Image Processing Reference
In-Depth Information
From Table 3 , compare the corresponding values of m and σ of each image, we can find that
most of the m outside , m inside and almost all σ outside , σ inside values obtained by modified method are
much smaller than the typical results. The m value shows how much the curves deviated, and
the smaller m indicates that the curve is much close to the real ones. The σ can exactly relects
the level of similarity of two curves, the smaller σ is, the much similar to the manual-segmen-
ted results. Thus, we can come to the conclusion that the contours obtained by modified meth-
od are more similar to the manual-segmented results compared with the typical method.
Table 3
The Similarity of Different Methods
Evaluation
Image
( m outside / R equ ± σ outside / R equ ) − ( m inside / R equ ± σ inside / R equ )
Typical Method
Improved Method
Figure 5(a) (0.0813 ± 0.0055) − (0.1709 ± 0.0040) (0.0589 ± 0.0035) − (0.0637 ± 0.0016)
Figure 5(b) (0.1727 ± 0.0099) − (0.0628 ± 0.0147) (0.0298 ± 0.0018) − (0.0425 ± 0.0080)
Figure 5(c) (0.1021 ± 0.0124) − (0.0589 ± 0.0033) (0.0477 ± 0.0039) − (0.0456 ± 0.0014)
Figure 5(d) (0.2076 ± 0.0092) − (0.1180 ± 0.0087) (0.0758 ± 0.0003) − (0.0592 ± 0.0028)
Figure 6(a) (0.0505 ± 0.0020) − (0.1508 ± 0.0089) (0.0465 ± 0.0022) − (0.1691 ± 0.0036)
Figure 6(b) (0.1289 ± 0.0058) − (0.1008 ± 0.0074) (0.1171 ± 0.0054) − (0.0659 ± 0.0041)
Figure 6(c) (0.5395 ± 0.0201) − (0.2628 ± 0.0204) (0.1627 ± 0.0153) − (0.3878 ± 0.0185)
Figure 6(d) (0.1328 ± 0.0025) − (0.1239 ± 0.0044) (0.0510 ± 0.0027) − (0.1265 ± 0.0029)
4 Conclusions
In this work, we present an effective integrated approach based on the improved VFC Snake
model for mass automatic segmentation in mammogram which with low contrast and blurry
boundaries. First of all, the local threshold method, RS theory, and morphological filter are
applied to preprocess the mammograms to remove the labels and enhance the whole image.
Then we use the LHT and CHT algorithms to locate the massive lesions and the position of
which parametrically indicated as an approximate circle. The mass segmentation stage uses
the parametric circle to initialize the deformable method which is defined by improving the
force field of typical VFC Snake model and extract the mass boundary accurately. The pro-
posed approach is tested on DDSM and MIAS database, respectively, and the results show
that our algorithm achieves a higher detection rate and superior segmentation accuracy com-
pared with the typical VFC Snake model. What's more, the segmented contours are much sim-
ilar to the actual boundary of objects. In conclusion, the improved approach can not only loc-
ate and segment the mass automatically, but also in lower dependence on the initial active
contour and with stronger capability of convergence. Besides, this algorithm is robust to the
interference of blurry areas and tissue and able to converge precisely to the object. In addition,
the results conform to the pathology characteristics of actual masses to some extent and bene-

Search WWH ::

Custom Search