Iaetsd early detection of breast cancer

Page 1

INTERNATIONAL ASSOCIATION OF ENGINEERING & TECHNOLOGY

ISBN NO : 378 - 26 - 138420 - 8

This calls in the need to apply engineering principles in the bio-medical field. Computer aided diagnosis (CAD) systems are automated systems which are capable of identifying the tumor cells from the mammograms without the intervention of radiologists or oncologists. [6] The mammograms are analyzed for the presence of malignant masses, thickening of the tissue and deposits of calcium known as microcalcification.[7] The masses often occur in dense areas of the breast and have different shapes and spotted by the bright spots in the mammograms. The micro-calcification is generally identified in the classifier stage as either benign or malignant. Malignant micro- calcification generally has a diameter of less than 0.5 mm and is fine and stellar structured.

EARLY DETECTION OF BREAST CANCER USING CAD SYSTEM EMPLOYING SVM CLASSIFIER Authors S.DURGALAKSHMI ;nan_naned@yahoo.co.in, SHREYA REDDY ;shreyareddy.7676@gmail.com, Fourth year Panimalar Institute of Technology

ii.

DETECTION USING CAD SYSTEM:

ABSTRACT: Mammography is a low radiation process used to detect breast cancer. It is of two types – screening mammography and diagnostic mammography. Screening mammography is the procedure used to detect cancer in asymptotic population while diagnostic is used to analyze the patients with abnormal conditions. Usually diagnostic is used as a follow up to the screening method in abnormal patients. Interpretation of the mammography can be difficult due to the poor contrast and less difference between a healthy and abnormal breast. Due to which high rate of false positives and false negatives are seen owing to some women undergoing surgery unnecessarily. To increase the efficiency and accuracy, image analysis and classification are done with the help of a CAD system.[8][9]

Breast cancer is the most common cancer pathology detected in women. It is the second leading cause for morbidity and high mortality rates. The cause of the cancer is still unknown thus early detection of it is very important. This paves way to the need for early detection systems for breast cancer. Currently the most prevalent system of detection is mammogram. But often detection from the mammogram is not accurate which calls for a second opinion which is a costly affair. This paper presents a computer aided diagnosis system for early detection of the cancerous tissue. It employs an efficient image enhancement system using and classification using the support vector machine. KEYWORDS: mammogram, microcalcification, CADsystem, breast cancer. i.

The CAD process mainly uses 3 distinct phases which are the (1) image processing, (2) sedimentation and (3) classification. The first step is where the mammogram is processed and the contrast and quality of the image is enhanced for easy identification. In the second step, the separation of the suspicious cells from the background parenchyma is done. While in the third step the classification of the cells as either benign or malignant is done.

INTRODUCTION:

Breast cancer is the abnormal growth of cancerous cells in the mammary glands of the woman. It is reported that 1 in 12 woman are affected with breast cancer. The symptoms of which are formation of lumps, lesions, or abnormal skin texture or areolar swelling and change in size of the breasts.[1]The detection of these is often difficult as the density and shape of the breasts change from woman to women. This is done with the help of screening using a low radiation this technique is known as mammography. [2] Most common diagnosis of the cancerous growth is the biopsy, which is an invasive procedure by which a piece of tissue or a sample of cells is removed from your body so that it can be analyzed. [3]The drawback of such an approach is a high number of non-productive biopsy examinations with high economic costs. Statistics show that only 15-34% of breast biopsies are proved cancerous and that 10-30% of all cases of breast cancer goes undetected by mammography.[4] [5]

iii. IMAGE PROCESSING: The image processing phase is further split into steps as shown in the figure 1. [10][11]

Figure 1.Image processing phases. The original mammogram is taken as the input image to the CAD system in which it is passed through the Gaussian

International Conference on Advancements in Engineering Research 58

www.iaetsd.in


INTERNATIONAL ASSOCIATION OF ENGINEERING & TECHNOLOGY

ISBN NO : 378 - 26 - 138420 - 8

soothing filter. The filter smoothens the image by removing noise and is used to blur the image to obtain a smooth gray scale.[12] In the contrast stretching step, which is also known as the full scale histogram stretching, the image is stretched over the entire gray scale area. A well distributed is one which has a good visibility and high contrast. [13] Then the top hat operation is performed on it. It is a morphological method using mathematical operations. It divides the image into two sets – the object and the structuring elements. The top hat performs the difference of the original image and its opening points which are the collection of the foreground points. This highlights the bright spots in the image.[14]The DWT reconstructs the decomposed image back. Thus a reconstructed image is available at the end of the image processing by the CAD.

iv. SEGMENTATION:

Figure 5.Process from segmentation to classification In the segmentation phase in the CAD, the suspect cells are split from the background image using global and local thresholding as micro-calcification look brighter than the surrounding. In global thresholding is a technique in which the histogram reveals peaks corresponding to the background and the object. In the case of micro-calcification extra peeks are raised. Local thresholding is used to refine results obtained in the previous stage as the pixels need to have an enhanced intensity values. This is followed by feature extraction where the abnormalities are detected and extracted. The selection process in the CAD involves an optimum set of features from those available in the image after segmentation. The feature space can be divided into subspace based on intensity and texture. The scale space method is used to extract the abnormalities; it is based on Laplacian scale-space representation.[15] The possible micro-calcification are identified by local maxima on a range of scales in the filtered image. For finding the size, contrast, the response is used. The finding is demarcated as a micro-calcification if the contrast exceeds a predefined threshold value.

Figure 2.a)original b)filtered image

v. SUPPORT VECTOR CLASSIFIER: Support vector machine is a learning model with algorithm for analyzing data and recognizing patterns in them. It classifies the inputs given to it based on the pattern recognized. It is based on a decision plane model. [16]

Figure 3.a)original b)enhanced image

The SVM divides the inputs into two classes and builds a model which assigns the value to either one of the classes. It is a mapping of the values as points in space such that they can be divided into two distinct set by a wide margin between them. They points are identified to belong to a category based on which side they fall on. The SVM intakes a test set data which its interprets and a data set which it categories.[17] The SVM does linear classification using hyper planes while non-linear cluster classification using kernel functions. (1)LINEAR METHOD: Figure 4 a) reconstructed mammogram b) segmented output

International Conference on Advancements in Engineering Research 59

www.iaetsd.in


INTERNATIONAL ASSOCIATION OF ENGINEERING & TECHNOLOGY

ISBN NO : 378 - 26 - 138420 - 8

In the linear method, the inputs are clearly spread out in a linear manner and thus finding a plane that separates them properly is the only concern.

Figure 8.thehyperplane

The margin on one side is 1 / |w|, thus the total margin is 2*(1/|w|). Thus minimizing this |w| values will increase the separability.

Figure 6.a hyper plane dividing two data sets The decision plane separates the assets of different classes. For proper classification, the margin of the decision plane from the class objects must be greater. This would allow a better categorizing of the samples.

The width can be increased by the use of Lagrange multipliers. Using karush-kush-tucker conditions,

The best hyper plane is thus one with maximum margin from both the classes.

Applying this in the width equation, the width of the plane will be the dot product of the data objects. L= xi .xj where xi and xj are data objects. (ii) NON LINEAR METHOD: For the nonlinear data set, the classification cannot be done using linear SVM. This requires the special kernel function. In this case, the SVM introduces an extra set of dimension where the points can be projected on to in a better space.

Figure 7.Twohyper planes with varying margins

This space is calculated using the kernel functions There are number of kernels that can be used in SVM models. These include linear Polynomial, RBF and sigmoid:    

xi*xj - linear (γxixj+coeff) - polynomial Exp (-γ|xi-xj|2) - RBF Tanh (γxixj+coeff) -sigmoid

The Radial Base Function (RBF) kernel is the most suitable method with SVM. [18]

International Conference on Advancements in Engineering Research 60

www.iaetsd.in


INTERNATIONAL ASSOCIATION OF ENGINEERING & TECHNOLOGY

ISBN NO : 378 - 26 - 138420 - 8

For Gaussian radial basis function: K (x, x’) =exp (-|x-x’|2/ (2σ2)).

[7] M. Rizzi, M. D’Aloia and B. Castagnolo, “A fully automatic system for detection of breast microcalcification clusters,” J. Med. Biol. Eng., 30: 181-188, 2010.

vi. RESULT: [8] L.J.W. Burhenne,”potential contribution of computer aideddetection to the sensitivity of screening mammography”,Radiology, vol.215, pp.554-562, 2000.

The proposed method involves the mammogram image being filtered by a Gaussian filter based on standard deviations and the resultant image is context stretched. The unwanted background is removed using top hat approach. The output is decomposed into two scales which are reconstructed back using DWT. The image is then segmented and the features from the selected area are extracted in the fragment extraction phase and then the tumor classification is done using the SVM classifier.

[9] T.W.Freer and M.J.Ulissey, “Screening mammography with computer aided detection: prospective study of 2860 patients in a community breast cancer”, Radiology, vol.220, pp.781-786, 2001. [10]Howard Jay Siegel, Leah.J, “PASM: A Partitionable SIMD/MIMD System for Image Processing and Pattern Recognition” IEEE trans. [11]SumanThapar, ShevaniGarg, “Study and implementation of various morphology based image constrast enhancement techniques”, International Journal of Computing & Business Research

vii. CONCLUSION: The computer aided diagnostic system has tremendously helped the radiologists as they do not need a second opinion from doctors. Moreover the hit ratio is comparatively higher. This reduces the negative reports about the existence of breast cancer and helps identify cancer in early stages. Thus the treatment process can be more effective. It is also reported that the linear method of classification proves more efficient because of the linearity in the data set. The classifier was able to vividly demarcate the different cancer cells as benign or as malignant.

[12]Ahmed Elgammal, “Digital Imaging and Multimedia Filters”,Rutgers [13]Kumar, jagatheswari, “Contrast Stretching Recursively Separated Histogram Equalization for Brightness Preservation and Contrast Enhancement”,Advances in Computing, Control, & Telecommunication Technologies, 2009. ACT '09. International Conference [14]Jiang Duan, Chengdu, Wenpeng Dong, “Computational Intelligence for Multimedia, Signal and Vision Processing (CIMSIVP)”, 2011 IEEE Symposium

REFERENCES:

[15] T. Netsch and H. O. Peitgen, “Scale-space signatures for the detection of clustered microcalcifications in digital mammograms,” IEEE Trans.

[1]Mammography[Online] http://www.radiologyinfo.org/en/pdf/mammo.pdf [2] American college of radiology, Reston VA, Illustrated Breast imaging Reporting and Data system (BI-RADSTM) , third edition, 1998.

[16] C. Cortes, V. N. Vapnik, “Support vector networks”, Machine learning Boston, vol.3, Pg.273-297, September 1995

[3]E.C.Fear, P.M.Meaney, and M.A.Stuchly,”Microwaves forbreast cancer detection”, IEEE potentials, vol.22, pp.1218,February-March 2003.

[17]Zhang Xinfeng, ZhaoYan “Application of Support Vector Machine to Reliability, telkomnika ,2014 [18]JooSeuk Kim, Ann Arbor, ”Performance analysis for L2 kernel classification”

[4] Homer MJ. “Mammographic Interpretation: A practical Approach”, McGraw hill, Boston, MA, second edition, 1997. [5]Maria Rizzi*MatteoD’AloiaBeniaminoCastagnolo, “Review: Health Care CAD Systems for Breast Microcalcification Cluster Detection”,Journal of Medical and Biological Engineering,2012. [6] S.M.Astley,”Computer –based detection and prompting of mammographic abnormalities”, Br.J.Radiol, vol.77, pp.S194S200, 2004.

International Conference on Advancements in Engineering Research 61

www.iaetsd.in


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.