Automatic brain tumor detection from magnetic resonance image using unsupervised segmentation techniques

Abstract

The major goal of this dissertation is to present a new unsupervised newlinesegmentation algorithms used to detect brain tumor from Magnetic Resonance newlineMR Images We developed four new segmentation algorithms for newlineimprovement on the accuracy and computational time When applied to any newlinekind of brain tumor affected magnetic resonance images the algorithm newlineperformed with very good results and was able to find 81 percent to 98 newlinepercent accuracy newlineFirst method hierarchical self organizing map and fuzzy cmeans newlinealgorithm are used to classify the image layer by layer The lowest level newlineweight vector is achieved by the abstraction level We have also achieved a newlinehigher value of tumor pixels by this neurofuzzy approach The multilayer newlinesegmentation results of the neurofuzzy are shown to have interesting newlineconsequences from the viewpoint of clinical diagnosis Neurofuzzy newlinetechnique shows that MRI brain tumor segmentation using HSOMFCM also newlineperform more accurate one newline newline