Automatic mass segmentation method in mammograms based on improved VFC snake model - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 168

the distribution of which is disordered ( Figure 4(b) and (c) ) , and this causes the misleading of
the active contour to the interference tissues rather than the real mass boundaries. Thus, the
main idea of our proposed segmentation algorithm is to improve the force field by deining
a new and clear edge map f ( x , y ). First, the gradient value of each pixel is improved using RS
theory method by seting a gradient threshold to judge whether the pixel belongs to a poten-
tial boundary or not. If it is, then these pixels should be enhanced. Next, the edge map f ( x , y ) is
calculated by performing the Canny operator [ 19 ] , which is considered to have more excellent
features like stability and accuracy compared to the gradient operator. As shown in Figure
4(d) and (e) , it is obvious that the improved force field distribution turns to be more regular,
which appears much evenly near the boundaries and the areas mixed in normal glands.
FIGURE 4 Force field of typical and improved VFC Snake and the enlarged images.
Besides, the performance of typical VFC Snake model depends on the position of initial con-
tour around the mass, which usually is a circle formed by the coordinate position and radius.
Here, we set the initial contour by using the parametric circle obtained in the mass location
part.
3 Experiment results and discussion
Our proposed method was totally tested on 400 mammograms with abnormal breast regions
from DDSM and MIAS database and the experimental results are shown in Figures 5 and 6 ,
respectively. We can observe that our approach can achieve much beter results in comparison
with the typical VFC Snake method. During this test, we set the parameter of VFC Snake mod-
el as: α = 0.5, β = 0.2, and the iteration is about 30. To fully explain the superiority of our pro-
posed method in visual, the illustrated mammograms consist of craniocaudal and mediolater-
al oblique view, whose severity includes benign and malignant, and each case contains only
one abnormality.