| The segmenting medical images is a key procedure which exerts a great impact on wide-range application of image processing technique to medicine such as 3D reconstruction, quantitative analysis, visualization, and so on. The segmentation of MR brain image will be much more complex and difficult because of the following three factors: 1) Indistinct boundaries between brain tissues due to their overlapping and penetrating with each other. 2) Variance in brain structures among individuals. 3) Intrinsic uncertainty of MR images due to heterogeneity of magnetic field, partial volume effect and noise induced during imaging.In this thesis, the various segmentation methods of MR images previous proposed are reviewed firstly. Then, considering the multi-spectral feature of MR images, the algorithms of segmentation based on fuzzy clustering are analyzed and designed. Fuzzy clustering is not too sensitive to noise and good to handle the intrinsic uncertainty of the object. It describes a complex system in such a way that is not so exact, so that it can effectively segment such images whose boundary is indistinct. Fuzzy clustering is much more similar to the thinking way of human being because it uses logic of continuous value instead of binary logic. Therefore, the segmentation methods based-on fuzzy clustering is strongly suitable to handle such an image that has blurry boundaries like brain MRI.The axial plane MR brain images, which are from Visible Human Data Set, are determined as the study object in this thesis. In order to more exactly segment the white matter, gray matter and cerebrospinal fluid, the following two procedures are adopted to process the MR brain images: First, some image preprocessing methods are adopted so that the non-brain tissues such as skull and scalp are eliminated while only the brain structure can be remained in processed images. Secondly, the preprocessed images are further segmented using methods based on fuzzy clustering technique.
|
PDF Full Text Request