Abstract- Segmentation of the hip cartilages is clinically important. Automatic segmentation of the hip bones is the first step in segmenting the cartilages as they are stiffer and less probable to be damaged in diseases. In this study, we propose an automatic multi step technique for segmentation the femoral head and acetabulum using clinically obtained multi-slice T1- weighted MR data through the following steps: 1) We resample data sets with a modified Sinc interpolation technique To perform an isotropic dataset, 2) By assuming a spherical shape for the femur, we estimate the center of the femoral head by a Hough transform,3) We develop a multistage approach for bone segmentation that employs a modified self-adaptive on-line vector quantization technique for a low-level image classification and utilizes a region-growing strategy for a high-level feature extraction, 4) We localize the hip joint space edges by customizing the anatomical constraint, 5) Finally, from our recent research we use segmented articular space to segment the acetabular and femoral head cartilages from each other. The techniques are implemented in C++ and Matlab programming languages. The feasibility of the proposed techniques is successfully evaluated in the presence of 40 hips including 1200 MR images. Keywords - Hough Transform; Directional Derivative Filter; Cartilage Segmentation; Adaptive Thresholding; Vector Quantization. I. INTRODUCTION Hip joint has a main role in human locomotion and bearing the body weight. The common causes of hip joint are osteoarthritis and dysplasia [1]. Quantitative evaluation of cartilage is a great help for orthopedists to study the pathogenesis of joint dysfunction. The thickness of the cartilage is only a few millimeters, so accurate segmentation is important for corresponding quantifications [2]. MR imaging is preferred for cartilage imaging because of its noninvasiveness, high soft-tissue contrast ability, and its multi-planar and three-dimensional data acquisition [2], [3]. A normal hip joint is formed like a ball and socket shape. The femoral head is located in the acetabulum, in the pelvic bone. The head of the femur is reinforced in its position by very powerful ligament. The surfaces of femoral head and acetabulum are covered by cartilages. The articular space separate these two cartilages that can be appeared after continues leg traction [4]. The purpose of this study is to develop an automatic technique to segment the hip joint cartilages from multi slice MR data acquired by T1-weighted 3D fast spoiled gradient echo (SPGR). As the goal of hip cartilage segmentation and because of the bone stiffness we need to segment the femoral head and acetabular bone accurately. Thus, we apply a vector quantization algorithm to segment the bones. The hip joint space is finally segmented by labeling the obtained edges and imposing the anatomical constraint associated with the location of the cartilages and femoral head center in the hip joint. With using the segmented articular space localized in our previous work [6] we can segment the hip joint cartilages from each other. In Section II, we explain the acquired data sets and the proposed techniques. In Section III, we evaluate the performance of the developed techniques in the presence of the available data sets; while in our previous work we did not quantify the method. The paper ends with the concluding remarks mentioned in Section IV. II. METHODOLOGY Data Set MR imaging is performed with fat-suppressed 3D fast spoiled gradient echo (SPGR) sequence using a unilateral surface coil on a 1.5-T MR system [1]. In a typical MR data set of a hip joint, acetabular and femoral cartilages are attached to each other. To allow clear separation of acetabular and femoral cartilage on MR images, the original continuous leg traction technique is used during MR imaging [4]. The voxel dimensions of the MR data in our study were non-cubic. We up-sample the data in the sagittal direction by Sinc interpolation, i.e., zero expansion in the frequency domain to reduce the effect of unwanted Gibbs ringing. The resultant volume data contains 256x256x192 voxels. Proposed Technique The proposed technique is a multi-step algorithm for accurate segmentation of the hip joint cartilages. The steps are as follows. A. Femur Center Estimation The femoral head typically has a spherical shape with radius of around 20 to 25 mm. in our previous work, the center of the sphere approximated femoral head was estimated, by embedding this constraint into the mechanism of the Hough transform [6]. By employing the estimated femoral head center and imposing the anatomical information about the size of the femoral head, the region of interest (ROI) is automatically selected. Fig 1 shows a typical example AUTOMATED SEGMENTATION OF HIP CARTILAGES IN MULTI-SLICE MR DATA M. Khanmohammadi 1* , R. A. Zoroofi 2* , Y. Sato 3Θ , T. Nishii ¤ , K. Nakanishi ¤ , H. Tanaka ¤ , N. Sugano ¤ , H. Yoshikawa ¤ , H. Nakamura † , S. Tamura Θ * Control and Intelligent Processing Center of Excellence Electrical and Computer Engineering Department Faculty of Engineering, University of Tehran, Tehran, IRAN. Θ Division of Image Analysis ¤ Department of Orthopaedic Surgery † Department of Radiology Osaka University Graduate School of Medicine, Osaka , JAPAN. Email: 1 m.khanmohamadi@ece.ut.ac.ir 2 zoroofi@ut.ac.ir 3 yohsi@image.med.osaka-u.ac.jp PROC. CAIRO INTERNATIONAL BIOMEDICAL ENGINEERING CONFERENCE 2006© 1