142 Image Processing (NCIMP 2010) A Novel Method for Segmentation of the Hippocampus K. Somasundaram 1 and S. Vijayalakshmi 2 Department of Computer Science and Applications Gandhigram Rural Institute, Deemed University, Gandhigram, Dindigul, Tamil Nadu, India E-mail: 1 somasundaramk@yahoo.com; 2 Sviji_suji@yahoo.co.in ABSTRACT—Analysis of brain image is an important area under medical research. A lot of researches are going on for segmenting the medical images automatically. Hippocampus is a component of brain. It plays an important role in the normal behaviour of the human beings. It is a new region of research now. It takes many hours for a specialist in this area to segment the Hippocampus manually. There are many techniques available for the segmentation process. A modified approach based on the Watershed algorithm for segmenting the Hippocampus region of the brain has been presented in this paper. Initially the image is enhanced using the filters. The brain image is converted into its binary form using two approaches. In the first approach the block mean, mask and labelling concepts have been implemented and in the second approach tophat, mask and labelling concepts have been implemented. It is found that certain parts of the image contain holes which interrupt the segmentation process. So the image hole filling techniques are implemented and the related components are grouped into connected components. The gaps between the connected components are yet to be enlarged in order to isolate the hippocampus. The unwanted is removed using the run length encoding scheme. To start with only a single slice of the brain in two dimensions is considered. We have successfully extracted the hippocampus from one slice. KEYWORDS: Schizophrenia, Magnetic Resonance Imaging (MRI), Hippocampus, Neurodevelopmental, Gray Matter, Shape Asymmetry. I. INTRODUCTION Volumetric approaches for the analysis of Hippocampal Structures (HS) are required for the diagnosis of different diseases. In order to diagnose, evaluate and compare patient datasets, standardized and reproducible analysis are needed in clinical routine and research. MR-volumetric assessment of temporomesial structures, especially hippocampal structures, play an important role in the diagnosis of temporal lobe epilepsy degenerative diseases like Alzheimer dementia and the evaluation of their course of disease. Objective, reliable and reproducible semi- or fully automatic methods are required to record minor atrophies, evaluate chronological changes and confidently compare them among different observers and institutes. Resulting increased inter- and intra-observer variability of manual techniques limit the role of segmentation in clinical studies. Only valid results of hippocampal volume segmentation can be confidently correlated to severity of disease progression and to neurohistopathological results. Conventionally used manual segmentation is a time consuming procedure, which therefore cannot be performed in clinical routine. Fully automatic segmentation may sometimes fails if the hippocampus is relatively small and the shape of the object is highly variable. Compared to this, semiautomatic methods, may sometimes provide a more realistic approach because of the combination of human expertise and automatic techniques. The location of the hippocampus is shown in the following MRI scan of brain which is marked by an arrow. Fig. 1: Hippocampus II. METHODS AND TOOLS USED This paper contains the following sections. In section I we have presented the past history analysis of the some segmentation algorithms with their merits and demerits. In section II we have discussed the basic concepts involved in the proposed algorithm. In section III we have presented the results and discussionsand concluded with the future enhancements. Image Processing (NCIMP 2010) Editor: K. Somasundaram Allied Publishers