A New Local Adaptive Mass Detection Algorithm in Mammograms

Ehsan Koozegar

, Mohsen Soryani

and Ines Domingues

Iran University of Science and Technology(IUST), Narmak, Tehran, Iran

INESC TEC (formerly INESC Porto) and Faculty of Engineering, University of Porto, Porto, Portugal

Keywords: Mamograms, Masses, Detection.

Abstract: Mammography is the most effective procedure for an early detection of breast abnormalities. Masses are a

type of abnormality which are very difficult to be visually detected on mammograms. In this paper an

efficient method for detection of masses in mammograms is introduced and tested. The algorithm is inspired

by binary search and was evaluated both on mini-MIAS and INBreast databases. Mini-MIAS results show

that our algorithm outperforms other competing methods. For INBreast database there are no other

published mass detection results for comparison, but we believe that our algorithm has good performance.

1 INTRODUCTION

Breast cancer is one of the most lethal diseases in

various parts of the world especially in western

countries. Several reports about the outbreak and

severity of breast cancer are published by different

organizations (Oliviera et al., 2011). According to

some reports, breast cancer is the second most

common disease after lung cancer (10.9% of cancer

incidence in both men and women) and the fifth

most common cause of cancer death (Oliveira et al.,

2009). The National Breast Cancer Foundation has

estimated that 200,000 people suffer from the

disease and 20,000 die every year. Furthermore,

according to American National Cancer Institute,

every three minutes one woman is diagnosed with a

cancerous case and every 13 minutes one woman is

killed by the disease (Oliveira et al., 2009).

Among various modalities, mammography is the

most popular method to detect different

abnormalities in breasts. During the last two

decades, many scientists have been attempting to

help radiologists in the detection and diagnosis of

these anomalies. It is however important to note that

Computer Aided Diagnosis (CAD) systems are

designed to assist radiologists only as a second

interpretation and never as a substitute.

Masses and microcalcifications (MCCs) are two

most frequent findings in mammograms. Detecting

masses is more difficult than detecting MCCs

because mass features can be ambiguous or similar

to breasts’ parenchyma. Masses are usually located

in the dense regions of the breast. Furthermore, they

have smoother boundaries than MCCs and more

various shapes as well. These factors make mass

detection a challenging problem both for humans

(radiologists) and machines (CAD systems). It has

been reported that most abnormalities missed by

radiologists are related to cancerous masses

(Malagelada, 2007). Most of the available

commercial CAD systems for detecting MCCs have

reached 100% of detection rate, but the detection

rate of masses is still below 90%.

Mass detection has a vital role in full CAD

systems and many studies have been made during

the last two decades. A good review was published

in (Oliver et al., 2010) that covers mass detection

algorithms until 2008. Some of the more recent

works are worth mentioning. Martins team (Martins

et al., 2009) presented a methodology for detecting

masses in digitized mammograms using the growing

neural gas algorithm for image segmentation and

Ripley’s K function to describe the texture of

segmented structures. Growing neural gas is an

incremental and non-supervised clustering algorithm

while Ripley’s K function is a second-order tool to

analyze completely mapped spatial-point process

data. Using the digital database for screening

mammography (DDSM), the methodology reached

an accuracy rate of 89.3%, with 0.93 false positive

(FP) and 0.02 false negative (FN) per image.

In 2010, Gao (Gao et al., 2010) proposed two

concentric layer criteria to detect different types of

suspicious regions. After mammograms are

133

Koozegar E., Soryani M. and Domingues I..

A New Local Adaptive Mass Detection Algorithm in Mammograms.

DOI: 10.5220/0004218201330137

In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing (BIOSIGNALS-2013), pages 133-137

ISBN: 978-989-8565-36-5

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

separated into multi-intensity layers by using

different intensity thresholds, the real mass regions

should contain some concentric layers on different

intensity layers. If one of the following criteria is

satisfied, the region is deemed as a suspicious

region: Multilayer Criterion (focal regions with

concentric layers ൒1 are considered as mass regions

and the confidence increases with the increase in the

number of layers) and Single-Layer Criterion (focal

regions without concentric layers in their adjacent

lower intensity layer are considered as mass regions

only if their morphological features satisfy stricter

threshold conditions and the additional contrast

condition at the same time). The combination was

evaluated on DDSM, resulting in a sensitivity of

99% in malignant, 88% in benign, and 95.3% in all

types of cases.

Mencattini and Salmeri (Mencattini and Salmeri,

2011) developed a suspicious mass detection

scheme where, after smoothing images with a

Gaussian filter, it evaluates them with a Gradient

and Hessian matrix. FP rejection is achieved by

comparing three features against a threshold:

condition number, mean eigenvalues intensity map,

and Area. For a sensitivity of 0.9, results on DDSM

images exhibit a False- Positive per Image equal to

0.6 for cancer cases and 0.2 for normal cases.

Sampaio’s team (Sampaio et al., 2011) proposed

to use a cellular neural network to segment regions

that might contain masses. Two templates were

used: Textural template (is able to separate the mass

candidates but inserts pixels that do not belong to the

candidates) and Blur template (does not insert extra

pixels, but might remove several pixels from the

candidates). Images resulting from the use of these

templates are aggregated using the binary OR

operator. Sensitivity of 80% with rates of 0.84 FPs

per image and 0.2 FNs per image, and an area under

the ROC curve of 0.87 were obtained in the DDSM

database.

The remainder of this paper is organized as

follows: section 2 covers all the technical details like

used databases, mass detection algorithm description

and evaluation methodology; in section 3 results of

our method are shown and compared with those of

other competing methods. The paper closes in

section 4 with some final remarks and future work

directions.

2 MATERIALS AND METHODS

In this section we detail on the set of mammographic

images used, the mass detection algorithm and the

methodology used to evaluate the results.

2.1 Databases

Two mammogram databases were used, mini-MIAS

and INBreast. Mini-MIAS (Suckling et al., 1994)

consists of 330 images in which every image is of

size

1024*1024. These 330 images include 209

normal images, 56 images with at least one mass,

and the remaining have other types of anomalies.

One of the images (mdb059) was discarded in our

experiments because there is no information about

the center of the mass present in the mammogram.

INBreast database (Moreira et al., 2012) has a total

of 115 cases (410 images) of which 90 cases are

from women with both breasts (4 images per case)

and 25 cases are from mastectomy patients (2

images per case). Several types of lesions (masses,

calcifications, asymmetries, and distortions) are

included. In this work 107 images were used (all

images with at least one mass) with a total of 116

masses.

Note that, while mini-MIAS is a well-known

database, with the advantage of being already used

in several published works, it is a small database of

digitized mammograms having only the center and

radius information about the findings’ location.

INBreast, however, is a recent database having the

disadvantage of not being used by many works yet

making it more difficult to compare among different

algorithms. It has, as advantages, the fact that all the

images are Full-field digital mammograms and

accurate information on the form of detailed

contours on the shape and location of every finding

is available.

2.2 Detection

Our mass detection algorithm is a local adaptive

thresholding method which has been inspired from

binary search to determine an appropriate threshold

related to each local region (called cell).

The flowchart of the mammogram mass

detection algorithm (applied to each cell of the grid

respectively) is shown in Figure1.

Each image is first divided into equal non-

overlapping cells (a grid). In each cell of the grid,

the pixel with maximum gray level is found. The

location of the maximum pixel is shown as Index

and its value is named m.

First and Last are the bounds of the range which

is being explored and TH is the proper threshold.

First and Last are initialized to 0 and m respectively.

BIOSIGNALS2013-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

134

Figure 1: Flowchart of the mammogram segmentation algorithm.

In the first iteration, TH is assigned with the

middle value of the range [0;m] and then the

threshold is applied to the whole mammogram. After

that, the circularity measure is extracted from the

region that contains index (index is the location of

the maximum value):

Circularity =P

/ 4πA (1)

In Equation 1, P is the circumference of the region

and A is the area. In this equation, maximum

circularity is 1 and the less circular the region, the

bigger the circularity value will be. If the area or the

circularity of that region exceeds the corresponding

upper limit (Area

max, Circmax) then we should search

within the upper half of the previous range (i.e.

[TH;Last]). Else, if the area of the region is lower

than a threshold (Area

min), then the TH is too high

and we should search the proper TH in the range

[First;TH]. These operations are iterated until a

region with the area between Area

min and Areamax

and also less circularity than Circmax is found (if it

exists). In fact, masses generally have a radius

between a lower limit and an upper limit and are

also not very irregular. Although spiculated masses

are irregular in shape, their circularity can be lower

than a predefined limit and the irregularity occurs in

the margins of those masses and has a fairly small

effect on the whole circularity of those regions.

Moreover, we defined another measure as:

ratio

= Area / Circularity (2)

to filter curve-linear structures such as blood vessels

and milk ducts in mammograms. If a region has a

value lower than a predefined threshold, it is

considered as a curve-linear object and discarded.

2.3 Evaluation

We combine three typical rules to create a very

strong tagging rule. These rules are as follows:

• c

: if |Y

− Y

cad

| < max(R

,LY/2) and |X

− X

cad

< max(R

,LX/2)

• c

: if (X

cad

− X

)

+ (Y

cad

− Y

)

≤ R

• c

: if exists 50% overlap between biopsy-proven

mass and suspicious region

Where (LX, LY) are the length and width of the

detected ROI bounding box, (Xcad, Ycad) is the

region’s center of gravity and (Xb, Yb);Rb are center

and radius of the biopsy-proven mass. An extracted

ROI is labelled as True Positive (TP) if all the above

ANewLocalAdaptiveMassDetectionAlgorithminMammograms

135

rules are true. Otherwise, that ROI is labelled as FP.

3 EXPERIMENTS AND RESULTS

All images were scaled to 512*512 pixels and the

parameters needed for the mass detector were

empirically set as: Areamax = 8000 pixels, Areamin

=155 pixels and Circmax = 7. Some detection

examples are given in Figure 2 for mini-MIAS and

in Figure 3 for INBreast database.

In Table 1 some detection methods are shown for

comparison. From Table 1 it can be seen that only

the sensitivity of Density-Weighted Contrast

Enhancement (DWCE) filter is comparable with the

sensitivity of our mass detection algorithm. But,

with almost the same sensitivity, our method results

in fewer FPs per image (4.77) than those of DWCE

filter (12).

Figure 2: mini-MIAS mass detection examples. Left

column: original mammograms; right column: regions

obtained after segmenting the original mammograms.

Region growing (Eltonsy et al., 2007), Adaptive

thresholding (Kom et al., 2007) and Difference of

Gaussians (DoG) filter (Oliver et al., 2010) have

lower sensitivity and also more FPs in comparison

with our method. Although Template matching

(Nguyen et al., 2010) results in less FP, its

sensitivity is very low, making it unreliable.

By using the proposed method, the detection of

masses in 261 mammograms lasted 388 minutes,

i.e., one minute and 48 seconds for each image.

As far as we know, this work is the first

published work presenting mass detection results on

INBreast database. We have achieved a sensitivity

of 87% on INBreast (the algorithm missed 15

masses) and FP rate per image is 3.67.

Figure 3: INBreast mass detection examples. Left column:

original mammograms; right column: regions obtained

after segmenting the original mammograms.

Table 1: Results of different detection algorithms.

Detection method Sensitivity FP rate

Template matching (Nguyen et al., 2010) 0.38 2.9

Region growing (Eltonsy et al., 2007) < 0.6 > 8.5

Adaptive thresholding (Kom et al., 2007) < 0.65 > 9

DWCE filter (Petrick et al., 1996) < 0.9 > 12

DoG filter (Oliver et al., 2010) < 0.72 > 10.5

Our proposed method 0.91 4.77

4 CONCLUSIONS

In this paper, we tested a mass detection algorithm

BIOSIGNALS2013-InternationalConferenceonBio-inspiredSystemsandSignalProcessing

136

on two public mammogram databases. The

algorithm, inspired by binary search, reached a

sensitivity of 91% with false positive rate per image

of 4.77 in mini-MIAS database and a sensitivity of

87% with a false positive rate per image of 3.67 in

INBreast.

REFERENCES

Eltonsy, N. H., Tourassi, G. D., and Elmaghraby, A. S.

(2007). A concentric morphology model for detection

of masses in mammography. IEEE Transactions

on Medical Imaging, 26(6):880–889.

Gao, X., Wang, Y., Li, X., and Tao, D. (2010).On

combining morphological component analysis and

concentric morphology model for mammographic

mass detection. IEEE transactions on information

technology in biomedicine, 14(2):266–273.

Kom, G., Tiedeu, A., and Kom, M. (2007). Automated

detection of masses in mammograms by local adaptive

thresholding. Computers in Biology and Medicine,

37(1):37–48.

Malagelada, A. O. (2007). Automatic mass segmentation

in mammographic images. Ph.D. dessertation,

Department of Electronics-Computer Science and

Automatic Control, University of Girona.

Martins, L. D. O., Paiva, A. C. D., and Gattass, M. (2009).

Detection of breast masses in mammogram images

using growing neural gas algorithm and ripleys k

function. Journal of Signal Processing Systems,

55:7790.

Mencattini, A. and Salmeri, M. (2011). Breast masses

detection using phase portrait analysis and fuzzy

inference systems. International Journal of Computer

Assisted Radiology and Surgery, page 111.

Moreira, I. C., Amaral, I., Domingues, I., Cardoso, A.,

Cardoso, M. J., and Cardoso, J. S. (2012). INbreast:

toward a full-field digital mammographic database.

Academic radiology, 19(2):236–248.

Nguyen, V. D., Nguyen, D. T., Nguyen, T. D., Thi, N.,

and Tran, D. H. (2010). A program for locating

possible breast masses on mammograms. In

Proceeding of the 3rd International Conference on the

Development of BME in Vietnam, pages 11–14.

Oliveira, M. L. D., Braz, J. G., Cardoso, P., and Gattass,

M. (2009). Detection of masses in digital

mammograms using k-means and support vector

machine. Electronic Letters on Computer Vision and

Image Analysis, 8(2):39-50.

Oliver, A., Freixenet, J., Marti, J., Prez, E., Pont, J.,

Denton, E. R. E., and Zwiggelaar, R. (2010). A review

of automatic mass detection and segmentation in

mammographic images. Medical Image Analysis,

14(2):87–110.

Oliviera, J. d., Albuquerque, A. d., and Deserno, T. M.

(2011). Content-based image retrieval applied to

BIRADS tissue classification in screening

mammography. World Journal of Radiology, 3(1):24-

31.

Petrick, N., Chan, H., Sahiner, B., and Wei, D. (1996). An

adaptive Density-Weighted contrast enhancement

filter for mammographic breast mass detection. IEEE

Transactions on Medical Imaging, 15(1):59–67.

Sampaio, W. B., Diniz, E. M., Silva, A. C., de Paiva, A.

C., and Gattass, M. (2011). Detection of masses in

mammogram images using CNN, geostatistic

functions and SVM. Computers in Biology and

Medicine, 41(8):653-664.

Suckling, J., Parker, J., Dance, D. R., Astley, S., Hutt, I.,

Boggis, C., Ricketts, I., Stamatakis, E., Cerneaz, N.,

Kok, S. L., Taylor, P., Betal, D., and Savage, J.

(1994). The mammographic image analysis society

digital mammogram database. In Exerpta Medica.

International Congress Series 1069, volume 1069,

pages 375–378.

ANewLocalAdaptiveMassDetectionAlgorithminMammograms

137