Research On Brain MRI Segmentation Based On Finite Mixture Los

Accurate segmentation of medical images can provide physicians with effective data sup-port,effectively improving diagnostic accuracy.The main medical imaging techniques com-monly used today include digital shadow angiography,electronic computed tomography and magnetic resonance imaging.Among these techniques,MR images are widely used in the di-agnosis of brain diseases because of their ability to provide high-resolution image information of brain tissue.With continuous improvements in computer technology,deep learning techniques have been able to effectively improve the segmentation accuracy of brain tissue.However,deep learning-based image segmentation algorithms often rely on large amounts of labelled data,and in the field of medical imaging,obtaining large amounts of labelled data is often difficult and suffers from the small sample problem.In addition,brain MR images often contain bias fields,which can also affect the accuracy of segmentation.Therefore,to address the above problems,this paper aims to propose a semi-supervised method based on deep learning to solve the small sample problem and the bias field problem.The specific work is as follows.(1)For the small sample problem,this paper constructs a loss function based on a Gaus-sian mixture model.In the case of small samples,the model can couple not only the personality features of individual samples,but also the common features of all samples extracted by deep learning,thus improving the segmentation accuracy of the model.In addition,bias field re-covery is coupled into the Gaussian mixture model loss function,which can effectively reduce the influence of the bias field.The experimental results show that the method proposed in this paper can effectively improve the segmentation accuracy.(2)It has been found that brain MR images do not exactly obey symmetric distribution.To address this,we propose a loss function based on a skewed Gaussian mixture model to improve the segmentation accuracy of the model for brain MR images with asymmetric distribution.Similarly,we coupled bias field recovery in the model to effectively improve the final segmen-tation accuracy.Experimental results show that the method proposed in this paper can produce desirable results in the case of small samples.