A 3D Framework for the Musculoskeletal Segmentation of Magnetic Resonance Images

Abstract

In this thesis a new framework is proposed for obtaining the spongy bone, cortical bone, muscle and adipose tissue from MRI data. The method focuses on the accurate extraction of the edges of the target tissues, which is the main drawback of previous works. In this framework six new methods, as listed in section 1.3, are utilized together for improving the result of the segmentation by detecting the relational position of the tissues, acquiring the best possible contribution from the operator in terms of time and efficiency, forward and backward transfer of the segmented tissues at the seed slice and using the newly proposed Deformable Kernel Fuzzy-C Mean (DKFCM) method for improving the result of segmentation on the edges. This method first limits the searching area for the voxels of the target tissue from the whole data to a small strip around the edges of the target tissue. Then, it applies a very accurate segmentation on the searching area, using a deformable kernel, which is capable of adapting itself with the shape of the edge. Comparing the results of this work with some popular MRI segmentation methods like active contour, watershed, FCM and also some heuristic methods, which are proposed in literature for segmenting the MRI to four tissues, demonstrates the superiority of the proposed method especially on the edges.