3D Simultaneous Segmentation and Registration of Vertebral Bodies for Accurate BMD Measurements

In this paper, a novel 3D vertebral body (VB) segmentation method in computed tomography (CT) images is presented. The proposed approach depends on both intensity and shape information. The intensity information is managed by embedding an edge-mounted Willmore energy into the level set segmentation framework. Shape information is obtained from a set of training shapes. Shape variations are estimated using a probabilistic model which approximates the marginal densities of the vertebral body and its background in the variability region using a Poisson distribution. The proposed methods are validated with 40 clinical CT images as well as on a phantom with various Gaussian noise levels. The experimental results show that the segmentation accuracy of our framework is much higher than other alternatives. Applications on BMD measurements of vertebral body are given to evaluate the accuracy of the proposed segmentation approach.