Study On Segmentation Of Brain MR Images Based On Fuzzy Clustering

Recently, Magnetic resonance imaging (MRI) has been widely and deeply used in clinical medicine, due to its unique advantages, such as non-harmful, non-invasive, and seldom affected by the motions of target. Segmentation of MR images plays an important role in anatomic structure, tissues quantitative measurement and disease diagnosis, and it has become the primary means in studying the function, pathology and anatomic of body.Because fuzzy clustering can deal with the fuzzy boundary between the brain internal tissues and the inherent uncertainty properly, it has been widely used in segmentation of MR images. Among the fuzzy clustering, fuzzy c-means (FCM) clustering obtains the most widely used. It achieves the segmentation through minimizing the objective function, and it has advantages such as unsupervised study, simple realization, fast operation, etc. However, because of factors about radio frequency, noise and bias usually intrude into the MR images, thus the conventional FCM algorithm obtain the unsatisfying results. For that, this paper wants to introduce the spatial information of images to improve the objective function of traditional algorithm, and integrate the bias correction into the framework, so that it can reduce the effect of noise, and recover the bias field. Our work includes the following aspects:(1) A fuzzy c-means model based on the spatial structure information is proposed. This model takes both the non-local information and spatial structural similarity measurement (SSIM) between the image patches into consideration, and then a new distance function is established for image segmentation. The proposed model can reduce the noise effectively, and keep more structural information as well.(2) An image segmentation and bias correction model based on improved FCM with non-local information is proposed. The non-local information and bias correction is integrated into the model to reduce the effect of noise and intensity inhomogeneity as well as keep the image structures, and this model introduce membership regularized term to obtain crisp membership degree, so that the effect of membership at the transition are can be reduced, and the result of classification can be improved.(3) A novel model based on FCM with weighted image patch is proposed for image segmentation and bias correction. This model utilizes image patches to replace pixels in the fuzzy clustering, and construct a weighting scheme to able the pixels in each image patches to have anisotropic weights, and hence reduce the impact of noise and keep the image structures. Meanwhile bias field is taken into the model to reduce the effect of intensity inhomogeneity.