Segmentation of pelvic bone in STIR MRI with active contour

Intensity inhomogeneities often occur in STIR Sequence MR images. For this reason, segmentation of these images is challenging. In STIR images from pelvis, signal of bone marrow is similar to subcutaneous fat and adjacent muscular structures so bone marrow pathologic lesions could be hidden. In this paper, we propose a method for image segmentation of pelvic bones in STIR MR images. This method is an extension of active contour methods. In this method, radius of localization is adaptively determined for each point of the contour. We combine this local structure with RSF model and we use the edge information to reach an appropriate segmentation. The proposed model outperforms RSF model in segmenting the images that have intensity inhomogeneities. We apply our algorithm on more than 120 MRI slices of 6 subjects to show its accuracy. Finally, we compare the result of the proposed model with those of two other active contours models.