BRAIN TUMOR SEGMENTATION USING ASYMMETRY BASED HISTOGRAM THRESHOLDING AND K-MEANS CLUSTERING

Maya U.C, Meenakshy K

Abstract: Segmentation has become an important objective and still remains as a challenging area in medical image analysis. An accurate segmentation of the objects in medical images helps the doctors for proper diagnosis, surgical and treatment planning. The result of segmentation can be used by other processing techniques such as classification techniques to make the scope of segmentation wider. Segmentation result can also be used for quantification (volume calculation) purpose. An accurate segmentation is required for further manipulation of the problem. Moreover a small error in the segmentation process may get magnified in subsequent steps of processing. Segmentation extracts specific regions of interest from an image. The segmentation depth depends on the problem being solved. Several techniques exist for the segmentation of medical images. Various combinations of techniques have also been tried. Still the problem remains as a challenge. Even though it is simple, thresholding in its pure form does not give an accurate segmentation result in many cases, but when combined with other techniques of segmentation, it produces a highly accurate result. Human brain consists of soft tissues and Magnetic Resonance Imaging provides a better contrast for the same compared to other imaging modalities. Hence MRI is widely used in brain studies. Unlike other imaging modalities MRI is based on magnetic property of water molecules in human body and does not use radiation making it safer. This paper focuses on segmenting tumour affected region of brain from a Magnetic Resonance Image using thresholding and k-means clustering techniques. The proposed method contains eight important steps after which a segmented tumor region is obtained.

Keywords: Brain Tumor, MRI, Segmentation, Histogram Thresholding, K-means clustering

DOI: https://doi.org/10.15623/ijret.2014.0327012