Segmentation Algorithm Of Brain MRI Image Based On FCM

Precise segmentation of brain MRI image is the precondition and key of 3D reconstruction, quantitative analysis, visualization for brain MRI image. However, the complexity and uncertainty of the MRI imaging process and the complexity and individual differences of the brain structure makes automatic and accurate the brain tissue of interest from MRI brain images segmented extremely complex and difficult. MRI brain image segmentation is a very important practical significance and value. It has become the field of medical research and computer image processing a hotspot.In this paper, the segmentation is to label WM (White Matter), GM (Gray Matter) and CSF (Cerebrospinal Fluid) of brain tissues. The difficulty can be summed up as two points. First is the image noise including the additive noise and multiplicative noise called bias field noise. Second is the complex structure and individual differences of the brain. In different anatomical position, even with the same kind of brain tissue, its composition and microstructure build local factors vary. Mainly considering additive noise and the impact of bias field, we described a common mathematical model of brain MRI image segmentation issues which can provide a good mathematics framework of MRI image segmentation.Fuzzy C-means(FCM) clustering algorithm iterative optimization objective function to obtain the fuzzy partition of the data sets. It has a good convergence. FCM algorithm used in image segmentation provides the advantages of 1) Avoid setting the threshold value. Can resolved the problem of various branches of the division; 2)It is Suitable for processing image of the uncertainty and ambiguity features; 3) Non-supervision division. Manual intervention does not need in clustering process. It is suitable to the application of the automatic segmentation.In this paper, we proposed a fast FCM algorithm called hfKFCM (histogram fast KFCM) to solve the problem of the heavy and the big number of iterations of FCM process. The improvements are 1) using the image histogram peak as fuzzy clustering initial cluster centers to reduce the number of iterations. 2) using the rapid clustering algorithm based on statistical histogram information to reduce the amount of calculation of a single iteration.The simulation results show that hfKFCM algorithm significantly reduce the number of iterations and the amount of computation comparing with standard FCM algorithm. However, this algorithm has some limitations. When the MRI image noise was severely affected by the noise that the histogram peak information can not be responsible for pixel distribution, this method will not provide effective segmentation results.Only using the gray scale information and without considering the pixel spatial information makes standard FCM algorithm for image segmentation apply to image with little noise.Simulation results show that a large number of isolated noise exists in the segmentation results of FCM algorithm when the image contains additive noise. The continuity of regional division and the quality are poor.Pixels located close to each other have the greater probability to belong to the same type. DFCM and sFCMpq algorithm use a rectangular window to designate a local neighborhood. The distance Measurement between pixels or the degrees of membership were smoothed to make a partial classification smoother. It improves the misclassification caused by noise and uneven distribution of image intensity. But in the neighborhood system, each pixel affected by pixel in its neighborhood is the same. Thus, the edges will be smoothed over, giving rise to the erroneous classification of the region at risk. IFCM algorithm also improves the performance by smooth operation of the local neighborhood. The difference is that IFCM algorithm considers the distance similarity and impact of relative position of the adjacent pixels. In this way, pixels with similar eigenvalue and location tend to be divided into the same cluster.We present a modified FCM algorithm called ssFCM . It uses the method of smoothing the membership that used in sFCMpq and the method of restricting the spatial information of neighborhood used in IFCM algorithm for reference. It considered the differences of pixels inside a class tissue and pixels at edge area in the processing of smooth. We use region statistical information to distinguish noise pixels, pixels inside a class and pixels at edge area. For the pixels inside a class and noise pixels, we use the smooth method to improve the smoothness and continuous of the segmentation results. For the pixels at edge area, we do not use the smooth method in order to save the detail information. In this way, we get a more reliable algorithm.The results of simulation show that the algorithm can get the content segmentation of brain MRI image even when the image is added to a noise of 9%. We also simulate the ssFCM algorithm when the brain MRI image image is affected by the bias field. It concludes that the ssFCM algorithm can reduce the effect of the bias field in some certain degree. However When the bias field is 40%, it can not provide content segmentation results.It provides much more continuous and smooth classification segmentation results. But the number of calculation is increasing because of the calculating of the statistical information of neighborhood.