Center-free PFCM for MRI brain image segmentation

The Fuzzy C-Means clustering (FCM) and the possibility FCM (PFCM) are popular methods in MRI brain image segmentation. However, using the Euclidean squared-norm distance as the similarity criterion makes FCM and PFCM only suitable for clustering the hyperspherically distributed data groups. The MRI brain image does not distribute hyperspherically, which means FCM and PFCM have intrinsic deficiency for the segmentation of MRI brain image. The center-free FCM could segment the non-linearly separable data. But, it does not consider the spatial information and is very sensitive to noise. In order to segment the non-linearly separable data groups with noise, a center-free PFCM is proposed in this paper. Firstly, we modify the center-free FCM to deal with the non-linearly separable data. Then, we combine the improved center-free FCM with PFCM to make the new method less sensitive to noise. Experimental results on artificial datasets and MRI brain images show that our method is effective and outperforms the conventional FCM methods in the segmentation of the MRI brain images with noise.