Research Of Fuzzy Clustering Based Image Segmentation In Medical Image Analysis

With the rapid development of computer technology, automatization and electronics,medical imaging technics are highly developed. This makes it possible for clinicians toexamen pathological changes more and more directly, accurately, and clearly. In practicesand researches of clinical medicine, measuring of shape, boundary, sectional area or volumeof tissues and organs is often required, so that important information of the organ'spathology and function can be acquired. Accurate measuring is of great signi?cance fordiagnoses and therapy.Medical image segmentation technics are far from perfectness although various meth-ods are developed because of the distinct individual di?erence and the strict requirementfor precision and speed of segmentation algorithm. For all the reasons above, segmentationalgorithm is still an emphasis in research of medical analysis.There is often much noise in medical images which seriously a?ects segmentationresults, but conditional fuzzy Kohonen's competitive learning algorithm won't resist itsatisfactorily. Aiming at this problem, a Modi?ed Fuzzy Kohonen's Competitive Learningalgorithm is proposed in this thesis which can perform automatic segmentation on brainMRI images. Problems mainly discussed include:1. Comparing conditional FKCL algorithm with FCM algorithm under the precondi-tion of lucubrating both. Experimental results show that FKCL algorithm can not resistnoise satisfactorily enough, though it out performs FCM.2. Concluding the relationship between the central pixel and neighbor pixels in alocal window. A local similarity measure factor is introduced as the local informationbased on the conclusion above.3. Embedding local similarity measure factor into conditional FKCL algorithm, sothat noise can be restrained.Modi?ed FKCL algorithm is implemented to apply to medical image segmentation,and it has been proved e?ective.