Automated segmentation of brain tumors in MRI using potential field clustering

Abstract

We propose potential field clustering, a new algorithm based on an analogy with the concept of potential field in Physics. By viewing the intensity of a pixel in a FLAIR MRI image as a “mass” that creates a potential field, the algorithm is used for tumor localization. The center of the localized tumor cluster is then used as the initial seed in a region growing segmentation algorithm. We evaluate the performance of this segmentation approach on the publicly available brain tumor image segmentation MRI benchmark. The performance of the proposed approach is compared with that of the Force clustering algorithm by Kalantari et al. (2009). Experimental results show that the proposed algorithm is more accurate in localizing tumor centers, which, in turn, results in better segmentations.