Brain Tumor Segmentation in MR Images Based on an Improved Two-Phase Deformable Model

Due to the complex structure of the brain, it is a complicated task to accurately locate and segment tumors in magnetic resonance (MR) images. Even though computer-aided diagnosis has been a flourishing research field in the past decades, it still encounters many challenges in MR image segmentation. Consequently, this paper proposes an automatic and accurate algorithm for brain tumor segmentation in the hope to reduce the burden of physicians. Firstly, we employ mathematical morphology, histogram analysis, and a markercontrolled watershed method to obtain the initial surface contour. A modified edgeless active contour model is then applied to obtain final tumor segmentation results. The Brain Tumor Segmentation (BraTS) dataset provided by Medical Image Computing and Computer Assisted Intervention (MICCAI) in 2012 was utilized for performance evaluation. Segmentation results of the whole brain tumor in the BraTS2012 dataset reached an average Dice of 85.79%, which outperformed many state-of-the-art methods. This fully automatic brain tumor segmentation algorithm is potential for a good auxiliary tool to assist physicians in diagnosis and to facilitate brain cancer-related research.