Extremity bones segmentation in cone beam computed tomography, a novel approach

Bone segmentation in Cone Beam Computed Tomography (CBCT) may appear as a simple and efficient task. Unfortunately, this is not true for complex areas such as the hand, wrist or feet, where there are many small and thin bones that are very difficult to distinguish. In this article an efficient graph cut method to segment complex districts in CBCT has been developed. A graph cut approach is initialized with an automatic ”background” and ”object” evaluation, carried out with pixel-based techniques, such as Otsu and Yen thresholding. Segmentation result is post-processed with morphological operation, then connected components with 26-connectivity are evaluated to separate the different bone. Finally, the user can isolate the bone (or bones) of interest and display a 3D model. The approach is compared with a standard graph cut approach whit requires user scribbles to segment a single bone, both in terms of accuracy and usability.