Bone micro architecture is believed to play a key role in determining bone quality. We propose a new method of segmentation based on local shape classification. Bones samples are thus described into their basic elements (rods and plates). On each bone voxel we calculate the inertia moment of a neighborhood obtained by local geodesic dilation in the bone volume. The dilated volume is obtained through a homothopic dilation using the Fast Marching algorithms. The size of the dilated volume is choosen from local aperture diameter in order to be scale independent. The bone cross-section is calculated using an optimized granulometry algorithm. The segmentation has been carried on a wide range of human trabecular bone with varied structure. Voxels are then classified according the ratio between the inertia moments of the dilated volumes.

Date Published: 12 March 2010
PDF: 8 pages
Proc. SPIE 7623, Medical Imaging 2010: Image Processing, 76234O (12 March 2010); doi: 10.1117/12.843804

Published in SPIE Proceedings Vol. 7623:
Medical Imaging 2010: Image Processing
Benoit M. Dawant; David R. Haynor, Editor(s)