Abstract
Image segmentation is an important and challenging factor in the medical image segmentation. Thispaper describes segmentation method consisting of two phases. In the first phase, the MRI brain imageis acquired from patients’ database, In that film, artifact and noise are removed after that HSom isapplied for image segmentation. The HSom is the extension of the conventional self organizing mapused to classify the image row by row. In this lowest level of weight vector, a higher value of tumorpixels, computation speed is achieved by the HSom with vector quantization.Key words: Image analysis, segmentation, HSOM, tumor detection.
INTRODUCTION
Brain tumor is one of the major causes for the increase inmortality among children and adults. A tumor is a mass oftissue that grows out of control of the normal forces thatregulates growth (Pal and Pal, 1993). The complex braintumors can be separated into two general categoriesdepending on the tumors origin, their growth pattern andmalignancy. Primary brain tumors are tumors that arisefrom cells in the brain or from the covering of the brain. Asecondary or metastatic brain tumor occurs when cancercells spread to the brain from a primary cancer in anotherpart of the body. Most Research in developed countriesshow that the number of people who develop braintumors and die from them has increased perhaps asmuch as 300 over past three decades.The National Brain Tumor Foundation (NBTF) forresearch in United States estimates that 29,000 people inthe U.S are diagnosed with primary brain tumors eachyear, and nearly 13,000 people die. In children, braintumors are the cause of one quarter of all cancer deaths.The overall annual incidence of primary brain tumors inthe U.S is 11 - 12 per 100,000 people for primarymalignant brain tumors, that rate is 6 - 7 per 1,00,000. Inthe UK, over 4,200 people are diagnosed with a braintumor every year (2007 estimates). There are about 200other types of tumors diagnosed in UK each year. About16 out of every 1,000 cancers diagnosed in the UK are in the brain (or 1.6%). In India, totally 80,271 people areaffected by various types of tumor (2007 estimates).“Artificial Neural Networks (ANNs) are mathematicalanalogues of biological neural systems, in the sense thatthey are made up of a parallel interconnected system ofnodes, called neurons. The parallel action is a differencebetween von Neumann computers and ANNs. CombiningANN architectures with different learning schemes,results in a variety of ANN systems. The proper ANN isobtained by taking into consideration the requirements ofthe specific application, as each ANN topology does notyield satisfactory results in all practical cases. Theevolution of digital computers as well as the developmentof modern theories for learning and informationprocessing led to the emergence of ComputationalIntelligence (CI) engineering. Artificial Neural Networks(ANNs), Genetic Algorithms (GAs) and Fuzzy Logic areCI non-symbolic learning approaches for solvingproblems (Mantzaris et al., 2008). The huge mass ofapplications, which ANNs have been used withsatisfactory results, has supported their rapid growth.Fields that ANNs were used are image processing(Gendy et al., 2001), environmental problems(Bandyopadhyay and Chattopadhyay, 2007;Chattopadhyay and Chattopadhyay, 2009), Climate study(Chattopadhyay, 2007), financial analysis (Papadourakiset al., 1993). In this paper, a new unsupervised learningOptimization algorithm such as SOM is implemented toextract the suspicious region in the Segmentation of MRIBrain tumor. The textural features can be extracted fromthe suspicious region to classify them into benign ormalign.
RELATED WORK
The Segmentation of an image entails the division orseparation of the image into regions of similar attribute.The ultimate aim in a large number of image processingapplications is to extract important features from theimage data, from which a description, interpretation, orunderstanding of the scene can be provided by themachine. The segmentation of brain tumor from magneticresonance images is an important but time-consumingtask performed by medical experts The digital imageprocessing community has developed several segmentationmethods[8], many of them ad hoc. Four of the mostcommon methods are: 1.) amplitude thresholding, 2.)texture segmentation 3.) Template matching, and 4.)region-growing segmentation. It is very important fordetecting tumors, edema and necrotic tissues. Thesetypes of algorithms are used for dividing the brain imagesinto three categories (a) Pixel based (b) Region orTexture Based © Structural based. Several authorssuggested various algorithms for segmentation (Hillips etal., 1995; Aidyanathan et al., 1995; Sai et al., 1995;HanShen et al., 2005; Livier et al., 2005).Suchendra et al. (1997) suggested a multiscale imagesegmentation using a hierarchical self-organizing map; ahigh speed parallel fuzzy c-mean algorithm for braintumor segmentation (Murugavalli and Rajamani, 2006);an improved implementation of brain tumor detectionusing segmentation based on neuro fuzzy technique(Murugavalli and Rajamani, 2007) while Chunyan et al.(2000) designed a method on 3D variationalsegmentation for processes due to the high diversity inappearance of tumor tissue from various patients.
Image acquisition
The development of intra-operative imaging systems hascontributed to improving the course of intracranialneurosurgical procedures. Among these systems, the 0.5T intra-operative magnetic resonance scanner of theKovai Medical Center and Hospital (KMCH, Signa SP,GE Medical Systems) offers the possibility to acquire256*256*58(0.86 mm, 0.86 mm, 2.5 mm) T1 weightedimages with the fast spin echo protocol (TR = 400, TE =16 ms, FOV = 220*220 mm) in 3 min and 40 s. Thequality of every 256*256 slice acquired intra-operatively isfairly similar to images acquired with a 1.5 T conventionalscanner, but the major drawback of the intra-operativeimage is that the slice remains thick (2.5 mm). Images donot show significant distortion, but can suffer fromartifacts due to different factors (surgical instruments,hand movement, radio frequency noise from bipolarLogeswari and Karnan 007coagulation). Recent advances in acquisition protocol(Naylor and Li, 1988) however make it possible to acquireimages with very limited artifacts during the course of aneurosurgical procedure. The choice of the number andfrequency of image acquisitions during the procedureremains an open problem. Indeed, there is a trade-offbetween acquiring more images for accurate guidanceand not increasing the time for imaging.Images of a patient obtained by MRI scan is displayedas an array of pixels (a two dimensional unit based on thematrix size and the field of view) and stored in MATLAB7.0. Here, grayscale or intensity images are displayed ofdefault size 256 × 256. The following figure displayed aMRI brain image obtained in Mat lab 7.0.A grayscaleimage can be specified by giving a large matrix whoseentries are numbers between 0 and 255, with 0corresponding, say, to black, and 255 to white. A blackand white image can also be specified by giving a largematrix with integer entries. The lowest entry correspondsto black, the highest to white. In routine, 21 male andfemale patients were examined. All patients with findingnormal for age n = 20 were included in this study. Theage of patients ranged from 20 - 50 years. All the MRIexaminations were performed on a 1.5 T magneto visionscanner (Germany). The brain MR images are stored inthe database in JPEG format.


Download full report
http://academicjournalsjcreo/PDF/Pdf2010/Mar/Karnan%20and%20Logeswari.pdf