A Novel Automatic Method of Cortical and Cancellous Bone Segmentation Based on Marrow Compartment

Abstract

Our novel cortical-cancellous bone segmentation method can segment cancellous bone from cortical bone automatically. This method is based on the observation that any bone tissue that is surrounded by marrow can be defined as cancellous bone and bone tissue that is outside of the marrow region can be defined as cortical bone. The marrow region is found by identifying the largest continuous three-dimensional structure that contains only cavities existing within the region of interest. For each two-dimensional image slice, a mask (CMask) that covers only marrow and cancellous bone is created by dilating-filling-eroding the marrow region. The bone elements covered by the CMask are considered cancellous bone, and the bone elements that are not covered by the CMask are considered cortical bone.
Visual inspection has confirmed successful segmentation of all data sets using the new method for two years in our lab. The segmentation results from the new method were also compared with those from a manual method and those from a published method, the so-called Helterbrand method. The mean values of the cancellous bone volume measured by the new method were not significantly different from those measured by the manual method, while there were significant differences between the mean values measured by the Helterbrand method and the manual method. The coefficient of variation was approximately 1% for the new method, demonstrating good precision. The variance of the new method was statistically significantly smaller than that of the manual method, which confirmed the better repeatability of the new method compared with the manual technique.