# INTRODUCTION n Neonatal Brain MRI Segmentation By Building Multi -Region-Multi -Reference Atlases, Brain tissue segmentation, which classifies brain tissues into meaningful structure such as gray matter (GM), white matter (WM), and then cerebrospinal fluid (CSF).The segmentation is performed in this structure is difficult in neonatal brain image due to low spatial resolution ,insufficient tissue contrast ,and ambiguous tissue intensity distribution. Due to these problem difficulties image intensity is insufficient for effective neonatal brain MRI segmentation. The knowledge-based algorithm is seems to be effective. The atlas is build with multiple individual atlases with decision fusion strategies. The strategy implies that individual atlas-based segmentation fuse the segmentation into final result. Prastawa constructed an atlas by averaging 3 semiautomatic segmented neonatal image are alignment using affined transformation. Weisenfeld obtained an unbiased atlas by averaging the probability maps of 20 newborn subjects. Which are non-rigidly aligned with a simultaneous group-wise registration .a multi-region -About ? -M.phil Scholar, P.S.G.R. Krishnammal College for Women, Coimbatore. Email id: janumphil@gmail.com About ? -Assistant Professor BSc Dept., P.S.G.R. Krishnammal College for Women, Coimbatore. multi-reference approach, which estimates multiple atlases for different anatomical regions. Subject specific atlas is constructed for more effective neonatal segmentation. There are two issues, taking brain as a single entity assign weight globally to all vowels so that a local shape patterns in the brain will be desired. The parcellation is performed to separate the brain into multiple sub regions. So that atlas can be build for each region separately. A cluster technique called affinity propagation is used to cluster the images. In In atlas based image segmentation algorithms, the segmentation performance is affected by the registration procedure. The image acquired at late time brain image such as two-years old can achieve high accuracy using the existing segmentation method like fuzzy clustering. The proposed method is to use latetime point image in conjunction with its segmentation result as subject-specific tissue probabilistic atlas to guide tissue segmentation of neonatal image. The subject-specific atlas can be used within a jointregistration-segmentation. In Construction of Multi-Region-Multi-Reference Atlases For Neonatal Brain MRI Segmentation, Atlas can be grouped into two categories1) average -shape atlas method (prastawa et.,2005;song et al.,2007;xue et., 2007)2)multi-classifier decision fusion methods, multi subjects in a population are selected as individual atlases to independently guide segmentation. Single atlas may not sufficiently characterize shape variation in a population; the atlas-based segmentation approach has the drawbacks. The brain is taken as a single entity, different brain images regions have different anatomical pattern as region-wise comparison approach may be more appropriate. A single average shape atlas is generating from a population, it is better to construct multiple atlases. To I overcome these 2 issues a method for each query image a subject specification is accommodated to the structural shapes of the query image. First the averageshape atlas of a population images is divided into multiple regions. Each sub-population is represented an exemplar and each its regions is represented by multiple exemplars. Collection of regional exemplar is called multi-region-multi-reference atlas. A query image, one best match exemplars is selected for each region and the selected exemplars for all regions are combined to form the final subject-specific atlas. A jointregistration-segmentation strategy is finally used to segment the query image. Experiment result indicates that, in significant segmentation accuracy improvement can be achieved. # II. # DECISION FUSION METHOD Detecting edges in each image separately and then fusing the results is called decision fusion method. # a) Process of Decision Fusion In neonatal brain MRI segmentation by building multi -region-multi -reference atlases, To build the atlas as prior knowledge and to aid segmentation three strategies are commonly used 1) single individual atlas 2) average-shape atlas 3) multiple individual atlases with decision fusion. The category 3 implies that the individual-atlas-based segmentation multiple times with different atlas subject and then fuse the multiple image segmentation into a final result with majority voting rule. It is to be noted that computation cost is quite high due to multiple segmentation. In neonatal brain image segmentation in longitudinal MRI studies, The decision fusion is widely used to combine multiple segmentation into final decision with compensation for errors in single segmentation (Heckerman et al., 2006;Warfield et al., 2004). Decision fusion technique could be used to achieve better neonatal segmentation. The concept of decision fusion is used to fuse the multiple image segmentation into a single segmentation, with the single segmentation the neonatal brain MRI can be segmented easily. The need of decision fusion is to fuse multiple image segmentation and to get the final result. In Construction of multi-region-multi-reference atlases for neonatal brain MRI segmentation, Atlas construction methods can be roughly grouped into two categories1) average-shape atlas methods 2) multiclassifiers decision fusion methods. In multi-classifier decision fusion methods, multiple subjects in a population are selected as individual atlases to independently guide segmentation. All segmentation results from different atlases can then fused by a majority-voting rule. # III. # METHODOLOGY In Neonatal Brain MRI Segmentation By Building Multi -Region-Multi -Reference Atlases, The multi-region-multi-reference framework for neonatal segmentation is carried out using neonatal images of 10 neonatal subjects (6 males & 4females) with age ranging from 26 to 60 days. For evaluation process 2 sagittal, 3covonal, &3 transverse slices of images are manually segmented by expert. The proposed method was compared with manual segmentation .The Dice ratio (DR) is used to measure tissue overlap rate for manual segmentation and automatic segmentation. The approach was evaluated with 2 other atlases. The first method (population A) was created 76 infants with ages ranging from 9 to 15 months. The second method (population B) uses the population atlas. To compare population A&B the joint registration -segmentation strategy is used to segment the brain images. It is to be said proposed method yield a good result. Decision fusion is used with multiple atlas because single atlas does not give a good result. Multiple atlases are carried out independently. A multi-region-multi-reference framework for neonatal brain image segmentation is proposed in this paper. For representing the local shape variation, multiple atlases are selected. Experimental results demonstrate that our method yield the highest agreement with manual segmentation and brings out two population-atlas based segmentation methods. In Neonatal Brain Image Segmentation in Longitudinal MRI Studies, MRI images of neonates were performed with more than 180 subjects. MRI scanning was performed using a 3T siemens scanner .In 10 subjects (4 females and 6 males) their neonatal images have been manually segmented.manual image segmentation was mainly focused on 2 sagittal slices,3 coronal slices and 3 axial slices. Segmentation was based on intensity based clustering method and then manually edited with ITK-SNAP software (yushkevich et al., 2006). In 10 subjects with both one-year-old and two-year-old images. We use both of them to guide neonatal image segmentation separately. To measure the overlap rate between two segmentation we use dice ratio (DR).The decision fusion technique could be potentially used to achieve better neonatal segmentation performance, by combining the segmentation results from multiple subject-specific atlases. A framework is presented by using subjectspecific tissue probabilistic atlas. The experimental results demonstrate that subject-specific atlas has superior performance compared to the populationbased atlases, and the proposed algorithm achieves comparable performance manual raters in neonatal brain image segmentation. The average total computation time is around 28 min for segmentation of a 256×256×198 image with 1×1×1 spatial resolution on a pc with 2.5 GHZ Pentium 4 processor.3 min is used for segmentation of a late time point image for generating a subject-specific atlas,14 min is used for atlas-to subject registration ,and 11 min are used for atlas based neonatal image segmentation. It is to be concluded that proposed segmentation framework is able to achieve satisfactory segmentation results with reasonable computational time. In Construction Of Multi-Region-Multi-Reference Atlases For Neonatal Brain MRI Segmentation, The proposed multi-region-multi-reference neonatal segmentation framework was applied to 10 subjects.10 image were manually segmented by expert rater using ITK-SNAP (yushkevich et al., 2006). Central brain region was not segmented due to extremely low tissue contrast. The proposed segmentation algorithm was compared with that of manual segmentation. The tissue overlap rate is compared with dice ratio (DR). The decision fusion is used to fuse the multiple image segmentation into single segmentation and the fused images are been used for manual segmentation and the result yields a good result. Our method yields the highest agreement with manual segmentation and outperforms the two average-shape atlas-based segmentation method .If the given population includes subject with a broad range of ages, the constructed multiple atlases in each region will learn all the shapes from different ages. It is to be concluded that multiregion-multi-reference atlas makes it adaptable to a large range of datasets. The methods such as brain parcellation, similarity measurement and image clustering can be future refined and optimized. From this survey on image segmentation using decision fusion gives good result on neonatal images and brain tissues. This method can be further used to get the better performance for even very small images. The manual segmentation are done for 10 subjects in Neonatal brain image segmentation in longitudinal MRI studies, it is to be said that manual segmentation can be evaluated for more subjects when decision fusion technique is used. In Construction of Multi-Region-Multi-Reference Atlases for Neonatal Brain MRI Segmentation, large range of data set is adaptable for manual segmentation when decision fusion technique is used to fuse the multiple image segmentation into a single segmentation to bring a final decision. The neonatal brain image when done with manual segmentation gives good result but additionally when decision fusion technique is used it yield a better result. It is to be concluded that manual segmentation with decision fusion yields a good result. All these work can be proceeded to get better result. May©2011 Global Journals Inc. (US) Email: kavisha58@gmail.com MayA Survey On Image Segmentation Using Decision Fusion Method ©2011 Global Journals Inc. (US) MayA Survey On Image Segmentation Using Decision Fusion Method ©2011 Global Journals Inc. (US) * Multiatlasbased segmentation of brain images: atlas selection and its effect on accuracy PAljabar RAHeckemann AHammers JVHajnal DRueckert NeuroImage 46 2009 * Infant brain probability templates for MRI segmentation and normalization MAltaye SKHolland MWilke CGaser NeuroImage 43 2008 * The ontogeny of human gyrification EArmstrong ASchleicher HOmran MCurtis KZilles Cereb. Cortex 5 1995 * Unified segmentation JAshburner KJFriston NeuroImage 26 2005 * Topologypreserving tissue classification of magneticresonance brain images PLBazin DLPham IEEE Trans. Med. Imag 26 2007 * FCM: the fuzzy c-means clustering algorithm JCBezdek REhrlich Comput. Geosci 10 1984 * A fully automatic and robust brain MRItissue classification method CACocosco APZijdenbos ACEvans Med. Image Anal 7 2003 * Measures of the amount of ecologic association between species LRDice Ecology 26 1945 * Mapping the early cortical folding process in the preterm newborn brain JDubois MBenders ACachia FLazeyras Ha-Vinh RLeuchter SVSizonenko CBorradori-Tolsa JFMangin PSHuppi Cereb. Cortex 18 2008 * Automatic anatomical brain MRI segmentation combining label propagation and decision fusion RAHeckemann JVHajnal PAljabar DRueckert AHammers NeuroImage 33 2006 * Detailed semiautomated MRI based morphometry of the neonatal brain: preliminary results MNishida NMakris DNKennedy MVangel BFischl KSKrishnamoorthy VSCaviness PEGrant NeuroImage 32 2006 * A survey of current methods in medical image segmentation DLPham CXu JLPrince Annu. Rev. Biomed. Eng 2 2000 * Automatic segmentation of MR images of the developing newborn brain MPrastawa JHGilmore WLin GGerig Med. Image Anal 9 2005 * HAMMER: hierarchical attribute matching mechanism for elastic registration DShen CDavatzikos IEEE Trans. Med. Imaging 21 2002 * Adaptive, template moderated, spatially varying statistical classification SKWarfield MKaus FAJolesz RKikinis Med. Image Anal 4 2000 * Simultaneous truth and performance level estimation (STAPLE): an algorithm for the validation of image segmentation SKWarfield KHZou WMWells IEEE Trans. Med. Imaging 23 2004 * Automatic segmentation of newborn brain MRI NIWeisenfeld SKWarfield NeuroImage 47 2009 * Joint registration and segmentation of neuroanatomic structures from brain MRI FWang BCVemuri SJEisenschenk Acad. Radiol 13 2006 * Automatic segmentation and reconstruction of the cortex from neonatal MRI HXue LSrinivasan SJiang MRutherford ADEdwards DRueckert JVHajnal NeuroImage 38 2007 * Effects of registration regularization and atlas sharpness on segmentation accuracy BT TYeo MRSabuncu RDesikan BFischl PGolland Med. Image Anal 2008 * A nonparametric method for automaticcorrection of intensity nonuniformity inMRI data JGSled APZijdenbos ACEvans IEEE Trans.Med.Imaging 17 1998 * Userguided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability PAYushkevich JPiven HCHazlett RGSmith SHo JCGee GGerig NeuroImage 31 2006 * A variational framework for integrating segmentation and registration through active contours AYezzi LZollei TKapur Med. Image Anal 7 2003