Research On Automatic Segmentation Of Brain Tumors Based On Mri Images

At present brain tumor is a common brain disease.Magnetic Resonance Imaging(MRI)plain scan and enhanced MRI scan are the top choices for brain tumor examination,they can identify the location and cumulative range of tumors and help determine the nature of tumors.However,biased fields and noises would emerge during MRI imaging,making imaging uncertain and fuzzy.To automatically segment brain tumors in MRI,this thesis adopts Fuzzy C-means(FCM)algorithm,which is characterized by fuzziness and no supervision and can accomplish the automatic segmentation of MRI brain tumors.FCM algorithm has such advantages as few parameters,fast speed,and high accuracy.However,traditional FCM algorithm is sensitive to the initial value and can easily fall into local optimum value.For this reason,we can firstly improve the DA algorithm and then uses the improved dragonfly algorithm to optimize the parameters of the FCM algorithm,finally study and apply segmentation of MRI brain tumors with the improved FCM algorithm.Major work and innovation of the thesis are listed below:(1)Dragonfly Algorithm(DA)and the proposed improved algorithm PVDA.To solve traditional DA’s problems such as slow prematurity and convergence and low precision of solution seeking,this thesis firstly introduces the algorithm of particle swarm optimization(PSO)for each round of iteration,replaces the worst dragonfly with optimal particles,provides elegant solutions to DA,and improves the precision of optimization;then,the thesis improves the inertia weight,and defines an update strategy based on non-linear weight.In the process of optimization,global optimization plays a dominant role in the early stage to find all local extremums;in the late stage of iteration,local development plays a dominant role to find the global optimum.The improvement strategies can help speed up convergence to avoid fal ing into local extremums;in the end,the thesis introduces mutation strategy,increases population diversity.Tests are conducted with classical testing functions and the results show: PVDA is better than the original algorithm in convergence speed and solutionseeking precision,can jump out of local extremums,and help find the global optimum.(2)FCM algorithm based on PVDA optimization(PVDA＿FCM).To solve FCM algorithm’s problem of sensitivity to the initial value and prevent it from fal ing into local optimum,PVDA is adopted to optimize the initial cluster center and fuzzy weighted exponent,achieve automatic selection of initial cluster center and fuzzy weighted exponent,and avoid fal ing into local optimum value owing to improper initialization.The optimization results are then tested on UCI dataset and results show: PVDA＿FCM has a higher cluster accuracy than comparing algorithms including FCM algorithm based on PSO optimization(PSO＿FCM)and FCM algorithm based on ant colony optimization(ACO＿FCM)and displays outstanding performance in cluster evaluation index.(3)PVDA＿FCM optimization algorithm’s application in MRI brain tumor.To realize automatic segmentation of MRI brain tumors,PVDA＿FCM is used to conduct clustering segmentation.The simulation results show: in terms of direct segmentation results,PVDA＿FCM optimization algorithm’s segmentation is more accurate with better edge segmentation effects;in terms of the segmentation evaluation indicators,PVDA＿FCM algorithm displays better performance than its comparing algorithms including PSO＿FCM and ACO＿FCM.