Breast cancer is one of the leading cancers in woman worldwide both in developed and developing nations as per the records from World Health Organization WHO. Many studies have shown that mammography is very effective for the breast cancer diagnosis. Mass detection poses a big challenge in detection because of its varying shape and density. Mass segmentation plays an important step for the cancer detection. Notable researches were done and still moving towards the effective detection of masses in mammograms. In this paper a novel segmentation of mass and its classification using Adaptive Neuro Fuzzy Inference System (ANFIS) is proposed. The segmentation includes two main steps. First, a rough initial segmentation through iterative thresholding, and second, an active contour based segmentation. The relevant statistical features are extracted and the classification is done by using ANFIS, which yields an accuracy of 91.30%.
Mammography, Mass segmentation, Iterative thresholding, Active contour, ANFIS.
Get your grade
or your money back
using our Essay Writing Service!
Over the past two decades, cancer has been one of the biggest threats to human life and it is expected to become the leading cause of death over the next few decades. There is no way of preventing cancer but can be examined early.
A mammogram is an X-ray of the breast. One can diagnose at very early stage and hence there is a chance for healthy survival. Breast cancer can be traced from mammogram before it can be felt. When mammography is combined with clinical breast exam the chances for finding cancer are even greater. For women with dense breast tissue, digital mammography may be more accurate than standard mammography. There are many other breast imaging tests  that can provide valuable information, however, these find difficult to tell the difference between dense breast tissue, benign (non-cancerous) lumps and cancer. And, sometimes they miss tiny calcium deposits that are the earliest sign of a tumor. Thus mammogram plays as a vital tool in breast cancer detection and identification . Masses are defined as space occupying lesions that are described by their shape and marginal properties. A benign is characterized by smooth margination, whereas a malignancy is characterized by an indistinct border that becomes more spiculated with time. Many researches were done to segment the mass in the mammograms , , , . Of which Active contour based segmentation provides the best way to segment the images whose background and foreground that are statistically different and homogeneous . The contour based segmentation retains the original information of region, edge and shape of the mass , . The region based active contour ,  model holds good than the edge based contour model , . Considering all the above advantages of the region based active contour, the segmentation technique combines the classical segmentation technique with the region based active contour. The rest of the paper is organized as follows. Section II details about the proposed method of mass segmentation and classification. Section III discusses about the statistical features extracted from the segmented images. Result and discussion are drawn in section IV. Conclusion is produced in section V.
Proposed Method of Mass Classification
Mammographic images are collected from the Imaging center, Coimbatore and from Mammographic Image Analysis Society (MIAS). The images are subjected to preprocessing which includes cropping (256 X 256) and enhanced by performing histogram equalization. Then the preprocessed image is subjected to iterative active contour based segmentation . From the segmented image Gray Level Co-occurrence Matrix (GLCM) is formulated. Features are extracted from GLCM. These features are fed to the ANFIS for classification.
Classification using ANFIS
Region Based Active Contour
Figure 1: Block Diagram of the Proposed Mass Segmentation and Classification
The block diagram of the proposed iterative active contour based segmentation and classification is shown in figure 1. The result obtained in various stages of preprocessing are depicted in figure 2.
(a) (b) (c)
Figure 2: (a) Original Image, (b) Cropped Image and (c) Histogram Equalized Image
The preprocessed image and the iterative thresholded image are shown in figure 3 (a) and (b) respectively. The average of the maximum pixel and minimum pixel value is set as initial threshold (T1). The value above threshold is set as foreground and below the threshold value as background. Average value of foreground and background is obtained (T2). If the value of T1 is equal to T2, T1 is the final threshold obtained. If T1 is not equal to T2, then T2 is assigned with T1.
Always on Time
Marked to Standard
Figure 3: (a) Preprocessed Image, (b) Iterative Thresholded Image
The segmentation technique involves initialization of the mask. The square mask of the size 111 x 111 is defined within the mass. This mask is allowed to deform until the minimum energy is encountered. The binary segmented image is obtained. This is compared with the gray scale preprocessed image and then the gray level segmented image is obtained. Figure 4 shows the results obtained in various stages of segmentation.
Figure 4: Image Results Obtained in Various Stages of Mass Segmentation
This method of segmentation is preferred as it holds good when compared to the existing techniques like Fuzzy C Means based segmentation and Level set based segmentation .
From the gray level segmented image shown in Figure 4, GLCM are formulated. Four matrices are formulated for,, and orientations. The average of the above four matrix is formulated as mean GLCM. From the formulated GLCM sixteen statistical features are extracted and are listed in table 1. From these 16 features the optimal features are obtained by calculating the variance and the features with higher value of variance are selected and are shown in table 2. These include Contrast, Dissimilarity, Sum average, Sum variance and Correlation .
Table 1: Feature Selection from GLCM
Feature Extracted from GLCM
Information measure of correlation
Inverse difference normalized
Inverse difference moment normalized
Table 2: Sample Features Extracted
Result and Discussion
The statistical features are extracted from 61 images. Table 3 illustrates the number of images considered for training and testing. Number of training samples considered are 39. For testing 22 samples are considered. Three triangular membership functions are used. The testing error for the triangular membership function is 0.4603 which is lesser when compared with other membership functions (0.5701-trapeziodal, 1.1941-gbell, 0.8864-gaussian, and 1.483-gaussian2). Total number of rules framed is 729. The classification result obtained through ANFIS is shown in the figure 5.
. Testing data
* FIS output
Figure 5: ANFIS-Classification Plot
Table 3: Classification Result of ANFIS
No. of training images
No. of testing images
With the information available from table 3 sensitivity, specificity and overall accuracy of the proposed classification system are calculated using the equations (1), (2), (3) and it is tabulated in table 4.
Sensitivity: X 100 (1)
Specificity: X 100 (2)
Accuracy: X 100 (3)
Table 4: Performance Metrics
The sensitivity obtained is 100%, the specificity obtained is 87.5% and the overall classification accuracy obtained through ANFIS is 91.30%.
In this paper a new computer aided mass segmentation and classification scheme is proposed. Five statistical features were selected for the classification of mass. The classification is done by using ANFIS. The implementation of the proposed method was carried out using MATLAB software. Result shows that the suggested features can give acceptable accuracy for the classification of mass. The result shows that the overall accuracy is 91.30%. With increase in the number of test samples the accuracy may still be improved to a great extent.
This Essay is
a Student's Work
This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.Examples of our work