Published on

  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide


  1. 1. R.Ramani, Dr.S.Suthanthiravanitha, S.Valarmathy / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 2, Issue 5, September- October 2012, pp.1124-1129 A Survey Of Current Image Segmentation Techniques For Detection Of Breast Cancer R.Ramani*, Dr.S.Suthanthiravanitha**, S.Valarmathy*** * (Assistant Professor, Vmkv engineering college) ** (Professor/EEE, Knowledge institute of technology) *** (Assistant Professor, Vmkv engineering college)AbstractBreast cancer is one of the most common forms radiologists can miss the abnormality easily if theyof cancer for women. Accuracy rate of breast only diagnose by experiences. Computed aidedcancer in mammograms depend on the detection technology can help doctors andsegmentation of images. Mammography is radiologists getting a more reliable and effectiveespecially used only in the breast tumor detection diagnoses there are numerous tumor detectionMammogram breast cancer images have the techniques have been analyzed [4-5]. The key taskability to assist physicians in detecting disease in designing such image processing and computercaused by cells normal growth. The goal of vision application is the accurate segmentation ofsegmentation is to simplify and or change the medical images. Image segmentation is the processrepresentation of an image into something that is of portioning different regions of the image basedmore meaningful and easier to analyze, but the on different criteria[6].breast cancer imagesame time image segmentation is very difficult segmentation from mammography images isjob in the image processing. The main aim of this complicated and challenging but its precise andpaper is to review existing approaches of exact segmentation is necessary for tumor detectionspreprocessing and current segmentation and their classification of tissues for early detectiontechniques in mammographic images. The of abnormalities in breast. Mammography imagingobjective of preprocessing is to improve the is most efficient imaging techniques.image quality, removing the irrelevant noises and Mammography is highly accurate but like mostunwanted parts in the background of the medical tests, it is not perfect [7].on averagemammogram. There are different types of mammography will detect above 80% - 90% of thesegmentation algorithms for mammogram image. breast cancer in women without symptoms. ToTheir advantages and disadvantages are accurate segmentation of breast images challenge,discussed. however, accurate image segmentation of the mammography images is very important and crucialKeywords: Breast Cancer, Mammogram, for the exact diagnosis by computer aided clinicalPreprocessing, Segmentation algorithms tools lot of Varity of image segmentation algorithms have been developed for mammography images. InI.INTRODUCTION this paper we present a review of the methods used The many of research works conducted in in mammography image segmentations. The reviewthe area of breast cancer detection and classification covers mammography images and noise reductionmuch university, commercial institution and and image segmentation approaches. The paperresearch centers are focused on this issue because of concludes with a discussion on the upcoming trendthe fact that breast cancer is becoming the most of advanced researches in breast cancer imagecommon form of cancer disease of today’s female segmentation.population. Breast cancer is the second mostcommon cancer in Indian women. in the united II LITERATURE REVIEWstates alone a most recent survey estimated 2, Segmentation is the process of partitions an07,090 new cases of breast cancer and 39,840 death image to several small segments the mainin women during 2010[1].the average incidence difficulties in image segmentation are, noise, biasrates varies from 22-28 per 10,00,000 women per field, partial volume effect (a voxel contributes inyear in urban setting to 6 per 1,00,000 women per multiple tissue types)year in rural areas’ X-ray mammography is the mostcommon investigation techniques used by A.De-noising methods [Decreases the noise]radiologists in interpretation of the mammograms. It In image pre-processing techniques areis necessary to detect the presence or absence of necessary in order to find the orientation of thelesions from the mammograms [2-3].however the mammogram to remove the noise and to enhanceappearance of breast cancer is very subtle and the quality of the image[8].the pre-processing stepsunstable in their early stages. Therefore doctors and are very important in order to limit the search for 1124 | P a g e
  2. 2. R.Ramani, Dr.S.Suthanthiravanitha, S.Valarmathy / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 2, Issue 5, September- October 2012, pp.1124-1129abnormalities without undue influence from back maximum or minimum value in a neighborhoodground of the mammograms. The main objective of around that pixel. The neighborhood stands for thethis process is to improve the quality of the image to shape of the filter, maximum and minimum filtersmake it ready to further processing by removing the have been used in contrast enhancement.unrelated and surplus parts in the back ground of themammograms [9]. B. IMAGE SEGMENTATION The main objective of image segmentation1. Adaptive median filter is to extract various features of the images which Adaptive median filter works on a can be merged or split in order to build objects ofrectangular region Pxy, it changes the size of Pxy interest on which analysis and interpretation can beduring the filtering operation depending on certain performed. Image segmentation refers to the processconditions such as of partitioning an image into groups of pixels whichZmin = minimum pixel value in Pxy are homogeneous with respect to some criterion.Zmax = maximum pixel value in Pxy The result of segmentation is the splitting up of theZmed = median pixel value in Pxy image into connected areas. Thus segment isPmax = maximum allowed size of Pxy concerned with dividing an image into meaningfulEach output contains the median value in 3 by 3 regions. The image segmentation techniques such asneighborhoods around the corresponding pixel in thresholding, region growing, statistics models,the input images. The edges of the image however active control modes and clustering have been usedare replaced by zeros [10].adaptive median filter has for image segmentation because of the complexbeen found to smooth the non repulsive noise from intensity distribution in medical images,2D signals without blurring edges and preserve thresholding becomes a difficult task and often failsimage details. This is particularly suitable for [14-15].enhancing mammograms images. 1. Region growing segmentation2. Mean filter Region growing is an approach to image The mean filter replaces each pixel by the segmentation in which neighboring pixels areaverage value of the intensities in its neighborhood. examined and added to a region class if no edges areIt can localy reduce the variance and is easy to detected. This process is iterated for each boundaryimplement [11]. pixel in the region. If adjacent regions are found, a region merging algorithms is used in which weak3. A markov random field method edges are dissolved and strong edges are left intact. In this method spatial correlation The region growing starts with a seed which isinformation is used to preserve fine details [12].in selected in the centre of the tumor region. Duringthis method regularization of the noise estimation is the region growing phase, pixels in the neighbor ofperformed. The updating of pixel value is done by seed are added to region based on homogeneityiterated conditioned modes. criteria thereby resulting in a connected region [15].4. Wavelet methods 2. Random walk method In frequency domain these method is used The random walk method is used tofor de-noising and preserving the signal application segment the breast tissues for detection of cancerousof wavelet based methods on mammography image cells. Random walk is defined as discrete randommakes the wavelet and scaling coefficient biased. motion in which a particle repeatedly moves a fixedThis problem can be solved by squaring distance up, down, east, south and north. This is amammograms images by non central chi-square region growing based image segmentation methoddistribution method [13]. based on random walk of a particle. In this method the initial position at which a particle is initially 5. Median filtering present is known as seed point movement from one A median filter is a non linear filter is position to another is based on the probabilityefficient in removing salt and pepper noise median calculation. After the seed point has been detectedtends to preserve the sharpness of image edges random walk method is to be performed forwhile removing noise. The various of median filter segmentation and ten fine segmented [16].are i) centre-weighted median filter ii) weightedmedian filter iii) max-median filter, the effect of 3. Watershed algorithmincreasing the size of the window in median filtering The watershed is a powerful tool for imagenoise is removed effectively. segmentation in watershed the image is considered as a topographic surface. The watershed algorithms6. Max-Min filter have been developed and tested on Varity of Maximum and minimum filter attribute to mammogram breast cancer images [17]. It has beeneach pixel in an image a new value equal to the found that the result of segmentation gives very 1125 | P a g e
  3. 3. R.Ramani, Dr.S.Suthanthiravanitha, S.Valarmathy / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 2, Issue 5, September- October 2012, pp.1124-1129good suggestion to a radiologist and doctors to 5. Wavelet based adaptive windowing methodfurther investigate on the presence of micro The wavelet based adaptive windowingcalcifications in the breast tissue. Various steps method is used for the segmentation of bright targetsinvolved in watershed algorithms such as in an image. In these method two types ofInput image. segmentation is used for mammogram to detectInput image converted to gray scale image. tumor. Coarse segmentation is implemented byFilter is used for noise removed. using wavelet based histogram thresholding, whereMedian filter is used to enhance the quality of the threshold value is chosen by performing inimage. wavelet based analysis of pdf of waveletApply threshold segmentation. transformed images at different channels and secondApply watershed segmentation. one is fine segmentation which is obtained byMorphological operation. choosing threshold by using windowing method.Finally output will be a tumor region. The wavelet based adaptive windowing method is effective to segment the tumor in mammograms and4. Adaptive mean shift algorithm it can also be used in other segmentation Adaptive mean shift algorithm is obtained applications. This method of segmentation yieldfrom mean shift clustering mean shift algorithm. significantly superior image quality when it’sThe compared to the global threshold method andBasic mean shift clustering algorithm maintains a window based adaptive thresholding method [19].set of data points the same size as the input data set.A mean shift algorithm that is similar then to k- 6. K-Means clustering method.means called likelihood mean shift, replaces the set The k-means algorithms are an iterativeof points undergoing replacement by the mean of all technique that is used to partition an image into k-points in the input set that are within a given cluster. In statistics and machine learning, k-meansdistance of changing set. One of the advantages of clustering is a method of cluster analysis which canmean shift over k-means is the there is no need to to portions n observation into k cluster in whichchoose the number of cluster, because mean shift is each observation belongs to the cluster with thelikely to find only a few clusters if indeed only a nearest mean [20-21]. The basic algorithms is givensmall number exist. However, mean shift can be belowmuch slower than K means. Mean shift has soft - Pick k cluster centre’s either randomly orvariants much as K-means does. based on some heuristic. - Assign each pixel in the image to the cluster that minimum the distance between the pixels cluster centre. - Re-compute the cluster centre’s by averaging all of the pixels in the cluster. --- Repeat last two steps until convergences are1 attained. The most common algorithm uses anThis is the difference between the weighted mean iterative refinement technique; due to this ambiguityand x, known as mean shift vector. Since the it is often called the k-means algorithms.gradient of the density estimator always pointstowards that direction in which the density rises 7. Fuzzy c-meansmost quickly, from the above Equation the mean The fuzzy c means algorithms also knownshift vector always points towards the direction in as fuzzy ISODATA is one of the most frequentlywhich the density rises most quickly. This is the used methods in pattern recognition fuzzy c-meansmain principle of mean shift-based clustering. This is a method of clustering which allows one piece ofequation is generalized into data to belong to the two or more cluster[22]. It is based on the minimization of objective function to achieve a good classification. J is a squared error clustering criterion and solutions of minimization are least squared error stationary point of J in the following equation ---2This is referred to as adaptive mean shift vector. Theadaptive mean shift vector always points towardsthe direction in which the density rises most quickly,which is called the mean property. This is the basicprinciple of adaptive means shift based clustering ---3[18]. 1126 | P a g e
  4. 4. R.Ramani, Dr.S.Suthanthiravanitha, S.Valarmathy / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 2, Issue 5, September- October 2012, pp.1124-1129Fuzzy partitioning is carried out through an iterative 9. Segmentation by Morphological Algorithmoptimization of the objective. The clustering method Mathematical morphology is used as a toolfor both k means algorithms and fuzzy c means for extracting image components such as boundariesalgorithms is same but in k means algorithm when it in image segmentation. Since language ofcluster, it takes the mean of the weighted cluster so mathematical morphology is set theory, thisas easy to identify masses or the origin point of segmentation approach is based on binary image.cancer and tumor. Similarly in FCM, it considers This algorithm includes two major steps:that each point has weighted value associated with preprocessing and segmentation. Thresholding iscluster. Doing this were able to find out how much used to convert input image into binary image. Sincethe cancer has spread out, this is helped to us to find tumor tissue tends to have maximum intensity inout the stages of breast cancer[23]. mammograms, normally closed to 1 in gray level, a global threshold could serve as the first cut in the8. Vector Quantization process and convert the image into binary image. The main objective of the vector Dilation and erosion are two basic morphologicalquantization segmentation method to detect operations defined by equation (2) and (3)cancerous mass from MRI images. In order to respectively [26].increase radiologist’s diagnostic performance,computer-aided diagnosis (CAD) scheme have beendeveloped to improve the detection of primarysignatures of this disease: masses and microcalcifications. During the past many researchers in The simplest way to realize boundary extraction of athe field of medical imaging and soft computing binary image A is given by equation (4) [26],have made significant survey in the field of imagesegmentation. Image segmentation techniques canbe classified as based on edge detection, region or where B is a suitable structuring element. However,surface growing, threshold level, classifier such as there could be noise present by this method, insteadHierarchical Self Organizing Map (HSOM), and of using original binary image A; dilation of A, thatfeature vector clustering or vector quantization. is, A⊕ could be BVector quantization has proved to be a very used. Equation (5) is the resulting edge detectioneffective model for image segmentation process. formula.Vector quantization is a process of portioning an n-dimensional vector space into M regions so as tooptimize a criterion function when all the points ineach region are approximated by the representationvector associated with that region. There are two 10. Level Set Modelprocesses involved in the vector quantization: one is Many of the PDEs used in imagethe training process which determines the set of processing are based on moving curves and surfacescodebook vector according to the probability of the with curvature based velocities. In this area, theinput data, the other is the encoding process which level set method developed by Osher andassigns input vectors to the code book vectors [24]. Sethian[27] was very influential and useful . TheTumors or calcifications are embedded in an basic idea is to represent the curves or surfaces asinhomogeneous background. In mammograms, the zero level set of a higher dimensional hyperbackground objects may even appear brighter. surface. This technique not only provides moreTherefore, global threshold methods suffer accurateconsiderable drawback. Vector quantization Numerical implementations but also handlesegmentation algorithm attempts to overcome such topological change very easily. It has severaldrawbacks. Vector quantization is based on advantages; its stability and irrelevancy withclustering algorithm. Clustering is the process of topology, displays a great advantage to solve thegrouping a data set in a way that the similarity problems of corner point producing, curve breakingbetween data within a cluster is maximized while and combining etc. Since the edge-stopping functionthe similarity between data of different clusters is depends on the image gradient, only objects withmaximized and is used for pattern recognition in edges defined by gradients can be segmented.image processing. Quantization methods are used Another disadvantage is that in practice, the edge-for tumor detection in MRI images. From results it stopping function is never exactly zero at the edges,is observed that proposed method gives far better and so the curve may eventually pass through objectresults compared to morphological Segmentation. boundaries [28]The Algorithms are Identification rate for vectorquantization method is 71.5% 11. Multiobjective Image Segmentationand for Morphological Segmentation Identification Earlier image segmentation problem hasRate is 66.7%.[25] been treated as mono objective. Mono-objective 1127 | P a g e
  5. 5. R.Ramani, Dr.S.Suthanthiravanitha, S.Valarmathy / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 2, Issue 5, September- October 2012, pp.1124-1129images considers only one objective, because a REFERENCESsingle segmentation image. Such type of segmented [1]. The American college of radiology (ACR),images are of good quality but may not allow a level process (as image segmentation [2]. Stems .e.e(1996) can j. sung. Vol 33-considered as low level process & pattern pp128-132recognition, object tracking & scene analysis as high [3]. Highmam R and Brady M(1999)level process ) to extract all information included mammogaraphic image analysis,klworerswithin the image. So different segmentation results academic publishers,isbn:0-7923-5620-9.are calculated. Image segmentation is a multi [4]. Sheng zhow Xu and hong Liu&Enminobjective optimization problem [29]. The song “marker-controlled watershed forconsideration of multiple criteria (objectives) starts lesion segmentation in mammograms”.Jfrom the understanding of image pattern to its digital imaging 24:754-763,2011selected segmentation process involved (feature [5]. M.Grgic et al.[eds];recc Advan.inselection/extraction, similarity/ dissimilarity mult.sig.processing and communicationmeasure) and finally the assessment of its output sci231,pp631-657,2009(validity assessment). As there are possibilities of [6]. chang PL .Teng wg(2007) exploting themultiple sources of information for a segmentation self organiszing map for medical imageproblem, thus multiple representations have to be segmentation,in,twentieth IEEEconsidered, for example, feature selection is the international sysmposium on computerprocess of identifying similarity criteria used in based medical systems.pp281-288segmentation process, now either only single criteria [7]. Guido .M.te Brate and Nicois used, that is intensity of pixels, or to make it a Karssemrjes,single and multiscalemulti objective problem consider several similarity detection of masses in digitalcriteria to segment same image, which can be mammograms,IEEE Transations onintensity, color, texture, shape, spatial information. Medical imaging Vol.18,noFor instance, in segmenting a medical image based 7,july1999,628-638.on CT scan, multiple features like intensity, shape [8]. Samir kumar bandyopathyay,pre-and spatial processing of mammogaramRelationship could be considered. Similarly criteria images,international journal of engineeringfor inter pattern similarity that is grouping can be science andmultiple, spatial coherence vs. feature homogeneity, technology,vol.2(11),2010,6753-6758.connectedness vs. compactness, diversity vs. [9]. D. narvain ponraj,m.evangelinaccuracy. For image segmentation multiple methods jenifer,p.poongodi,j.samuvel manoharan.acan be used for getting appropriate output, and there survey on the pre-processing techniques ofmay be a tendency for multiple optimizations and mammogram for the detection of breastdecision making processes where multiple validity cancer, journal of emerging trends inassessment should be used.[28] There are two computing and information sciences.volgeneral approaches for Multi objective optimization 2,no:12,December 2011.problem, the first approach is to combine multiple [10]. Jwad nagi automated breast profileobjective functions into a single composite function, segmentation for ROI detection usingand the second is to determine a set of solutions that digital mammogram IEEE EMBSare non-dominated with respect to each objective. confrerence on bio-medic al engineering and sciences(Iecbes2010),kuala lumpur.III. CONCLUSION [11]. Junn shan wen ju,yanhui, guoa, ling zhang, Image segmentation is one of the most h.d.cheng. automated breast cancerchallenging and active research areas in the field of detection and classification usingmedical image processing. Mammogram Image ultrasound images-a survey. Patternsegmentation is a challenging task and there is a recognition 43,(2010) 299-317need and huge scope for future research to improve [12]. an d(1984) stochastic relaxation gibbsthe accuracy, precision and speed of segmentation distribution and the Bayesian restoration ofmethods. Thus there is no single method which can images,IEEE trans pattern and mach intelbe considered good for neither all type of images, 721-741.nor all methods equally good for a particular type of [13]. nowak rd (1999) wavelet based ricianimage. Due to all above factors, image segmentation noise removal for MRI images,IEEE transremains a challenging problem in image processing image processing.8(10);1408-1419.and computer vision and is still a pending problem [14]. Suzuki h.torkakij (1991) automaticin the world. Further works may be conducted to segmentation of head MRI images bydevelop efficient segmentation methods. knowledge guided thresholding,computer medical imaging graphic15(4);233. 1128 | P a g e
  6. 6. R.Ramani, Dr.S.Suthanthiravanitha, S.Valarmathy / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 Vol. 2, Issue 5, September- October 2012, pp.1124-1129 [15]. Dr.samir kumar bandhyopathyway,tuhin image processing, Prentice Hall, 2002. utsab paul,segmentation of brian MRI [27]. X. Jiang, R. Zhang, S. Nie,“Image imges research in computer science and Segmentation Based on PDEs Model: a software engineering,volume 2,issue Survey”, IEEE conference, pp. 1-4, 2009. 3,march2012,issn;2277 128x. [28]. Rajeshwar Dass, Priyanka, Swapna Devi,[16]. Mona choubey,suyash agarwal, an Image Segmentation Techniques, implementation of randow walk algorithms International Journal of Electronics & to detect Brain cancer in 2d MRtionI , Communication Technology(IJECT) Vol. international journal of engineering 3, Issue 1, Jan. - March 2012 ISSN : 2230- research and applications. Vol 2,issue 7109 (Online) | ISSN : 2230-9543 (Print) 1,jan-feb 2012,pp998-1001, [29] B. C. Wei, R. Mandava,“Multiobjective[17]. Vincent L and soiled P(1996) IEEE trans optimization Approaches in Image on machine intel vol 18,no 12,pp.1163- Segmentation- the Direction and 1173. Challenges”, ICSRS Publication, 2010.[18]. M.C. jobin Christ,Dr.R.M.Parvathi,an adaptive mean-shift algorithm for MRI BRAIN segmentation. International journal of engineering science and technologies.issn;0975-5462, vol 3,no 7, july 2011[19]. G.bharatha sreeja, Dr.P.Rathika, Dr.D.Devaraj, “detection of tumors in digital mamoograms using wavelet based adaptive windowing method”,international journal modern education and computer science,2012, 3,57-65.[20]. T. Kanungo. D.M Mount ,N.Netanyahu,C.Piatko,R.Silverman and A.Y.wa(2002),an efficient k-means clustering algorithms,analysis and implementation proc.IEEE conference computer vision and pattern recognition pp.881-892.[21]. Bhagwati charanpatel, Dr.G.R.sinha, an adaptive k-means cluster algorithms for breast image segmentation, international journal of computer applications(0975- 8887),vol 10-n 4 ,nov-2010[22]. Thangavel. K. mohideran,a. k ,dept.of comp.sci,periyar university,salem,india. “ semi-supervised k means clustering for outlines detection in mammogram classification” appears in treandz in information science and computing(2010)[23]. An edge detection method for microcalcification cluster in mammogram by yu guang zhang wen Lu Fu yun cheng LI song taishan med.univ.taian,china, in second international conference on bio- medical engineering and informatics 2009[24] Donald Mcrobbie. Medical Image Segmentation [Online]. Available: http://[25] Dr. H. B. Kekre et. al, Detection of tumor in mri using vector quantization Segmentation, International Journal of Engineering Science and Technology, Vol. 2(8), 2010, 3753-3757[26] Rafael C. Gonzalez and Richard E. Woods, Digital 1129 | P a g e