Bio-Inspired Computing For Brain Magnetic Resonance Image Segmentation

Brain magnetic resonance(MR)images segmentation into Gray Matter(GM),White Matter(WM)and Cerebrospinal Fluid(CSF)is not only a research topic of common concern in the field of medical image processing,but also an important foundation for associative diagnosis and treatment of brain related diseases.So far,there have been a large number of brain MR images segmentation algorithms,which mostly involve optimization problems.Compared to traditional numerical optimization methods,bio-inspired algorithms have stronger adaptability and the potential of searching the global optimum,so there has been an ever-increasing interest in bio-inspired algorithms.In this paper,bio-inspired algorithms for brain MR images segmentation have been studied to improve segmentation accuracy and efficiency.The main research results of this paper are as follows:(1)Local Variational Bayesian Inference Using Niche Differential Evolution(NDE-LVB)for brain MR images segmentation: To address image degradation caused by bias field and partial volume effect(PVE),each MR image is divided into many small data volumes characterized by local variational Bayes(LVB)models,which is inferred by the niche differential evolution(NDE)technique to avoid local optima.The experimental results on simulated and clinical data show that NDE-LVB can obtain higher accuracy than GA based segmentation methods.(2)Clonal Selection Algorithm(CSA)Combined with Differential Evolution(DE)and Estimation of Distribution Algorithm(EDA)-CSA/DE/EDA: to solve the limitations of hypermutation and receptor editing in traditional CSA,CSA/DE/EDA replaces hypermutation and receptor editing with DE and EDA,respectively,to extract the local information and the global information in current population.So the proposed algorithm should have both local and global searching ability.The comparison on five commonly used optimization functions indicates CSA/DE/EDA has stronger optimization ability that DE,EDA and CSA.(3)Hidden Markov Random Field(HMRF)Based Brain MR image Segmentation Using CSA/DE/EDA(HMRF-CSA/DE/EDA): Since the degradation factors in MR images may affect the stability of HMRF model parameters,HMRF-CSA/DE/EDA consists of two parts: the region-based and pixel-based HMRF segmentation.The region-based HMRF model can be used to locate search space for objective function,then CSA/DE/EDA is employed to train the pixel-based HMRF model.The comparison on simulated and clinical data shows that HMRF-CSA/DE/EDA outperforms other segmentation methods...