SEGMENTATION OF VERTEBRAE USING LEVEL SETS WITH EXPECTATION MAXIMIZATION ALGORITHM Melih S. Aslan 1 , Aly A. Farag 1 , Ben Arnold 2 , and Ping Xiang 2 1 University of Louisville, Computer Vision and Image Processing Lab., Louisville, KY, 40299, USA 2 Image Analysis, Inc., 1380 Burkesville St., Columbia, KY, 42728, USA ABSTRACT In this paper, we propose a robust level sets method to seg- ment vertebral bodies (VBs) in clinical computed tomography (CT) images. Since the VB and surrounding organs have very close gray level information and there are no strong edges in some CT images, the initialization of level sets method be- comes very crucial step. If the object and background re- gions are not initialized correctly, the results would not be acceptable. Also, the size and place of the initial seed may give non-reproducible results. To solve these problems, we use a statistical level sets method which uses the Expectation- Maximization (EM) algorithm for the initialization and pa- rameter estimation. Validity was analyzed using ground truths of data sets (expert segmentation) and the European Spine Phantom (ESP) as a known reference. The proposed method is compared with other known alternatives. Index Terms— Spine bone, vertebral body (VB) segmen- tation, statistical level sets, expectation-maximization. 1. INTRODUCTION The spine bone consists of the vertebral body (VB) and spinal processes. Bone mineral density (BMD) measurements and fracture analysis of the spine bones are restricted to the VBs. In this paper, we are interested in computed tomography (CT) images of the vertebral bone. The primary goal of the pro- posed work is in the field of spine densitometry where BMD measurements are restricted to the vertebral bodies (see Fig. 1 for regions of spine bone). Osteoporosis is a bone disease characterized by a reduc- tion in bone mass, resulting in an increased risk of fractures. Therefore, correct VB segmentation takes an important step to identify vertebral fractures and to measure BMDs which are used in evaluating new osteoporosis therapies [1]. Seg- mentation is an important method for feature extraction, image measurements, and image display. Segmentation has been used in many applications such as detection of coronary border in angiograms, multiple sclerosis lesion qualification, surgery simulations, surgical planning, measuring tumor vol- ume, functional mapping, automated classification of blood cells, studying brain development, detection of micro cal- cification on mammogram, and image registration. Various Fig. 1. An example view of a spine bone. Left and right im- ages show axial and sagittal views of the spine bone, respec- tively. Pink color shows the VB region (The image is adopted from [4]). (a) (b) (c) (d) Fig. 2. Typical challenges for vertebrae segmentation. (a) Inner boundaries. (b) Osteophytes. (c) Bone degenerative disease. (d) Double boundary. approaches have been introduced to tackle the segmentation of spine bones. For instance, Klinder et al. [2] developed an automated model-based vertebra detection, identification and segmentation approach. Kang et al. [3] proposed a 3D segmentation method for skeletal structures from CT data. Their method is a multi-step method that starts with a three dimensional region growing step using local adaptive thresh- olds followed by a closing of boundary discontinuities and then an anatomically-oriented boundary adjustment. Appli- cations of this method to various anatomical bony structures are presented and the segmentation accuracy was determined using the ESP [5]. Later, Mastmeyer et al. [6] presented a hierarchical segmentation approach for the lumbar spine in order to measure bone mineral density. They reported that their algorithm can be used to analyze three vertebrae in less than 10min. This timing is far from the real time required for clinical applications but it is a huge improvement compared to the timing of 1 − 2h reported in [7]. Recently, in the con- text of evaluating the Ankylosing Spondylitis, Tan et al. [8] 2010 978-1-4244-4128-0/11/$25.00 ©2011 IEEE ISBI 2011