method
results in less discomfort, costs less money and has
no negative health repercussions [13].
On the other side, misleading positive results from
ultrasound images may result in misdiagnosis [14].
Additionally, ultrasonography is not frequently used
in clinical settings, and using the equipment requires
highly qualified professionals [14], [15]. Fig. 4
displays some breast ultrasound imaging samples
[16].
Fig. 4: Breast ultrasound image examples [16]
2.3 MRI Screening
A non-invasive imaging method is MRI screening.
It has been extensively utilized for medical imaging,
including breast screening and imaging of the brain,
spine, bones, and joints. It is based on magnetic and
radio frequency fields. A discernible signal is
created as a result of how the radio frequency pulses
affect how the resonant nuclei are arranged [17].
Fig. 5 displays some breast MRI image samples
[18].
MRI, on the other hand, offers bright, sharp images
that provide an improved contrast between various
types of soft tissues, whereas mammogram images
show the contrast between soft tissue and bone
density. Because of this, MRI is employed in breast
screening to examine the minute intricacies within
breast tissues. Although this is important
knowledge, the radiologist still needs to analyze the
supplied data [19]. MRI radiologists employ CAD
algorithms to analyze breast MRIs and to lessen the
incidence of false-positive diagnoses [6].
Fig. 5: Breast MRI picture examples [18]
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2023.11.1
Ali Al-Faris
E-ISSN: 2415-1521
2
Volume 11, 2023
3 Breast MRI with CAD
To aid MRI radiologists in enhancing the accuracy
of breast MRIs, detecting tumor masses, and
lowering the incidence of false-positive detection,
CAD systems are utilized in conjunction with image
processing algorithms [6].
To find malignancies inside bodily organs, CAD
algorithms are created [20], [21]. For the various
modalities of medical pictures, including X-Ray and
Ultrasound, a variety of segmentation and
classification approaches are used. Wavelets,
fractals, statistical techniques, and vision-based
techniques have all been presented recently for
breast mass identification [22], [23], [24].
Additionally, approaches based on artificial
intelligence, such as Fuzzy Logic and Artificial
Neural Networks (ANN), have been developed for
classification [22], [25], [26].
Studies [10], [20], [27], [28] have established the
benefits of adopting CAD systems for breast tumor
identification in screening technologies. Fast
detection, accuracy, and helping radiologists locate
dense breasts that could be missed are some of the
benefits of CAD. However, to overcome the
drawbacks of present systems, CAD systems still
require upgrades.
The most frequent drawbacks of breast CAD
systems include the production of false-positive
results in many breast images, the failure to detect
tiny tumors, and the requirement for human user
engagement [10]. In contrast to another human
anatomy, there haven't been many studies on breast
MRI CAD systems
4 Breast MRI Tumour Segmentation
Methods
Supervised and unsupervised techniques are the two
primary subcategories of image segmentation
systems. Similar classification algorithms are used
in breast MRI tumor segmentation systems.
Additionally, some researchers have suggested
mixed systems or semi-supervised methods, which
will be discussed later in this section.
4.1 Supervised Methods
The analyst knows in advance the numerical
features, such as mean and variance, of the classes
in the image and uses them in the training stage
when using the supervised approach [29]. Learning
the individual items to be detected is done during
the training phase. The system must then be ready to
recognize and categorize new input images based on
the occurrence or nonappearance of comparable
items. Examples with and without the item are
covered in the training step [30]. Popular supervised
algorithms include the Bayesian Method, Support
Vector Machine, and K-Nearest Neighbours (KNN).
Support vector machine classifier-based supervised
technique for breast MRI cancer segmentation was
proposed by Jianhua et al. [31]. The chest and out-
of-body portions are first segregated in this
approach, leaving only the breast region for
subsequent processing. The texture features are
extracted for each pixel. To extract frequency
features, the wavelet transform is used. In the
training stage, a committee of Support Vector
Machines is created as the classifier after a
progressive feature selection is carried out to pick
useful features. To classify new data at the pixel
level, this classifier is applied. Various picture
protocols can be addressed using this technique. It
also lowers the number of features that are chosen.
To get the desired results, it must, however, be
trained on at least ten different cases.
Rabiei [32] uses the K-Nearest Neighbours (KNN)
classifier. They employed contextual data based on
the temporal kinetic signal and the geometry of the
items of interest in their work. The method is
illustrated by utilizing machine learning to divide
breast diseases into four categories using a KNN
classifier. In complex backdrops, the system
achieved high tumor segmentation results. The
fundamental drawback of this approach is that the
user must manually detect a binary window to begin
the initial segmentation, pick the breast region, and
ignore the remaining parts of the images that are
related to the tissues of the chest and heart.
Another supervised technique was put forth by Wu
et al. [33] and is based on the Bayesian method and
Markov random field model. With this method, the
characteristics of the breast MRI images were
analyzed and classified as tumor or non-tumor
regions. The Iterative Conditional Mode (ICM)
approach is used to estimate class membership.
Modeling the prior distribution of the class's
membership as a multi-level logistic model using a
Markov Random Field assumes that the class's
composition depends only on its immediate
neighbors. It is assumed that the likelihood
distribution is gaussian. This method could be
successfully used for real-time segmentation in
healthcare facilities. However, each Gaussian
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2023.11.1
Ali Al-Faris
E-ISSN: 2415-1521
3
Volume 11, 2023
distribution's parameters are explicitly chosen as an
approximation of its class representative.
4.2 Unsupervised Methods
Unsupervised segmentation is the process of
dividing an image into a collection of sections that
are distinct and constant in terms of certain
properties, such as intensity level, size, or texture
[30], [34], [35]. The unsupervised segmentation
family includes clustering, region-based approaches,
thresholding, and contour methods.
Compared to supervised approaches, unsupervised
methods have several advantages. With supervised
approaches, the segmentation must begin with the
analyst determining the features of the images in the
dataset beforehand. Contrarily, unsupervised
algorithms automatically identify unique classes,
significantly reducing the analyst's workload.
Additionally, for the supervised approaches, some
object attributes might not be known beforehand.
Unsupervised algorithms, however, automatically
identify these features in the image [35], [36].
A fuzzy c-means (FCM) clustering-based technique
for the segmentation of breast tumors in MRI
images was presented by Chen et al. [37]. The
suggested tumor segmentation approach requires
input from a person to choose the ROI, then picture
enhancement within the chosen ROI. The increased
ROI is then classified using FCM. By implementing
thresholding to the tumor membership map,
connected component labeling, and hole filling to
the chosen object, the tumor is finally segmented.
The technique can segment breast MRI tumors in a
precise, effective, and reliable manner. This
method's fundamental flaw is that it requires manual
input to identify the ROI as a rectangular shape
before segmentation can begin.
Cui et al. [38] suggest using a marker-controlled
watershed technique to separate malignant tumors
from breast MRI images. The semi-automatically
method begins by manually defining the ROI
ellipse. Then, using Gaussian mixture modeling, the
internal and exterior markers for the tumor's
watershed segmentation are found. The results
demonstrate good segmentation outcomes that are
consistent with the radiologist's manual tumor
volume description. The use of a mouse to line an
ROI in the shape of an ellipse on a chosen area that
contains the suspected tumor is the main weakness
of this method.
Militello et al. [39] investigated and compared four
unsupervised segmentation algorithms. These
include k-means, fuzzy c-means, spatial fuzzy c-
means, split-and-merge combined with region
growing (SMRG), and (sFCM). The observed
experimental results support the use of unsupervised
pattern recognition methods for segmenting medical
images using area- and distance-based metrics. In
particular, clustering-based segmentation
approaches outperformed the SMRG. Therefore, for
medical pictures that are characterized by
uncertainty/variability (sometimes connected to
noise), crisp segmentation techniques—such as k-
means and SMRG—are not well-suited, producing
erroneous borders and poorly defined details. Fuzzy
modeling, which has inherent flexibility, was used
in both FCM and sFCM clustering techniques to
greatly improve performance. Breast images are
characterized by essential variability, which
underperforms in complex cases and may cause
unsupervised methods to fail to identify borders or
anatomical details. Additionally, noise tampers with
digital photos, changing some aspects of the original
image. Dealing with noisy or low-contrast images is
typical. Lesion segmentation in these kinds of
images is not a simple task that can be done without
user input. This is because it is important to get
accurate results.
4.3 Semi-Supervised Methods
The area of machine learning known as semi-
supervised learning is focused on employing both
labeled and unlabeled data to carry out certain
learning tasks. It allows using the substantial
amounts of unlabeled data accessible in several use
cases in blending with smaller sets of labeled data. It
is conceptually located between supervised and
unsupervised learning.
Semi-supervised learning is frequently utilized to
decrease time-consuming and expensive manual
pixel-level annotation. Consistency regularisation
places restrictions on the consistency of predictions
made using perturbations to inputs, features, and
networks [40].
Oh et al. [40] proposed a semi-supervised breast
MRI segmentation approach that can be trained with
small amounts of annotation. A time difference map
is also proposed to incorporate the distinct time-
varying enhancement pattern of the tumor. In their
work, they also presented a novel loss function that
efficiently distinguishes breast tumors from those
without tumors based on triple loss. This loss
reduces the potential for false positives. The
proposed method produces better segmentation
results with fewer annotations, particularly for
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2023.11.1
Ali Al-Faris
E-ISSN: 2415-1521
4
Volume 11, 2023
boundary-based metrics relevant to spatially
continuous breast tumors.
To obtain a high performance, Azmi et al. [41] offer
a semi-supervised classification method to segment
breast tumors in MRI based on texture analysis. The
Improved Self-Training (IMPST) classifier is
trained solely with a labeled image in the first stage
of this two-stage procedure. The classifier is then
retrained to achieve high accuracy using
nondeterministic unlabeled data that is obtained in
the subsequent stage using a straightforward
thresholding method. The drawback of this method
appears in the requirement for the user to create a
small window to identify the cancer ROI region,
even though the accuracy and precision of
segmented images have increased based on reported
results.
In the work by Azmi et al., the supervised,
unsupervised, and semi-supervised approaches are
examined [41]. The supervised segmentation
methods, such as K-Nearest Neighbors (KNN),
Support Vector Machine (SVM), and Bayesian, as
well as the semi-supervised methods, such as self-
training and improved self-training (IMPST), lead to
high accuracy, according to their comparison study
on the MRI Breast RIDER dataset [18]. But prior
knowledge is necessary. As a result, the procedure
becomes challenging, costly, and time-consuming.
Contrarily, unsupervised techniques like fuzzy C-
means (FCM) do not require prior information, yet
they perform poorly [41].
5 Discussion
The methods under the supervised category use the
training phase to learn the system before segmenting
the tumors. This phase requires previous knowledge.
Among the advantages of using supervised methods
are that they can be applied to several image
protocols. Also fixable in terms of choosing features
ranging from few to many, with the choice based on
the needs of the system and the type of images used.
Furthermore, these methods produced excellent
tumor segmentation results in complex
systems. However, the drawbacks of using
supervised methods consist of the requirement of
training in at least ten different situations to produce
the desired results. In addition, in most of the
supervised applications, a window must be
manually created to choose the breast area while
ignoring the other areas of the image that contain
heart and chest tissue. Also, the parameters should
be calculated carefully and chosen so that they are
typical of the class.
In the unsupervised category, methods do not
require a training phase; they divide the images into
a collection of classes that are distinct and constant
in terms of certain properties, such as intensity level,
size, or texture. Unsupervised methods can segment
breast MRI tumors in a precise, effective, and
reliable manner. Also, results demonstrate good
segmentation outcomes that are consistent with the
radiologist's manual tumor volume description. On
the other hand, the drawbacks of unsupervised
approaches are that the segmentation procedure
cannot begin unless the user recognizes ROI areas.
The methods underperform in complex cases and
may cause unsupervised methods to fail to identify
borders or anatomical details. Segmenting noisy
images or low-contrast images is challenging, and it
might need user input.
The methods under the semi-supervised category
employ both labeled and unlabeled data to carry out
certain learning tasks. They are a hybrid of
supervised and unsupervised learning methods.
Based on recorded findings, the accuracy and
precision of segmented images have improved.
Most of their systems require user input to either
create windows to identify the tumor regions and/or
select parameters to start the process.
The three categories' descriptions, benefits, and
drawbacks are summarised in Table 1.
6 Conclusion
This paper reviewed earlier research on MRI breast
tumor segmentation systems along with associated
image processing methods and algorithms. The
study included several topics, including the history
of breast cancer, breast screening, CAD systems,
and methods for breast MRI tumor segmentation.
The various breast screening methods, such as
mammography, ultrasonography, and MRI, are
discussed. Previous research revealed that CAD
algorithms are crucial for assisting radiologists in
reading images and lowering the incidence of false-
positive diagnoses.
Approaches for segmenting breast MRI tumors have
been discovered; they are divided into supervised,
unsupervised, and semi-supervised approaches.
High accuracy is achieved during supervised
segmentation. However, due to the need for prior
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2023.11.1
Ali Al-Faris
E-ISSN: 2415-1521
5
Volume 11, 2023
knowledge, the procedure becomes challenging,
expensive, and time-consuming. Unsupervised
methods, on the other hand, do not require prior
information, although they do less well than other
approaches. The exclusion of other parts of the
breast is a crucial pre-process in tumor segmentation
systems. This procedure is crucial because, in the
majority of MRI breast cases, the feature levels
between the tumor regions and other regions are
comparable. Several methods have been developed
to exclude these unwanted regions from breast
images. Even though these methods typically
succeeded in their aim of exclusion. The approaches
suffered from the fact of creating them for a specific
type of image solely or the requirement for user
input are still their primary drawbacks. It is advised
to look at the potential for developing new CAD
systems using a combination of supervised and
unsupervised techniques to get highly accurate
segmentation results without the requirement for
prior knowledge. Fully automatic systems that do
not require user inputs might also be regarded as
potential study topics in the future.
Table 1. Summary of MRI breast tumors
segmentation approaches
Method
Description
Benefits
Supervis
ed
methods
[31],
[32],
[33]
Methods
that use the
training
phase to
learn the
system
before
segmenting
the tumors.
● They can
be applied
to several
image
protocols.
● Fixable in
terms of
selecting
features
ranging
from few
to many.
● In
complex
systems,
they
achieved
high tumor
segmentati
on results.
carefully
computed
and chosen
as being
typical of
the class.
Unsuper
vised
methods
[37,
[38],
[39]
Methods do
not require a
training
phase, and
they divide
the images
into a
collection of
sections that
are distinct
and constant
in terms of
certain
properties,
such as
intensity
level, size,
or texture.
● The ability
to segment
breast MRI
tumors in a
precise,
effective,
and
reliable
manner.
● Results
demonstrat
e good
segmentati
on
outcomes
that are
consistent
with the
radiologist'
s manual
tumor
volume
description
.
● The
segmentatio
n procedure
cannot
begin unless
the user
recognizes
ROI areas.
● The
methods
underperfor
m in
complex
cases and
may cause
unsupervise
d methods
to fail to
identify
borders or
anatomical
details.
● Segmenting
noisy
images or
low-contrast
images is
challenging,
and it might
need user
input.
Semi-
supervis
ed
methods
[40],
[41]
The
methods
employ both
labeled and
unlabeled
data to carry
out certain
learning
tasks. They
are a hybrid
of
supervised
and
unsupervise
d learning
methods.
Based on
recorded
findings, the
accuracy and
precision of
segmented
images have
improved.
Most of their
systems
require user
input to either
create
windows to
identify the
tumor regions
and/or select
parameters to
start the
process.
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2023.11.1
Ali Al-Faris
E-ISSN: 2415-1521
6
Volume 11, 2023
References:
[1] B. Stewart and C. P. Wild, "World Cancer
Report," in "International Agency for
Research on Cancer," 2018.
[2] Z. Sultana, M. A. R. Khan, and N. Jahan,
"Early Breast Cancer Detection Utilizing
Artificial Neural Network," WSEAS
Transactions on Biology and Biomedicine,
vol. 18, pp. 32-42, 2021.
[3] C. E. DeSantis, S. A. Fedewa, A. G. Sauer, J.
L. Kramer, R. A. Smith, and A. Jemal,
"Breast cancer statistics 2015," Cancer
Journal for Clinicians, vol. 66, no. 1, pp. 31-
42, 2016.
[4] J. Ferlay et al., "Estimating the global cancer
incidence and mortality in 2018:
GLOBOCAN sources and methods," Int J
Cancer, vol. 144, no. 8, pp. 1941-1953, Apr
15, 2019, doi: 10.1002/ijc.31937.
[5] F. Bray, J. Ferlay, I. Soerjomataram, R. L.
Siegel, L. A. Torre, and A. Jemal, "Global
cancer statistics 2018: GLOBOCAN estimates
of incidence and mortality worldwide for 36
cancers in 185 countries," CA Cancer J Clin,
vol. 68, no. 6, pp. 394-424, Nov 2018, doi:
10.3322/caac.21492.
[6] C. D. Lehman, S. Peacock, W. B. DeMartini,
and X. Chen, "A New Automated Software
System to Evaluate Breast MR Examinations:
Improved Specificity Without Decreased
Sensitivity," Am. J. Roentgenol., vol. 187, no.
1, pp. 51-56, July 1, 2006 2006, doi:
10.2214/ajr.05.0269.
[7] R. Hussain, S. Arif, M. A. Sikander, and A. R.
Memon, "Fuzzy Clustering Based Malignant
Areas Detection In Noisy Breast Magnetic
Resonant (MR) Images," International
Journal of Academic Research, vol. 3, no. 2,
March, 2011 2011.
[8] R. Croshaw, H. Shapiro-Wright, E. Svensson,
K. Erb, and T. Julian, "Accuracy of Clinical
Examination, Digital Mammogram,
Ultrasound, and MRI in Determining
Postneoadjuvant Pathologic Tumor Response
in Operable Breast Cancer Patients," Annals
of Surgical Oncology, vol. 18, no. 11, pp.
3160-3163, 2011.
[9] R. Bronstein, D. Reyes, and T. Kaufman,
"Breast Imaging," in Clinician's Guide to
Diagnostic Imaging, W. R. Reinus Ed. New
York, NY: Springer New York, 2014, pp. 83-
95.
[10] G. I. Andreea, R. Pegza, L. C. Lascu, S.
Bondari, Z. Stoica, and A. Bondari, "The Role
of Imaging Techniques in Diagnosis of Breast
Cancer," 2012.
[11] J. Suckling et al. Mammographic Image
Analysis Society (MIAS) database,
[12] S. P. Foroozandehasl, M. W. Marashdeh, M.
S. Jaafar, and L. Ghasemi, "Assessment of an
Anthropomorphic Ultrasound Breast
Phantom," WSEAS Transactions on
International Journal of Electrical
Engineering and Computer Science, vol. 1,
pp. 56-60, 2019.
[13] C. H. Lee et al., "Breast cancer screening with
imaging: recommendations from the Society
of Breast Imaging and the ACR on the use of
mammography, breast MRI, breast
ultrasound, and other technologies for the
detection of clinically occult breast cancer,"
(in eng), J Am Coll Radiol, vol. 7, no. 1, pp.
18-27, Jan 2010, doi:
10.1016/j.jacr.2009.09.022.
[14] S. Li and P. H. Wu, "Magnetic resonance
image-guided versus ultrasound-guided high-
intensity focused ultrasound in the treatment
of breast cancer," (in eng), Chin J Cancer,
vol. 32, no. 8, pp. 441-52, Aug 2013, doi:
10.5732/cjc.012.10104.
[15] J. G. Elmore, K. Armstrong, C. D. Lehman,
and S. W. Fletcher, "Screening for Breast
cancer," (in 0098-7484), JAMA, the journal of
the American Medical Association, vol. 293,
no. 10, p. 12, 2005.
[16] S. Prapavesis et al., "Breast ultrasound and
ultrasound-guided interventional procedures
in the breast: A CD-ROM multimedia
teaching file," 2003: European Congress of
Radiology-ECR 2003.
[17] J. B. Weaver, Y. Xu, D. M. Healy, and L. D.
Cromwell, "Filtering Noise From Images
With Wavelet Transforms," Magnetic
Resonance in Medicine, vol. 21, no. 2, pp.
288-295, 1991, doi:
10.1002/mrm.1910210213.
[18] U.S. National Cancer Institute. Reference
Image Database to Evaluate Therapy
Response (RIDER) MRI Breast [Online]
Available: from National Biomedical Imaging
Archive (NBIA) Web site
https://imaging.nci.nih.gov/ncia,
[19] I. Gardiner. (2010, February 4, 2010) CAD
Improves Breast MRI Workflow: Increasing
Throughput While Maintaining Accuracy in
Breast MRI Reads Requires Powerful
Workflow Tools. Imaging Technology News.
Available:
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2023.11.1
Ali Al-Faris
E-ISSN: 2415-1521
7
Volume 11, 2023
http://www.itnonline.com/article/cad-
improves-breast-mri-workflow
[20] L. Li, R. A. Clark, and J. A. Thomas,
"Computer-Aided Diagnosis of Masses with
Full-Field Digital Mammography," Academic
Radiology, vol. 9, no. 1, pp. 4-12, 2002.
[21] B. Verma and J. Zakos, "A Computer-Aided
Diagnosis System For Digital Mammograms
Based on Fuzzy-Neural and Feature
Extraction Techniques," IEEE Transactions
on Information Technology in Biomedicine,
vol. 5, no. 1, pp. 46-54, 2001.
[22] D. R. Chen, R. F. Chang, W. J. Kuo, M. C.
Chen, and Y. L. Huang, "Diagnosis of breast
tumors with sonographic texture analysis
using wavelet transform and neural
networks," (in eng), Ultrasound Med Biol,
vol. 28, no. 10, pp. 1301-10, Oct 2002, doi:
10.1016/s0301-5629(02)00620-8.
[23] D. R. Chen et al., "Classification of breast
ultrasound images using fractal feature," (in
eng), Clin Imaging, vol. 29, no. 4, pp. 235-45,
Jul-Aug 2005, doi:
10.1016/j.clinimag.2004.11.024.
[24] W.-J. Kuo, R.-F. Chang, D.-R. Chen, and C.
C. Lee, "Data mining with decision trees for
diagnosis of breast tumor in medical
ultrasonic images," Breast Cancer Research
and Treatment, vol. 66, no. 1, pp. 51-57,
2001/03/01 2001, doi:
10.1023/A:1010676701382.
[25] R. Mousa, Q. Munib, and A. Moussa, "Breast
cancer diagnosis system based on wavelet
analysis and fuzzy-neural," Expert Systems
with Applications, vol. 28, no. 4, pp. 713-723,
2005/05/01/ 2005, doi:
https://doi.org/10.1016/j.eswa.2004.12.028.
[26] H. D. Cheng and M. Cui, "Mass lesion
detection with a fuzzy neural network,"
Pattern Recognition, vol. 37, no. 6, pp. 1189-
1200, 2004/06/01/ 2004, doi:
https://doi.org/10.1016/j.patcog.2003.11.002.
[27] I. Leichter, S. Buchbinder, P. Bamberger, B.
Novak, S. Fields, and R. Lederman,
"Quantitative characterization of mass lesions
on digitized mammograms for computer-
assisted diagnosis," (in eng), Invest Radiol,
vol. 35, no. 6, pp. 366-72, Jun 2000, doi:
10.1097/00004424-200006000-00005.
[28] S. Arora, J. Acharya, A. Verma, and P. K.
Panigrahi, "Multilevel thresholding for image
segmentation through a fast statistical
recursive algorithm," Pattern Recognition
Letters, vol. 29, no. 2, pp. 119-125,
2008/01/15/ 2008, doi:
https://doi.org/10.1016/j.patrec.2007.09.005.
[29] C. K. Narayanappa, G. R. Poornima, and B.
V. Hiremath, "Estimation of the Ambit of
Breast Cancer with a Modified Resnet
Analysis Using Machine Learning Approach,"
WSEAS Transactions on Biology and
Biomedicine, vol. 18, pp. 183-190, 2021.
[30] A. Oliver et al., "A review of automatic mass
detection and segmentation in mammographic
images," Medical Image Analysis, vol. 14, no.
2, pp. 87-110, 2010/04/01/ 2010, doi:
https://doi.org/10.1016/j.media.2009.12.005.
[31] J. Yao, J. Chen, and C. Chow, "Breast Tumor
Analysis in Dynamic Contrast Enhanced MRI
Using Texture Features and Wavelet
Transform," Selected Topics in Signal
Processing, IEEE Journal of, vol. 3, no. 1, pp.
94-100, 2009, doi:
10.1109/jstsp.2008.2011110.
[32] H. Rabiei, A. Mahloojifar, and M. E. Farmer,
"PROVIDING CONTEXT FOR TUMOR
RECOGNITION USING THE WRAPPER
FRAMEWORK," in Biomedical Imaging:
From Nano to Macro, 2007. ISBI 2007. 4th
IEEE International Symposium on, 12-15
April 2007 2007, pp. 1252-1255, doi:
10.1109/isbi.2007.357086.
[33] Q. Wu, M. Salganicoff, A. Krishnan, D. S.
Fussell, and M. K. Markey, "Interactive lesion
segmentation on dynamic contrast enhanced
breast MRI using a Markov model," 2006,
vol. 6144, pp. 61444M-61444M-8. [Online].
Available:
http://dx.doi.org/10.1117/12.654308
[34] K. S. Fu and J. K. Mui, "A survey on image
segmentation," Pattern Recognition, vol. 13,
no. 1, pp. 3-16, 1981/01/01/ 1981, doi:
https://doi.org/10.1016/0031-3203(81)90028-
5.
[35] M. G. H. Omran, "A PSO-based Clustering
Algorithm with Application to Unsupervised
Image Classification," University of Pretoria
etd, 2005.
[36] R. W. Bogue, "Machine Vision Theory,
Algorithms, Practicalities," Assembly
Automation, vol. 25, no. 3, 2005, doi:
10.1108/aa.2005.03325cae.001.
[37] W. Chen, M. L. Giger, and U. Bick, "A Fuzzy
C-Means (FCM)-Based Approach for
Computerized Segmentation of Breast
Lesions in Dynamic Contrast-Enhanced MR
Images1," Academic radiology, vol. 13, no. 1,
pp. 63-72, 2006.
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2023.11.1
Ali Al-Faris
E-ISSN: 2415-1521
8
Volume 11, 2023
[38] Y. Cui et al., "Malignant lesion segmentation
in contrast-enhanced breast MR images based
on the marker-controlled watershed," Medical
Physics, vol. 36, no. 10, pp. 4359-4369, 2009,
doi: doi:http://dx.doi.org/10.1118/1.3213514.
[39] C. Militello et al., "On Unsupervised Methods
for Medical Image Segmentation:
Investigating Classic Approaches in Breast
Cancer DCE-MRI," vol. 12, no. 1, p. 162,
2022.
[40] Y.-T. Oh, E. S. Ko, and H. Park, "Semi-
supervised Breast Lesion Segmentation using
Local Cross Triplet Loss for Ultrafast
Dynamic Contrast-Enhanced MRI," in Asian
Conference on Computer Vision (ACCV),
2022, pp. 2713-2728.
[41] R. Azmi, R. Anbiaee, N. Norozi, L. Salehi,
and A. Amirzadi, "IMPST: A New Interactive
Self-Training Approach to Segmentation
Suspicious Lesions in Breast MRI," J Med
Signals Sens, vol. 1, no. 2, pp. 138-48, 2011.
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
Ali Al-Faris is the sole writer of this paper. He
reviewed the studies, compared them, and discussed
and concluded the work.
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
WSEAS TRANSACTIONS on COMPUTER RESEARCH
DOI: 10.37394/232018.2023.11.1
Ali Al-Faris
E-ISSN: 2415-1521
9
Volume 11, 2023
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 author has no conflict of interest to declare that
is relevant to the content of this article.