Fuzzy Clustering Algorithm Based Brain MRI Image Segmentation

Magnetic resonance imaging (MRI) is a popular non-invasive imaging technique. It has been applied to brain tissue imaging widely for its good resolution for soft tissue. The correct segmentation of brain tissue helps to research brain development and human aging, brain disease diagnose, location of lesion and surgical planning. Due to limitation of device technique and motion of the object during the imaging process, MRI is affected by noise, bias field and partial volume effect which degrade the automatic segmentation result. The task of this paper is to study automatic algorithm classifying brain MRI image into white matter, gray matter and cerebrospinal fluid when the noise and bias field exist.Fuzzy clustering algorithm is suitable to brain MRI image segmentation, because of the soft classification of samples, which describes the blur of MRI image caused by partial volume effect. Fuzzy C-Means (FCM) is the most studied fuzzy clustering algorithms. Due to lack of neighbor information, standard FCM is sensitive to noise, bias field and will get inaccurate segmentation results.To improve the robustness of FCM to noise, introducing neighborhood information into the objective function is the general method. In this paper, we firstly analyze the ideas of several improved FCM algorithms by studying the algorithms’theories and, comparing the experimental results. We then propose a new algorithm edge sensing fuzzy local information c-means clustering algorithm for image segmentation. The new algorithm, which utilizes the impact of local spatial, gray information and the homogeneity degree of neighbor window to the objective function, is robust to noise. The segmentation results of synthetic images and brain MRI images show that our proposed algorithm is more accurate than several other algorithms.Generally, the algorithm which can reduce the influence of noise and bias field is benefit for accurate brain tissue segmentation, because noise and bias field often exist in brain MRI image simultaneously. To solve this problem, we present a new algorithm by combining edge sensing fuzzy c-means and coherent local information clustering. The new algorithm can remove noise and correct bias field at the same time. The segmentation results of large synthetic images, simulation brain MRI images and real brain MRI images demonstrate that the new algorithm has better segmentation performance and can classify brain tissue correctly.