Clinically desired segmentation method for vertebral bodies

In this paper, we propose a clinically desired segmentation method for vertebral bodies (VBs) in computed tomography (CT) images. Three pieces of information (intensity, spatial interaction, and shape) are modeled to optimize a new probabilistic energy functional; and hence to obtain the optimum segmentation. The information of the intensity and spatial interaction are modeled using the Gaussian and Gibbs distribution, respectively. A shape model is proposed using a new probabilistic function to enhance the segmentation results. This model is a generic shape information which is obtained using the cervical, lumbar, and thoracic spinal regions. We propose a semiautomated segmentation algorithm which uses limited interventions only in the VB separation process. The overall segmentation process completes the task in very low execution time which is one of the most important contribution of this paper. The proposed method is validated with clinical CT images and on a phantom with various Gaussian noise levels. This study reveals that the proposed method is robust under various noise levels, less variant to the initialization, and quite faster than alternative methods. One of the most important contributions of our paper is to offer a segmentation framework which can be suitable to the clinical works with acceptable results.