Research Of Brain MR Image Segmentation Algorithm Based On Fuzzy Clustering Theory

Brain disease is a severe threat to human health.How to use medical imaging to analyze the brain tissue qualitatively and quantitatively has become a hotspot.Magnetic Resonance Imaging(MRI)is particularly effective for the checking of soft tissue such as brain.Therefore,MRI has been widely used in clinical practice.Accurate segmentation of the brain tissue is the premise for the subsequent processing including brain analysis and 3D visualization.It can improve the reliability of brain diseases diagnosis and the effectiveness of treatment programs.First this thesis gave an overview on image segmentation methods at home and abroad,among which methods based on fuzzy clustering theory behave well and have a great prospect because of their many advantages.So this thesis will focus on fuzzy clustering algorithm,and the main work and research results are as follows:For the segmentation of brain MR images affected by noise,most existed algorithms based on FCM only smooth the membership or introduce spatial information controlled by a fixed parameter,which lead to different performance towards images affected by different level noise and bad performance towards much more complicated images.These algorithms cost too much running time as well.So an improved FCM algorithm named GKSFCM has been proposed in this thesis.To conquer the problem of ignoring spatial information in FCM,which leads to poor anti-noise performance,the algorithm introduces spatial information controlled by a noise-depending parameter.And the algorithm uses kernel induction distance instead of Euclidean distance in FCM to improve the classification ability.Moreover,to save the algorithm's running time,membership regularization is also introduced to speed up the convergence of the algorithm,obtaining a clearer membership distribution as to improve the efficiency of the algorithm.Simulation results show that compared with FCM and those based on FCM,the proposed algorithm performs better in anti-noise,segmentation accuracy and algorithm speed.For the segmentation of brain MR images affectd by bias field,most existed algorithms based on FCM can not accurately segment images affected by severe bias field or both affected by bias field and noise,or not suitable for all three sections of brain MR images.So an improved CLIC algorithm named W_GCLIC has been proposed in this thesis.The introduction of global information makes the segmentation objective function controlled by both regional information and global information.The global information is controlled by a parameter representing the similarity of pixel and pixel in its neighbor.And the parameter updates automatically in every iteration process.So the structure information is well considered and detais is well maintained,and as a result the bias field estimated from the image is much more accurate.Simulation results show that the compared with some of the existing fuzzy clustering algorithms algorithm,the proposed algorithm performs better in the intensity inhomogeneity image segmentation.