Modified Finite Mixture Model Combining Spatial Information For Brain MR Image Segmentation

With the increasing degree of aging in China,brain disease has become one of the major diseases that threaten the health of the people.Magnetic resonance imaging(MRI)technology because of its non-invasive,non-traumatic,less affected by target motion and can provide soft tissue clear features,are widely used in clinical diagnosis.However,due to the impact of imaging equipment itself,acquired images often contain noise,bias,the traditional methods are difficult to obtain ideal segmentation results.Therefore,segmentation of brain MR images is a hot and difficult part in medical image segmentation.Based on the finite mixed model,this paper makes full use of the pixel relations of each point of the image,and innovates the spatial constraint terms,and further improves the accuracy of the brain MR image segmentation.The main research contents of this paper include the following two aspects:(1)This paper proposes a model of brain MR image segmentation and bias field coupling recovery based on spatially constrained Gaussian model.This method based on Markov rand field(MRF)theory,using the prior pixels and their neighborhood points and posterior probability,construct new space constraints.At the same time,in order to overcome the phenomenon of intensity inhomogeneity,using the basic function fitting the bias field and coupling it to the improved Gaussian mixture model.Even if the segmentation image contains strong noise and bias field,the model can get more accurate results and recover the image offset field at the same time.(2)An improved spatial constraint term is introduced into SGM.A brain MR image segmentation and migration field coupling recovery model based on spatial variable skew Gauss mixture model is proposed.The model improves the noise robustness while using skewed Gauss distribution to reduce the influence of pixel skew distribution,thus further improving the segmentation accuracy.(3)A model of brain MR image segmentation and bias field coupling recovery based on SMM is proposed.This model has the ability to reduce the influence of noise and bias field and can more effectively depict the distribution information of brain tissue which obeys the heavy tailed distribution,so that more accurate results can be obtained.Experimental results of simulated brain images show that the proposed method is more accurate than existing similar methods.Experimental results of real brain MR images verify the effectiveness of the proposed method.