Professional Documents
Culture Documents
©iStockphoto.com/bestdesigns
An overview of
mammogram analysis
Huda Al-Ghaib, Reza Adhami, and Melanie Scott
A
mammogram is an X-ray
image of the human breast.
It is used for the detection
and diagnosis of changes
in breast tissues. Asymp-
tomatic women may be encouraged to
undergo screening mammography on
a regular basis after reaching a cer-
tain age. In most cases, mammo-
grams reveal information that could
yield the detection of suspicious
lesions by breast radiologists (BRs).
Hopefully, the detection is achieved
before the lesions have advanced to a
late stage of cancer, which makes it
difficult to cure.
Lesions can be characterized as
masses, calcifications, or architec-
tural distortions. A mass is an ac-
cumulation of cells in one specific
location that can be benign or ma-
lignant. Calcifications are tiny bright
deposits of calcified milk. Calcifica-
tion is characterized as benign or
image licensed by ingram publishng
malignant based on its shape, size,
contour, and its count in one loca-
tion. Architectural distortion appears Computer-aided diagnosis (CAD) has within the lobes to produce milk and
when lesions destroy the normal ar- been developed to assist radiologists transfer it to the nipple through the
chitecture of the breast with no pres- in detecting breast lesions. ducts. Figure 1 shows the anatomy
ence of a mass. Since a mammogram of the breast with a magnified TDLU.
is a two-dimensional (2-D) projection Breast anatomy The breast lies on the superfi-
of a complex three-dimensional (3-D) The breasts of female mammals are cial fascia that covers the anterior
object, this makes it challenging to active organs that are responsible and lateral parts of the pectoralis
identify all ill-defined lesions using for nourishing the newborns with major muscle. The space between
mammography. Many factors, such milk. The female hormones become the breast and the pectoralis major
as breast density, overlapping tis- active in the breasts during puberty. muscle is the retromammary space,
sues, and the lesion’s size, affect the These hor mones stimulate the which consists of adipose tissue and
accuracy of mammogram analysis. breast glands to grow into a spheri- connective fascia. The breast tissues
cal shape. When a gestation period extend vertically from the clavicle
Digital Object Identifier 10.1109/MPOT.2015.2396533
ends, the female hormones boost the bone located on the second or third
Date of publication: 14 November 2016 terminal ductal lobule units (TDLUs) rib to the abdominal wall located on
0278-6648/16©2016IEEE IEEE Potentials November/December 2016 ■ 21
the sixth or seventh rib. Horizontally,
the breast tissues extend from the
Intralobular axilla and the side muscles to the
Terminal Duct sternum bone.
The external anatomy of the breast
Segmental
Duct Ductules consists of the skin, nipple, areola,
tubercles, and several glands. The
breast skin contains sweat glands, oil
Collecting
Duct glands, and hair follicles. The nipple
is located at the center of the breast
or slightly below the center. The nip-
ple’s texture is soft to firm, with a flat,
round conical, inverted, or cylindrical
Lactiferous shape. The areola is the smooth, dark
Sinus
pink, circular area that surrounds
the nipple. The areola consists of sev-
Extralobular Lobule
Terminal Duct eral small bumps that are known as
the tubercles of Montgomery and are
responsible for lubricating the nipple
Terminal Duct
during the lactation period.
Lobular Unit
TDLU A network of lobes that is re-
sponsible for producing milk resides
Fig 1 The anatomy of the breast. (Image used with permission from L. Tabár, 2014.) within the connective and fatty tis-
sues. Each lobe is of a circular to
pyramid shape with an individual
duct that has its own opening on the
nipple surface. Figure 2 illustrates
a 3-D histology of a main duct. The
branching ducts inside each lobe
subdivide it into a number of lobules.
Each lobe may contain ~10–100 lob-
ules. Each lobule consists of several
acini (ductules) and intralobular ter-
minal ducts (ITDs). The ductules are
located at the far end of the lobule
and are the basic units responsible
for milk production. The extralobular
terminal ducts (ETDs) and the lob-
Fig2 The 3-D histology of a main duct. (Image used with permission from L. Tabár, ule form the TDLU. The connective
2014.) tissues hold together and support
the glands inside the breast. Fatty
tissue is part of the connective tis-
sues that gives the breast its smooth
contours. A network of nerves, blood
vessels, and lymph vessels reside
within the connective tissues. Fig
ure 3 displays lobules imaged using
(a) an X-ray mammography image,
(b) a galactogram, and (c) subgross
histology techniques, respectively.
Breast cancer
Breast cancer is a life-threatening
disease of unknown cause that
(a) (b) (c) affects women around the globe.
According to the American Cancer
Fig3 (a) A mammogram, (b) a galactogram, and (c) a subgross histology with numer- Society, breast cancer is the second
ous lobules outlined by fat. (Images used with permission from L. Tabár, 2014.) leading cause of ca ncer deat h
IS
ar
O
2%
r
la
l
N
bu
s
D
bu
ou
ry
al
Lo
ry
Tu
in
la
e
lla
uc
ul
in
D
ed
pi
er
M
th
Ap
IEEE Potentials November/December 2016 ■ 23
They account for 47.7% and 35.4% of tions are due to invasive cancer in the malignant deadly carcinomas may
the invasive cancers, respectively. breast tissue. not be possible.
Massive lesions are mostly con- According to the Breast Imag- In the screening mammogra-
sidered as the pathological evalua- ing Reporting and Data System (BI- phy procedure, a low-dose X-ray is
tion of calcifications. Calcifications RADS), an architectural distortion passed through the breast tissues.
are tiny particles with different pat- occurs when the normal architec- Connective and fatty tissues, mass-
terns. Calcifications are more likely ture of the breast is distorted with es, and calcifications have different
to occur in young patients, with no definite mass visible. BI-RADS is attenuation factors for the X-ray and
a probability of 63% for patients published and trademarked by the appear with different brightness
younger than 50 years old. Deter- American College of Radiology (ACR) levels on the mammogram. Masses
mining the malignancy of calcifica- based on the collaborative effort of and calcifications usually appear
tions is a challenging task because many health groups. Architectural brighter compared with other nor-
of their fuzzy appearance and dif- distortions include focal retraction mal tissues. Figure 7 shows the
ficulty of distinguishing them from or distortion at the edge of the paren- mammographic representation for a)
their surroundings, especially in chyma and spiculations radiating a mass, b) calcifications, and c) an
dense breasts. The location, distri- from a point. In addition, sometimes architectural distortion, respective-
bution, density, and the shape of the architectural distortions appear ly. There are two standard projec-
calcifications are commonly used as the initial stage of an obvious tions for screening mammography:
to classify them as benign or ma- mass shadow. Other signs of archi- craniocaudal (CC) and mediolateral
lignant. Linear and segmental cal- tectural distortions include asym- oblique (MLO) views. The CC view
cifications are of malignant nature metrical thickening and straight- shows the medial part of the breast,
while diffuse and regional are of be- ening of fibrous connective tissues, and the nipple must be in the profile.
nign nature. asymmetric density in one breast, The external lateral portion may also
Calcifications can be an early sign or fibrotic changes that produce a be included in the CC view, while the
of breast cancer, as 30–50% of the “purse-string” appearance. retromammary space is only shown
early detected breast cancers were in 20% of cases. The pectoralis ma-
through the appearance of clusters Screening mammography jor muscle appears in only 30–40%
of fine granular microcalcifications. The main aim of using screening of the cases. Most of the breast tis-
Calcifications formed in the ducts mammography is to detect malig- sues are captured within the MLO
through a bilateral process known as nant carcinomas while they are still view. The retromammary space, as
secretory disease and results in se- localized in the breast. Many health well as the pectoralis major muscle,
cretory-type calcifications. There are organizations encourage asymptom- must appear in the MLO to indicate
two mammographic presentations for atic women of a certain age to correct positioning of the breast dur-
the secretory type: intraductal and undergo screening mammography ing the MLO procedure.
periductal forms. Intraductal calcifi- on a regular basis. Despite its popu- An accurate mammogram analy-
cations have a long needle shape that larity, screening mammography is a sis depends on the quality of the im-
reflects the duct shape, while peri- controversial issue. Some experts ages and the radiologist’s skills and
ductal have hollow ring-like or tube- believe that the overdiagnosis and experience. Radiologists carefully ex
like shapes. false positive rates, associated with amine the mammogram views to
Another type of calcification formed using screening mammography search for suspicious lesions, mass-
in the duct is the casting type, which surpass its benefits. Others argue es with ill-defined margins, breast
can be further subdivided into frag- that screening mammography is not asymmetry, and microcalcifications.
mented and dotted casting types. Cal- 100% effective, since it produces Standard definitions established by
cifications formed in ducts are most numerous false positive rates. Some BI-RADS are used to interpret the
likely to be malignant. Calcifications experts are proposing individualiz- mammogram findings. These defini-
can be formed in the terminal duc- ing screening mammography based tions help to maintain standardized
tal lobule units (TDLUs) as well, and on the risk factors. At this point, terminology that facilitates commu-
examples of these calcifications are women have to decide how often nication between the radiologist and
crushed stone-like and powdery, and they want to perform screening the physician.
cotton ball-like calcifications mammography and embrace the
Architectural distortion is the risk associated with their decisions. Screening mammography
third-most prevalent sign of nonpal- Breast cancer is a disease with shortcomings and benefits
pable breast cancer. Architectural ambiguous symptoms. In most cas- Screening mammography does not
distortions accounted for almost 20% es, it takes several years for the dis- provide a perfect diagnosis. Similar to
of the detected cancers along with ease to present physical symptoms, any other diagnostic tools, it includes
asymmetry, single dilated duct, and which may manifest during its late false negative, false positive, and
developing densities. It is reported stages. Without the use of screening overdiagnosis rates. False negative
that 79% of the architectural distor- mammography, detection of these detection can result in g iving a
Fig7 A mammographic representation of (a) a mass, (b) calcifications, and (c) an architectural distortion. (Used with permission from
L. Tabár, 2014.)
patient the assurance of a cancer- further attention. Unfortunately, been conducted in the area of mam-
free body while she may be develop- 10–30% of breast lesions are over- mogram analysis for that purpose.
ing a carcinoma. On the other hand, looked by radiologists during routine The main goal of developing such a
a false positive places the emotional screenings. A retrospective study system is to improve the accuracy
pain of having cancer on the patient, (including malignant cases only) and consistency of mammogram in-
as well as having her go through showed that in 48% of the cases, terpretation by radiologists and to
multiple tests and biopsy procedures minimal signs were visible on prior detect small gradual changes in the
while her body is, in fact, cancer-free. mammograms. Furthermore, 9% of breast tissue.
Screening mammography by itself is malignant cases were visible on The developed system is known
a stressful procedure that costs time screening mammograms obtained as CAD. In the medical community,
and money and produces anxiety. two years earlier. Also, the sensitivity CAD is a second observer for screen-
This procedure requires compressing in locating lesion in mammograms ing mammography because human
the breast, which may be uncomfort- for women with BRAC1 or BRAC2 decision-making and observation
able, and, of course, exposure to gene mutations is found to be low in mechanisms are considered more
X-rays. Therefore, as mentioned pre- the 16–40% range. reliable and trusted in comparison
viously, individualizing the procedure One of the reasons behind the to those made by machines. Ideally,
may be the best option. high false negative rates is the inter- CAD should improve the quality of
One advantage for screening mam pretation of a large number of mam- service provided to patients, and pa-
mography is its effectiveness in de- mograms by radiologists on a daily tients should not suffer from faults
tecting nonpalpable malignant le- basis for detecting a small number due to device failures. CAD is con-
sions in asymptomatic women. It has of cancers. One obvious solution is sidered one of the major research
been documented that the carcino- to double read each mammogram subjects in medical imaging and di-
mas found during screening mam- by two radiologists to confirm the agnostic radiology.
mography are usually detected dur- findings. Clearly this solution pro- The CADs have two stages: de-
ing the early stages of the disease. duces more accurate interpretation tection of suspected lesion locations
Patients with early detected cancers for mammograms, from 81.4% to and reduction of false-positive rates.
have higher survival rates and more 88%. Additionally, the double read- The first stage should provide high
successful treatment options in ing of screening mammography has sensitivity in detecting carcinomas,
comparison to patients with cancer increased the detection rate of breast while the second stage should have
detected at later stages. cancer by 4–15%. high specificity to reduce the false-
This practice is not always imple- positive rate. The CA D outcome
Computer-aided diagnosis mented in countries that lack a suf- is illustrated as a set of marks on
On any given day, radiologists exam- ficient number of BRs. An alternative the mammogram identifying suspi-
ine numerous mammograms. Biop- solution may be through developing cious lesions.
sy, magnetic resonance imaging an automated diagnostic system
(MRI), and exaggerated mammogram that uses a fast computational envi- The importance of CAD
views are tools used by radiologists ronment for providing a second opin- In 1998, the U.S. Food and Drug
to examine breast areas that need ion. For many years, research has Administration (FDA) approved CAD
IEEE Potentials November/December 2016 ■ 25
p-value for breast density, histopa-
Without CAD With CAD Without CAD With CAD thology, and tumor size categories
were 0.274, 0.922, and 0.138, re-
86 160
84.29% spectively. CAD’s false positive rate
81.10%
84 140 was 2.3 per case, where a case in-
82 111.6 109.8
120 cludes four images: the right and left
135.5
80 100 94.3
views of both MLO and CC.
77.95%
(%)
78 80
76 75.38% 60 CAD systems methodology
74 40 A general block diagram for algo-
72 20 rithms used in CAD is demonstrat-
ed in Fig.11. This block diagram
70 0
BR Group RR Group BR Group RR Group consists of three phases: training,
(a) (b) testing, and evaluation. The train-
ing phase includes two steps: pre-
FIG8 (a) Sensitivity and (b) reading time (in minutes) for BR and RR groups. (Statistics processing and feature extraction.
courtesy of Young, 2014.) In the preprocessing step, image fil-
tering and transformations are
applied to provide higher-quality
100 mammograms and to filter out
90.40%
88% excess noise. In the feature extrac-
90 Single Reader with CAD
tion step, features that best charac-
80 Double Reader No CAD
terize the lesions and normal breast
70 tissue are extracted. Feature extrac-
60 tion can be done manually or by
(%)
Tumor Size
21–30 mm 100%
11–20 mm 97%
1–10 mm 89%
DCIS 93%
Histopathology
ILC 100%
IDC 94%
Mammographics
Masses 92%
Calcifications 93%
Fatty 100%
0 20 40 60 80 100 120
(%)
FIG10 CAD detection rate based on different factors. (Statistics courtesy of The, 2009.)
IEEE Potentials November/December 2016 ■ 27
stages, a screening program has been Read more about it • N. Young, B. Joo, H. Sook, E.
launched around the globe. In this • L. Tabár. (2014, Oct.). Hands on Suk, J. Hee, C. Suk, and J. Jeong,
Breast Imaging Course [Online]. Avail- “Who could benefit the most from using
procedure, modalities such as mam-
able: http://w w w.mammographyed a computer aided detection system in
mography are used to acquire X-ray full-field digital mammography?” World
.com
images for the breast tissue and to J. Surg. Oncol., vol. 12, no. 1, pp. 1–20,
• L. Hartmann and C. Loprinzi,
differentiate between normal and The Mayo Clinic Breast Cancer Book.
2014.
abnormal breast patterns. Screening Intercourse, PA: Good Books, 2012.
mammography could be helpful to • The American Cancer Society. About the authors
those with a family history of having (2014, Oct.). Breast cancer statistics Huda Al-Ghaib (hudaalghaib@
breast cancer or with signs that could [Online]. Available: www.cancer.org fulbrightmail.org) earned her under-
be indicators of developing cancer. • L. Bruce, Multiresolution Signal graduate degree in computer engi-
Analysis with Applications to Mammo-
Nevertheless, screening mam- neering from the University of
graphic Lesion Classification: A Disser-
mography remains a controversial tation. Huntsville, AL: Univ. of Alabama T echnology in Baghdad, Iraq, in
issue; it lacks the consistency and Press, 1996. 2006. She worked in the Ministry of
suffers from factors including false- • H. S. Gallager and J. E. Martin, Higher Education and Scientific
negative, false-positive, and overdi- “The study of breast carcinoma by cor- Research from 2007 to 2009 and was
agnosis rates. Screening people at related mammopgrahy and subserial a recipient of a Fulbright Scholar in
whole organ sectioning: Early observa-
a higher risk of developing breast 2009 for which she earned her mas-
tions,” Cancer, vol. 23, no. 4, pp. 855–
cancer is more preferable than rec- 873, Apr. 1969. ter’s and Ph.D. degrees in electrical
ommending it for all women after a • O. Olsen and P. C. Gøtzsche, engineering from the University of Ala-
certain age. “Cochrane review on screening for bama in Huntsville in 2011 and 2015,
For every 1,000 screening cases, breast cancer with mammography,” respectively. She is currently an assis-
100 are called back for extra inves- Lancet, vol. 358, no. 9290, pp. 1340– tant professor at Utah Valley Universi-
1342, 2001.
tigations, and fewer than ten are of ty, College of Technology and Comput-
a malignant nature. Routine exami- • F. Nunes, H. Schiabel, and C. ing. She is a Member of the IEEE.
Goes, “Contrast enhancement in dense
nation of large numbers of mammo- breast images to aid clustered micro- Reza Adhami (adhamir@uah
grams by radiologists is a challenging calcifications detection,” J. Dig. Imag., .edu) earned his B.S.E., M.S.E, and
task. The advancement in technology vol. 20, no. 1, pp. 53–66, 2007. Ph.D. degrees in electrical engineer-
has urged scientists and engineers • S. Halkiotis, T. Botsis, and M. Ran- ing from the University of Alabama,
to develop an automated system that goussi, “Automatic detection of clus- Huntsville (UAH), in 1980, 1981,
tered microcalcifications in d igital
could assist radiologists to read a and 1985, respectively. He served as
mammograms using mathematical
large number of mammograms. The morphology and neural n etworks,” the Electrical and Computer Engi-
automated system is recognized as S ignal Process., vol. 87, no. 7, pp. neering Department chair at UAH
CAD and uses digital image process- 1559–1568, July 2007. from 1997 to 2010 and director of
ing algorithms to interpret mammo- • M. Gromet, “Comparison of com- the shared Ph.D. program between
gram information. puter-aided detection to double read- the University of Alabama in Bir-
ing of screening mammograms: Review
Many health facilities have found mingham and UAH. He is currently
of 231,221 mammograms,” Amer. J.
CAD to be helpful in reading screen- Roentgenol., vol. 190, no. 4, pp. 854– an emeritus professor of electrical
ing mammograms. However, CAD 859, 2008. and computer engineering at UAH.
lacks the consistency in interpreting • S. K. Yang, W. K. Moon, N. Cho, Melanie Scott (melaniehscott@
the overlapped challenging patterns J. S. Park, J. H. Cha, S. M. Kim, S. J. gmail.com) earned her undergradu-
of the breast. As a result, many nor- Kim, and J. G. Im, “Screening mam- ate degree (magna cum laude) from
mography-detected cancers: Sensitivity
mal cases have been called back for Vanderbilt University. She graduated
of a computer-aided detection system
extra unnecessary investigations. applied to full-field digital mammo- first in her class from the University
CAD is effective in identifying carci- grams,” Radiology, vol. 244, no. 1, pp. of Tennessee College of M edicine,
nomas associated with the presence 104–111, 2007. where she was a Dogett Fellow. She
of calcifications, masses of large sizes, • J. S. The, K. J. Schilling, J. W. completed her residency in diagnos-
and a mixture of masses and calcifi- Hoffmeister, E. Friedmann, R. McGinn- tic radiology at the University of
is, and R. G. Holcomb, “Detection of
cations. Nevertheless, many carci- Washington and her fellowship in
breast cancer with full-field digital
nomas are missed by CADs. Almost mammography and computer-aided breast imaging at the University of
all CADs neither incorporate algo- detection,” Amer. J. Roentgenol., vol. Alabama, Birmingham. She has
rithms to register temporal mammo- 192, no. 2, pp. 337–340, Feb. 2009. served as medical director for many
grams nor include algorithms to de- • J. M. Ko, M. J. Nicholas, J. B. breast centers. Her current practice
tect architectural distortions. Mendel, and P. J. Slanetz, “Prospec- draws upon past experiences in
tive assessment of computer aided
In conclusion, more research needs these centers to address individual
detection in interpretation of screen-
to be conducted for automated detec- ing mammography,” Amer. J. Roent- patient needs while providing high-
tion of breast architectural distortions genol., vol. 187, no. 6, pp. 1483–1491, quality care.
as well as mammogram registration. 2006.