MR brain image segmentation using a possibilistic entropy based clustering method

A novel pixel-intensity-based segmentation technique is presented for magnetic resonance (MR) brain images segmentation using possibilistic entropy clustering. A brief analysis of the problems of boundaries shifting and region blur in FCM based MR images segmentation is made, which reveals that the lack of robustness to noise and outliers and inappropriate membership assignment in intensity space lead to such problems. Within the framework of possibilistic entropy theory, the proposed algorithm inherits the merits of possibilistic theory and shows a great robustness to noise and outliers for class center estimation. It can also automatically control the resolution parameter during the clustering is progressing. Finally, the experiments of cerebrum region segmentation and lesion detection verify its effectiveness over the FCM algorithm.