The Potential Use of Bioinspired Algorithms Applied in the

Segmentation of Mammograms

David González-Patiño

, Yenny Villuendas-Rey

and Amadeo J. Argüelles-Cruz

Centro de Investigación en Computación del Instituto Politécnico Nacional, Avenida Juan de Dios Bátiz esq. Miguel

Othón de Mendizábal, Nueva Industrial Vallejo, Gustavo A. Madero, 07738, Ciudad de México, D.F., Mexico

Centro de Innovación y Desarrollo Tecnológico en Cómputo del Instituto Politécnico Nacional, Av. Juan de Dios Bátiz s/n,

Nueva Industrial Vallejo, Gustavo A. Madero, 07700 Ciudad de México, D.F., Mexico

Keywords: Segmentation, Mammography, Bioinspired Algorithm.

Abstract: In this article, we present the potential use of bioinspired algorithms for segmentation. The comparison is

done with 3 bioinspired algorithms and Otsu method, which is an algorithm currently used to perform image

segmentation. The vast majority of bioinspired algorithms were designed for optimization, however in this

work, an adjustment is done to use the function to be optimized as a function that allows us to segment an

image. The results in this work showed that the bioinspired algorithms are a good alternative to perform the

task of segmentation in medical images, specifically mammography.

1 INTRODUCTION

Breast cancer is a global problem and one of the

cancers that has caused more than 410,000 deaths

until 2002 (Ferlay et al., 2010). An early diagnosis

of breast cancer will result in more opportunities and

treatments which will reduce death risk of the

patient.

This is the reason that there are mammography

segmentation techniques, which aim to detect small

abnormal regions within the breast. This

segmentation process allows us to detect regions of

interest in the medical image and, in case of being

necessary, to carry out a subsequent treatment by the

medical specialist.

In this work we explore the possibility of using

bioinspired algorithms for the segmentation of

mammographies. These algorithms are based on a

biological process used to find an optimal solution to

a problem.

Although many of the bio-inspired algorithms

are mainly designed to solve optimization problems,

in this paper we propose to use a function that

allows us to transform the optimization process into

a segmentation process.

Nowadays, the use of segmentation algorithms in

medical areas has allowed the early diagnosis of

many diseases including breast cancer.

This paper is organized as follows, In section 2,

some of the previous works are presented. In section

3, the tested algorithms for segmentation are

explained as well as the materials used. The results

and comparison of the segmentation algorithms are

presented in section 4. And finally in section 5, we

present the conclusions and future work related to

this area.

2 PREVIOUS WORKS

In the existing field called Evolutionary

Computation, there has been many approaches such

as evolutionary methods and swarm intelligence

which have showed good performances solving

problems in computer vision (Vite-Silva et al.,

2007), clustering (Abraham et al., 2008), network

routing (Kassabalidis et al., 2001), among others.

Related to swarm intelligence, some algorithms

have been used in optimization of classifiers

(Ramirez et al., 2015), instance selection (Derrac et

al., 2012) and optimal parameter values selection

(Friedrichs and Igel, 2005).

Catarious et al. proposed a system to identify

suspicious masses in mammograms (Catarious et al.,

2004), similarly in 2006, Scharcanski and Jung

(2006) proposed a method to improve the contrast

and eliminate noise in mammographic images.

Related to bioinspired algorithms, Brodić and

González-Patiño, D., Villuendas-Rey, Y. and Argüelles-Cruz, A.

The Potential Use of Bioinspired Algorithms Applied in the Segmentation of Mammograms.

DOI: 10.5220/0006951103030306

In Proceedings of the 10th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2018) - Volume 1: KDIR, pages 303-306

ISBN: 978-989-758-330-8

303

Milivojević (2011) proposed a modified version of

the water flow algorithm which they applied to the

segmentation of texts.

Other approaches (Bevilacqua et al., 2009) have

been used to detect blood vessels in retinal fundus

using artificial neural networks.

Other works presented by Frausto-Solís et al.,

(2013)in 2013 and, Rad and Lucas (Rad and Lucas,

2007) in 2007 showed the use of bioinspired

algorithms (simulated annealing algorithm and IWO

algorithm respectively) showing good performances.

Although there are many algorithms to perform

the segmentation, there is not yet a perfect algorithm

that allows its use in all segmentation problems, that

is why this work presents modified bioinspired

algorithms that allow us to perform this task.

3 MATERIALS AND METHODS

Many of the bioinspired algorithms were mainly

designed for optimization tasks. However, in this

work, we propose the possibility of using these

algorithms for segmentation by using the Dunn

index and using each individual component as if it

were a gray level of the gray scale. In this work we

made a comparison of the Otsu method (Otsu, 1975)

against 3 bioinspired algorithms (Novel Bat

Algorithm (Meng et al., 2015), Invasive Weed

Optimization (Mehrabian and Lucas, 2006) and

Particle Swarm Optimization (Kennedy and

Eberhart, 1995)).

Given that the bioinspired algorithms used in this

work have as their main component Individuals, it is

necessary to present the representation that they will

have in the system. The composition of an individual

will be given by a vector composed of values that

represent gray levels as presented in figure 1.

Figure 1: Representation of an individual.

3.1 Novel Bat Algorithm (NBA)

Bat algorithm was proposed in 2010 by Yang (Yang,

2010), as an optimization algorithm based on the

behavior of bats and, specifically, the techniques

they use to identify food using echolocation.

Later, Meng (Meng et al., 2015) propose a

modification of the algorithm proposed in 2010,

which he called Novel Bat Algorithm (NBA), this

algorithm includes among other novelties, the

Doppler Effect and the ability to search the global

solution in different habitats. The implementation of

this latest version was the one used for this work.

3.2 Invasive Weed Optimization (IWO)

This algorithm of numeric optimization was

introduced by Mehrabian and Lucas (2006) in 2006

and is based on weed colonies and how they adapt to

the environment to spread their seeds. Unlike other

metaheuristics, this algorithm allows all individuals

(plants) to participate in later generations. This

aspect gives the opportunity to plants with better

performance to produce more seeds and plants with

lower performance, will produce fewer seeds.

Another notable difference of this algorithm is that

individuals produce seeds autonomously, that is, the

mating process does not occur.

3.3 Particle Swarm Optimization

(PSO)

This algorithm proposed by Kennedy and Eberhart

(1995), is based on the movement of the particles.

The search and refinement of candidate solutions is

based on a social swarm model (such as birds or

fishes) that pretends to find food.

3.4 Error Measures

According to Zhang et al., (2008), it is possible to

determine the quality of a segmentation by

calculating two error measures (F and F’). F

measures the average squared color error while F’ is

a modification of F that penalizes segmentations

with small regions.



√

















(1)







1000∗





























(2)

3.5 Dunn Index

Since the bioinspired algorithms used in this work

were designed for optimization, in this work, the

possibility of using them for segmentation is

proposed changing the function to be optimized as

the division of the minimum outercluster distance

between the maximum intercluster distance. This

150 50 240

KDIR 2018 - 10th International Conference on Knowledge Discovery and Information Retrieval

304

relationship is called the Dunn index (Dunn, 1973)

and is used as a function to be maximized since it is

intended to have the greatest distance between

clusters and the smallest distance between the

elements on the same cluster.

4 RESULTS

In this work, 362 images of real patients were

segmented, which belong to the Breast Cancer

Digital Repository database (Moura and López,

2013). This dataset contains the segmentation of said

images, carried out by a group of radiological

experts.

(a)

(b)

(c)

(d)

(e)

(f)

Figure 2: (a) Original image, (b) Overlap of the ROI

segmented by expert radiologist over original image, (c)

Image segmented by Otsu method, (d) Image segmented

by NBA, (e) Image segmented by IWO, (f) Image

segmented by PSO.

In Figure 2, we present a mammographic image

taken at random from the 362 images of the dataset.

This image was segmented by the 4 segmentation

algorithms and the image results are presented in the

image.

The values presented in Fig. 3, are expressed in

10



for F mean, and 10



for F’ mean. These

values were calculated for the 362 images and the

average is presented in Fig. 3 for each algorithm.

In Figure 3, it is observed that IWO obtained a

lower error than the other bioinspired algorithms.

The other bioinspired algorithms obtained lower

errors than Otsu method, which at the present time

remains as an algorithm widely used to segment.

Figure 3: Mean of the errors calculated for the

segmentation of the 362 mammographies.

Table 1: Mean time in seconds per image segmented.

Algorithm Mean Time per image (seconds)

Otsu method 0.051

Novel Bat Algorithm 53.83

Invasive Weed Optimization 412.71

Particle Swarm Optimization 36.93

362 images were processed with each algorithm and

the processing times were calculated for each image.

Later, the average time was calculated, these times

are presented in Table 1. It is observed that Otsu

method got a very small time to process each image.

It is relevant to mention that the Matlab

implementation of this algorithm was used. The

other algorithms were implemented and modified to

be used as segmentation algorithms.

5 CONCLUSIONS

In this work, a comparison of the bioinspired

algorithms applied to segmentation is presented,

something relevant given that most of these

algorithms are mainly designed for optimization.

We compared the errors calculated for each

segmentation performed by each bioinspired

The Potential Use of Bioinspired Algorithms Applied in the Segmentation of Mammograms

305

algorithm against the Otsu method, which is widely

used to perform image segmentation.

IWO algorithm got the lowest average error,

although it also got the highest time to process each

image.

It is important to clarify that, in the medical field,

there must be a balance between time and quality,

since we are dealing with real patients and the

incorrect classification of a disease is a very

important factor.

The results presented in this article show a good

performance of the bioinspired algorithms in the task

of segmentation, so it is a good alternative to carry

out a more detailed analysis in this field.

As future work, we intend to carry out a more

detailed study and explore the possibility of

automatically adjusting the parameters of the

algorithms.

ACKNOWLEDGEMENTS

The authors gratefully acknowledge the CONACYT,

Instituto Politécnico Nacional, COFAA-IPN and

Honeywell for their economical support to develop

this work.

REFERENCES

Abraham, A. et al., 2008. Swarm intelligence algorithms

for data clustering, in Soft computing for knowledge

discovery and data mining. Springer, pp. 279–313.

Bevilacqua, V. et al., 2009. A Comparison Between a

Geometrical and an ANN Based Method for Retinal

Bifurcation Points Extraction., Journal of Universal

Computer Science, 15(13), pp. 2608–2621.

Brodić, D. and Milivojević, Z., 2011. A new approach to

water flow algorithm for text line segmentation,

Journal of Universal Computer Science, 17(1), pp.

30–47.

Catarious et al., 2004. Incorporation of an iterative, linear

segmentation routine into a mammographic mass

CAD system, Medical physics. Wiley Online Library,

31(6), pp. 1512–1520.

Derrac, J. et al., 2012. Enhancing evolutionary instance

selection algorithms by means of fuzzy rough set

based feature selection, Information Sciences.

Elsevier, 186(1), pp. 73–92.

Dunn, J. C., 1973. A fuzzy relative of the ISODATA

process and its use in detecting compact well-

separated clusters. Taylor & Francis.

Ferlay, J. et al., 2010. Global burden of breast cancer, in

Breast cancer epidemiology. Springer, pp. 1–19.

Frausto-Solís, J. et al., 2013. Cluster Perturbation

Simulated Annealing for Protein Folding Problem,

Journal of Universal Computer Science, 19(15), pp.

2207–2223.

Friedrichs, F. and Igel, C., 2005. Evolutionary tuning of

multiple SVM parameters, Neurocomputing. Elsevier,

64, pp. 107–117.

Kassabalidis, I. et al., 2001. Swarm intelligence for

routing in communication networks, in Global

Telecommunications Conference, 2001.

GLOBECOM’01. IEEE, pp. 3613–3617.

Kennedy, J. and Eberhart, R., 1995. Particle swarm

optimization (PSO), in Proc. IEEE International

Conference on Neural Networks, Perth, Australia, pp.

1942–1948.

Mehrabian, A. R. and Lucas, C., 2006. A novel numerical

optimization algorithm inspired from weed

colonization, Ecological informatics. Elsevier, 1(4),

pp. 355–366.

Meng, X.-B. et al., 2015. A novel bat algorithm with

habitat selection and Doppler effect in echoes for

optimization, Expert Systems with Applications.

Elsevier, 42(17), pp. 6350–6364.

Moura, D. C. and López, M. A. G., 2013. An evaluation of

image descriptors combined with clinical data for

breast cancer diagnosis., International journal of

computer assisted radiology and surgery. Springer,

8(4), pp. 561–574.

Otsu, N., 1975. A threshold selection method from gray-

level histograms, Automatica, 11(285–296), pp. 23–

27.

Rad, H. S. and Lucas, C., 2007. A recommender system

based on invasive weed optimization algorithm, in

Evolutionary Computation, 2007. CEC 2007. IEEE

Congress on, pp. 4297–4304.

Ramirez, A. et al., 2015. Evolutive improvement of

parameters in an associative classifier, IEEE Latin

America Transactions. IEEE, 13(5), pp. 1550–1555.

Scharcanski, J. and Jung, C. R., 2006. Denoising and

enhancing digital mammographic images for visual

screening, Computerized Medical Imaging and

Graphics. Elsevier, 30(4), pp. 243–254.

Vite-Silva, I. et al., 2007. Optimal triangulation in 3D

computer vision using a multi-objective evolutionary

algorithm, in Workshops on Applications of

Evolutionary Computation, pp. 330–339.

Yang, X.-S., 2010. A New Metaheuristic Bat-Inspired

Algorithm, in González, J. R. et al. (eds) Nature

Inspired Cooperative Strategies for Optimization

(NICSO 2010). Berlin, Heidelberg: Springer Berlin

Heidelberg, pp. 65–74. doi: 10.1007/978-3-642-

12538-6_6.

Zhang, H. et al., 2008. Image segmentation evaluation: A

survey of unsupervised methods, computer vision and

image understanding. Elsevier, 110(2), pp. 260–280.

KDIR 2018 - 10th International Conference on Knowledge Discovery and Information Retrieval

306