3D Vertebrae Segmentation in CT Images with Random Noises Melih S. Aslan, Asem Ali, Aly A. Farag, Ben Arnold * , Dongqing Chen, and Ping Xiang * University of Louisville * Image Analysis, Inc. CVIP Lab., Louisville, KY,40299,USA 1380 Burkesville St. melih, asem@cvip.louisville.edu Columbia, KY, 42728, USA Abstract Exposure levels (X-ray tube amperage and peak kilo- voltage) are associated with various noise levels and ra- diation dose. When higher exposure levels are applied, the images have higher signal to noise ratio (SNR) in the CT images. However, the patient receives higher radiation dose in this case. In this paper, we use our robust 3D framework to segment vertebral bodies (VBs) in clinical computed tomography (CT) images with dif- ferent noise levels. The Matched filter is employed to detect the VB region automatically. In the graph cuts method, a VB (object) and surrounding organs (back- ground) are represented using a gray level distribution models which are approximated by a linear combina- tion of Gaussians (LCG). Initial segmentation based on the LCG models is then iteratively refined by using Markov Gibbs random field (MGRF) with analytically estimated potentials. Experiments on the data sets show that the proposed segmentation approach is more accu- rate and robust than other known alternatives. 1. Introduction The spine bone consists of the VB and spinal pro- cesses. Bone mineral density (BMD) measurements and fracture analysis of the spine bones are restricted to the Vertebral bodies (VBs). In this paper 1 , we are primarily interested in volumetric computed tomogra- phy (CT) images of the vertebral bones of spine column with a particular focus on the lumbar spine. (see Fig. 1 for regions of spine bone). Various approaches have been introduced to tackle the segmentation of skeletal structures in general and of vertebral bodies in particular for the anatomical def- inition of a VB. For instance, Kang et al. [1] proposed a 3D segmentation method for skeletal structures from CT data. Their method is a multi-step method that 1 This work has been supported by Image Analysis, Inc., Columbia, Kentucky, USA. Figure 1. Anatomy of a human vertebra (The image is adopted from [2]). starts with a three dimensional region growing step using local adaptive thresholds followed by a closing of boundary discontinuities and then an anatomically- oriented boundary adjustment. Applications of this method to various anatomical bony structures are pre- sented and the segmentation accuracy was determined using the European Spine Phantom (ESP) [3]. Later, Mastmeyer et al. [4] presented a hierarchical segmen- tation approach for the lumbar spine in order to mea- sure bone mineral density. This approach starts with separating the vertebrae from each other. Then, a two step segmentation using a deformable mesh followed by adaptive volume growing operations are employed in the segmentation. The authors conducted a perfor- mance analysis using two phantoms: a digital phantom based on an expert manual segmentation and the ESP. They also reported that their algorithm can be used to analyze three vertebrae in less than 10min. This tim- ing is far from the real time required for clinical appli- cations but it is a huge improvement compared to the timing of 1 − 2h reported in [5]. Other techniques have been developed to segment vertebral bones and skele- tal structures can be found for instance in [6, 7] and the 2010 International Conference on Pattern Recognition 1051-4651/10 $26.00 © 2010 IEEE DOI 10.1109/ICPR.2010.560 2282 2010 International Conference on Pattern Recognition 1051-4651/10 $26.00 © 2010 IEEE DOI 10.1109/ICPR.2010.560 2294 2010 International Conference on Pattern Recognition 1051-4651/10 $26.00 © 2010 IEEE DOI 10.1109/ICPR.2010.560 2290 2010 International Conference on Pattern Recognition 1051-4651/10 $26.00 © 2010 IEEE DOI 10.1109/ICPR.2010.560 2290 2010 International Conference on Pattern Recognition 1051-4651/10 $26.00 © 2010 IEEE DOI 10.1109/ICPR.2010.560 2290