Development of three dimensional binary patterns for local bone structure analysis

Three dimensional (3D) bone structure is important to assess the bone strength in a disease like osteoporosis. Conventionally, 3D bone structure has been modeled by “rod” and “plate” descriptors. However, subtle bone growth or resorption process occurring in the early stage of bone diseases may not be detected by this conventional model. In this paper, we describe a 3D local binary pattern (3dLBP) descriptor that should be particularly suitable for comparing the change of 3D local bone patterns in longitudinal studies. Individual 3dLBPs were designed on the basis of 6 neighbors in the 3D lattice. As a preliminary study, the difference between the cortical and trabecular bone patterns were investigated using of 3D human mandible images (n=9) obtained by cone beam computed tomography (CBCT). The cortical bones showed higher occurrences of “corner” and “edge” like local structures and higher heterogeneity in its 3D intensity distribution compared to the trabecular bones (Wilcoxon pairwise signed rank test, p<;0.05). The trabecular bones showed higher occurrences of “surface” like local structures and higher local uniformity in its intensity distribution (p<;0.05). Finally, 3d images were successively eroded to simulate the bone loss. The occurrence of “corner” like structures in the cortical bone was increased (Pearson´s analysis, rho=0.8, p<;0.01), while all other descriptors showed the changes to more homogeneous local structures during successive erosions. These results support that 3dLBP descriptors should be useful in comparing the local 3D structure changes that may occur during the bone loss.