adfa, p. 1, 2011.
© Springer-Verlag Berlin Heidelberg 2011
Segmentation of Mammography by Applying Extreme
Learning Machine in Tumor Detection
Cordeiro, F.R.
1
; Lima, S.M.L.
1
; Silva-Filho, A.G.
1
and Santos, W.P.
2
1
Federal University of Pernambuco - Informatics Center, Recife, Brazil
{frc, smll, agsf}@cin.ufpe.br
2
Federal University of Pernambuco - Department of Biomedical Engineering, Recife, Brazil
wellington.santos@ufpe.br
Abstract. Locating regions of tumor in digital mammography images is a hard
task even for experts. Consequently, due to medical experience, different diag-
noses to an image are commonly found. Therefore, the use of an automatic ap-
proach for detecting tumor regions is important to avoid misdiagnosis. In this
work, the Extreme Learning Machine (ELM) neural network was used to seg-
ment tumor regions of digitized mammograms available in the Mini-Mias data-
base. A set of images were selected for training, while different images were
used for testing. Results showed that ELM provides an over 81% classification
rate, being able to segment the region of tumor with high accuracy. By compar-
ing ELM with MLP network, it was possible to conclude that ELM has a faster
learning time, with a higher training and testing accuracy.
Keywords: Breast Cancer, Segmentation, Neural Network, Extreme Learning
Machine.
1 Introduction
The World Health Organization (WHO) estimates that 1,100,000 new cases of breast
cancer appear yearly worldwide [1]. Recent changes in lifestyle in society in-crease
obesity, which has a direct influence on breast cancer [2]. In developing countries,
like Brazil, this type of cancer is one of the main causes of death among women
[2][3].
It is estimated that the period between the beginning of the tumor and its growth
until it becomes palpable, reaching around 1cm, is of about 10 years [4]. During this
period, breast imaging is essential to the tumor attendance. The correct evaluation of
the tumor size takes an important role in the planning of the breast cancer treatment,
avoiding mutilating surgeries, such as mastectomy [5].
The tumor size is one of the most important factors in order to adopt or avoid
breast conservative therapies [5]. However, image devices determined by BMH (Bra-
zilian Ministry of Health) [6] for the detection of breast cancer are quite inefficient at
the evaluation of the nodule sizing. These methods depend substantially on profes-
sional examiner’s experience.