A New MR Brain Image Segmentation Using an Optimal Semi- supervised Fuzzy C-means and pdf Estimation

The work presented in this article concerns the classification of numeric data representing voxels of multimodal RM-Imaging. The procedure is partially supervised and it's not made any supposition on the number of classes and their correspondent's prototypes. The problem of initialization of the prototypes as well as their number is transformed in an optimization problem, besides the procedure is adaptive since it takes in consideration the partial and contextual information of some voxels by an adaptive and robust non parametric model. The procedure is founded on the evaluation of the probability density function (PDF) of the multimodal data. The originality of the method resides in a combination of evaluation of the PDF and the optimization of the number of classes by an energizing model applied to the fuzzy c-means classification algorithm (FCM) followed by a heuristic partial supervision. The quantitative and qualitative validation of this classification procedure of brain tissues and its performance are demonstrated through artificial MRI's data (125 cases) and real MRI's data (29 cases). The input images may be corrupted by noise and intensity nonuniformity (INU) artifact. This INU artifact is formulated as an additive bias field affecting the true MR imaging signal. A considerable improvement in the quality of the segmentation has been observed on it last at the time of the use of our algorithm.