Abstract
One of the important problems in medical diagnosis is the segmentation and detection of brain tumor in MR images. The accurate estimation of brain tumor size is important for treatment planning and therapy evaluation. In this regard, this paper presents a new method, termed as SoBT-RFW, for segmentation of brain tumor from MR images. It integrates judiciously the merits of rough-fuzzy computing and multiresolution image analysis technique. The proposed method starts with a simple skull stripping algorithm to remove non-cerebral tissues such as skull, scalp, and dura from brain MR images. To extract the scale-space feature vector for each pixel of brain region, the dyadic wavelet analysis is used, while an unsupervised feature selection method, based on maximum relevance-maximum significance criterion, is used to select relevant and significant textural features for brain tumor segmentation. To address the uncertainty problem of brain MR image segmentation, the proposed SoBT-RFW method uses the robust rough-fuzzy c-means algorithm. After the segmentation process, asymmetricity is analyzed by using the Zernike moments of each of the tissues segmented in the brain to identify the tumor. Finally, the location of the tumor is searched by a region growing algorithm based on the concept of rough sets. The performance of the proposed SoBT-RFW method, along with a comparison with related approaches, is demonstrated on a set of synthetic and real brain MR images using standard validity indices.