Several of the commonly used medical screening techniques used for breast screening
such are; mammography, ultrasound and Magnetic Resonance Imaging (MRI). Mammograms
and ultrasound are difficult to interpret because of the sensitivity of screening
is affected by the image quality and the radiologists level of expertise
(Rangayyan et al., 2007). While MRI images are
clear and have better contrast between most of the breast regions. For that
reason, MRI is used for breast screening. Although, some regions in MRI breast
images have the same intensity values and that may affect the segmentation process.
Therefore, the MRI breast images are enhanced using computer techniques of image processing in Computer Aided Detection (CAD) systems that help radiologists in detecting the tumour regions.
The intensity level between the breast skin-line and the tumor regions is similar
in the majority of MRI breast cases. Thus, segmenting and removing the breast
skin-line region methods will be discussed in this study. Without this, segmentation
errors may occur if not managed correctly (Solves Llorens
et al., 2012). This process would be more importance in the next
stage of the automatic tumor segmentation in CAD systems.
MATERIALS AND METHODS
The proposed approach of MRI breast skin line segmentation and removal consists of three main stages. These are: image pre-processing using the median filter, followed by skin line segmentation using Level Set Active Contour which is applied on the resultant image in the first stage. The last stage involves skin line removal using Morphological Thinning. The following is a detailed description of the processes involved.
Image pre-processing: As mentioned, the pre-processing phase is the first process that is executed. The median filter is applied in order to enhance the images resolution and to reduce the presence of the salt and pepper noise.
Breast skin-line segmentation: To segment the breast skin line, Level
Set Active Contour algorithm (Li et al., 2005)
is used. Level set formulation without re-initialization (which is specially
known as Chunmings algorithm) is included as one of the active contours
detection algorithms. Active contours are dynamic curves that move toward the
mass border. An external energy moves the zero level curves toward the mass
border using the edge indicator function g that is defined in Eq.
where, I is the image, Gσ is the Gaussian kernel with a standard deviation σ. By changing the σ parameter value and the number of the iterations NLS, the resultant detection is changed. Figure 1 shows the different results after Level Set Active Contour algorithm with different values of σ and NLS is applied.
Based on Fig. 1, it can be observed that the higher number of iterations NLS gives better detection. Therefore, the best value for NLS is 700 while the best value of σ is 1.5. These choices of the values are generic for all breast MRI images after resizing the image to be 288x288 pixels.
Breast skin-line removal: In order to delete the detected breast skin
border, the Morphological Thinning Algorithm as described by Lam
et al. (1992) is used. The image is divided into two distinct sub-fields.
Then in the first sub-iteration, pixel p from the first sub-field is deleted
if and only if the first three conditions as shown below are true.
||Different results after the application of level set active
contour algorithm with different values of σ and NLS. (a)
σ = 3, NLS = 100, (b) σ = 1.5, NLS =100,
(c) σ = 3, NLS = 300, (d) σ = 1.5, NLS =
300, (e) σ = 3, NLS = 700 and (f) σ = 1.5, NLS
In the second sub-iteration, pixel p from the second sub field is deleted
if and only if, the first two and fourth conditions are true.
x1, x2, ..., x8 are the values of the eight neighbours of p, starting with the east neighbour and numbered in counter-clockwise order.
The two sub iterations formulate the main iteration of the thinning algorithm.
The thinning level depends on the number of iterations. Whenever the number
of iterations is increased, there would be more shrinking of the images
border. The Morphological Thinning Algorithm only accepts a binary version of
the image. Therefore, the resultant image after the Level Set Active Contour
algorithm would be converted to binary image. Furthermore, after the thinning
procedure, the binary image is reconverted into its original grey scale representation.
Figure 2 shows the results after the Morphological Thinning
algorithm is applied using three different iteration numbers (NTh
= 1, NTh = 3 and NTh = 7).
||Results after applying morphological thinning algorithm with
three different iteration numbers on the resultant image of level set active
contour algorithm (a) After applying thinning (NTh = 1) on binary
image, (b) After reconverting thinning image (NTh = 1) to its
original grayscale, (c) After applying thinning (NTh = 3) on
binary image, (d) After reconverting thinning image (NTh = 3)
to its original grayscale, (e) After applying thinning (NTh =
7) on binary image and (f) After converting thinning image (NTh
= 7) to its original grayscale
From Fig. 2, NTh =7 is the best iteration number which enables the thinning algorithm whereby the deleted amount of pixels equals the thickness of the breast skin.
Evaluation approach: To evaluate the results of the segmentation and
the removal processes, pixel based evaluation approach (Rosin
and Ioannidis, 2003) is used. This approach compares between pixels of processed
image (Rt) and ground truth (Rg). The following measures
are used in this work to build the comparison; Dice index, Jaccard index, True
Positive Fraction (TPF), False Negative Fraction (FNF), False Positive Fraction
(FPF) and True Negative Fraction (TNF). The calculations are made using the
equations 10-15. (McNeil
and Hanley, 1984; Metz, 1986; Chalana
and Kim, 1997; Rosin and Ioannidis, 2003; Fenster
and Chiu, 2005; Prasad et al., 2011):
As the basis of measurements: for the Dice, Jaccard, TPF and TNF approaches, whenever the results are higher, the performance is better. Meanwhile for the rest of the basis of measurements which are MCR, FNF and FPF, the lower results indicate better performance.
The methodology explained earlier is applied and tested on the RIDER Breast
MRI dataset which is downloaded from the National Biomedical Imaging Archive
(NBIA)(US National Cancer Institute, 2007). This website
belongs to the U.S. National Cancer Institute. The dataset includes breast MRI
images for five patients. All images are Axial 288x288 pixels. Three sequences
are selected for each patient to be used in the experiments as test images.
Two Ground Truth (GT) images sets are prepared manually for the purpose of the
evaluation; the first GT is to evaluate the segmentation process while the second
GT set is to evaluate the removal process.
In the breast skin-line segmentation stage, the chosen parameters for Level Set algorithm are σ = 1.5 and NLS =700 while NTh =7 is the chosen parameter for the Thinning algorithm for the removal stage. The parameters have been selected using the Trial and Error method. Figure 3 shows five RIDER patients images and the processes of breast skin-line segmentation and deletion processes.
The proposed approach is applied on all the RIDER dataset images. Then the
calculations are made by comparing these with the ground truth using the evaluation
measures eq.10-15. The results of the
segmentation stage are tabled as in Table 1 while the results
of the skin-line removal are tabled as in Table 2.
||(a-e) Breast skin segmentation and removal processes on the
five RIDER images; original images (f-j) Five images after applying level
set algorithm with σ = 1.5 and NLS = 700 and (k-o) Then
applying the Thinning Algorithm with NTh = 7
|| Results of skin-line segmentation for RIDER MRI breast images
using evaluation measures (Dice, Jaccard, MCR, TPF, FNF, FPF and TNF)
|| Results of skin-line removal for RIDER MRI breast images
using evaluation measures (Dice, Jaccard, MCR, TPF, FNF, FPF and TNF)
From Table 1 and 2, it can be concluded that the proposed approached scored good results using the various measures. For the skin-line segmentation stage, the mean of results was high with (Dice = 0.9607), (Jaccard = 0.9275), (MCR = 0.0354), (TPF = 0.9646), (FNF = 0.0354), (FPF = 0.0401) and (TNF = 0.9599). For the skin-line removal stage the mean of results was high with (Dice = 0.9099), (Jaccard = 0.8394), (MCR = 0.1453), (TPF = 0.8547), (FNF = 0.1453), (FPF = 0.0186) and (TNF = 0.9814).
Then, Receiver Operating Characteristic (ROC) is applied to illustrate by drawing curve the True Positive Fraction compared with the False Positive Fraction as shown in Fig. 4 and 5.
From the curves, it can be observed that the Area under the Curve (AUC) for the segmentation stage is 0.9902 and AUC for the removal stage is 0.9507. The high AUC indicates improved segmentation performance.
|| ROC curve for MRI breast skin-line segmentation
|| ROC curve for MRI breast skin-line removal
An integration method of Level Set Active Contour algorithm and Morphological Thinning algorithm is presented in this work. The procedure for segmenting and removing the skin-line region from MRI breast images is a necessary process before the next level of the breast segmentation processes. The proposed approach is divided into three stages; pre-processing using median filter, skin-line segmentation using Level Set Active Contour which is then followed by skin-line removal using Morphological Thinning Algorithm.
The study is then tested by applying the methodology on the RIDER breast MRI dataset. The ground truth is manually prepared from the dataset as a benchmark. The evaluation statistic using six measures shows that the performance is significantly high compared with ground truth. The results of the segmentation stage are Dice = 0.9607, Jaccard = 0.9275, MCR = 0.0354, TPF = 0.9646, FNF = 0.0354, FPF = 0.0401 and TNF = 0.9599. While the results of the removal stage are Dice = 0.9099, Jaccard = 0.8394, MCR = 0.1453, TPF = 0.8547, FNF = 0.1453, FPF = 0.0186 and TNF = 0.9814. The proposed approach could also be implemented on other types of breast images such as the X-Ray mammogram images. However, it is necessary to take into account of changing the parameter values of the algorithms to fit specifications of the type of breast images.