Inner-Cluster-Structure Reconstruction Based Transfer Fuzzy Clustering and MRI Segmentation Applications

摘要

MRI automatically segmentation is very useful in clinical diagnosis. However, in some cases. MR images are contaminated by noises or lose pixels and then become sparse such that automatically segmentation by classical algorithms becomes difficult or impossible. In this study, we propose a transfer fuzzy clustering algorithm based on inner-cluster-structure reconstruction. Firstly, we use all objects to represent the inner cluster structure by assigning weights to all object. Secondly, in order to reconstruct the cluster structure in the target domain in which objects distribute sparsely or are contaminated by noises, we joint the two domains together and recalculate the weights of all objects in the target domain. Thirdly, the updated weights in the target domain are considered as transfer knowledge that is used for guiding the target domain learning. Experimental results on MR images and synthetic datasets indicate our novel algorithm achieves the best performance comparing with other similar algorithms.