Image Processing Reference
In-Depth Information
Foundation of China (Grant No. 61175012), Science Foundation of Gansu Province of China
(Grant Nos. 1208RJZA265 and 145RJZA181), Specialized Research Fund for the Doctoral Pro-
gram of Higher Education of China (Grant No. 20110211110026), and the Fundamental Re-
search Funds for the Central Universities of China (Grant No. lzujbky-2013-k06).
1 Introduction
Breast cancer is one of the common malignant tumors and remains the leading cause of cancer
death among females, accounting for 23% of the total cancer cases and 14% of the cancer
deaths in the world [ 1 ]. Mammography is a preferred method for early detection and also the
most efficient and reliable tool for early prevention and diagnosis of breast cancer [ 2 ] . Mam-
mograms are always with low contrast and the lesions are blurry and irregular, the shape
and size of each mass or calcification are changeful, which cause the high misdiagnosis rate
of breast cancer. For the past few years, computer-aid diagnosis has become the international
research hot spot worldwide [ 3 ] , which offers the doctors a reliable “second suggestion.”
Breast mass is an important symptom and its accurate segmentation is crucial to the treat-
ment of breast cancer. Different algorithms for early lesion area detection in mammograms
have been widely studied [ 4 - 7 ] . Kumar and Sureshbabu [ 4 ] detected mass in mammogram
automatically using wavelet transform modulus maximum (WTMM), which located the re-
gion of interest (ROI) by multithreshold and then extracted the contour of ROI by WTMM
method. Song et al. [ 5 ] proposed a hybrid segmentation method, which defined a local cost
function for dynamic programming based on the rough region of mass obtained by template-
matching technique, and the performance was evaluated by measuring the similarity. A
new mass segmentation and automatic estimation method were presented based on robust
multiscale feature-fusion and maximum a posteriori [ 6 ]. Before delineating the final mass, the
dynamic contrast improvement, template matching, and posterior probabilities were used to
obtain the mass candidate points. This method can segment the ill-defined or spiculated le-
Novel image detection methods are appearing along with the development of technology.
In recent years, the active contour model (Snake) [ 8 ] is widely used in image processing, com-
puter vision, etc. [ 9 , 10 ] , and among which VFC (vector field convolution) Snake [ 11 ] model
performs more excellent characteristics in segmentation of boundaries such as low dependen-
ce to initial contour, capability of convergence and superior noise robustness. While it doesn't
work well when we apply the typical VFC Snake method to extract the mass in mammograms,
because of that the mass boundaries are always with low contrast and appearing blurry in the
whole image.
Considering the disadvantage mentioned above, we proposed an integrated approach for
mass autosegmentation in breast based on the improved VFC Snake model. The present meth-
od can detect the regions of masses in mammograms automatically and achieve promising res-
ults. This chapter is organized as follows. In Section 2 , we present the methodology for mass
localization and segmentation. Section 3 illustrates some experiments to verify the proposed
method. Besides, the comparisons with typical VFC Snake model and the discussion also can
be found in Section 3 . Section 4 gives the conclusions of this chapter.
2 Methodology
The proposed methodology of breast mass segmentation can be schematically described in
Figure 1 . The method consists of four main processing steps: (1) obtaining the mammogram
Search WWH ::

Custom Search