Navigation System for ACL Reconstruction Using Registration between Multi-Viewpoint X-ray Images and CT Images Mamoru Kuga a*, Kazunori Yasuda b, Nobuhiko Hata a, Takeyoshi Dohi a a Graduate School of Information Science, The Univ of Tokyo b Dept of Medical Bioengineering and Sports Medicine, Hokkaido Univ Abstract. A high-precision navigation system for Anterior Cruciate Ligament(ACL) reconstruction surgery is presented. In this system, 3D CT data is used to visualize the structure of bones and optical localizer is used to detect the position of femur, tibia and surgical tools. Since CT images are acquired preoperatively and markers are fixed to bones intraoperatively, their relative position has to be initialized. To achieve this initialization, 2D/3D registration method between CT model and X-ray images was developed. The registration is performed by minimizing the difference between contour lines. As ACL reconstruction requires precision, three X-ray images taken from different viewpoints are used for the registration. Experimental registration shows the Fiducial Registration Error(FRE) error to be 1.2mm. Keywords: image-guided surgery; navigation; 2D/3D registration; X-ray fluoroscopy; CT image 1. Introduction Anterior Cruciate Ligament (ACL) reconstruction is performed for a person whose ACL is damaged and causes knee instability, pain and recurrent swelling. In this surgery, surgeon makes tunnels on femur and tibia, fixing ACL graft to them. The positional precision of the graft insertion tunnel is important (less than 1mm is required), since it directly contribute to the patient's rehabilitation after the surgery [1]. However, for minimally invasive surgery, this surgery is commonly performed under endoscope, which requires expert skills for surgeons to grasp the internal structure of knee joint. Although X-ray fluoroscopy can be used to assist surgeons intraoperatively, limited number of images due to radiational exposure and restriction to 2D images are the problems. Therefore, the objective of this study is to develop a navigation system which can display the real-time positions of femur, tibia and surgical tool in 3D view. * Corresponding author. E-mail address: mkuga@atre.t.u-tokyo.ac.jp (M.Kuga).
The clinical significance of this paper is that the navigation system can achieve accurate and designed positioning of the graft insertion tunnel thus enabling the best possible surgical outcome. The engineering contribution of this paper is that, to the authors' best knowledge, this is the first proposal to integrate navigation and 2D/3D registration in ACL reconstruction surgery. The navigation system utilizes CT data set for obtaining 3D models preoperatively and optical localizer for tracking bones. Since CT images are acquired preoperatively and optical markers are fixed to bones intraoperatively, their relative position has to be initialized. For this initialization, registration based on Iterative Closest Point algorithm [2], 2D/3D registration with endoscopic image [3] or X-ray image, are the candidate methods. We selected to use X-ray image based 2D/3D registration, because of its high resolution and accessibility of images from various viewpoint. 2. Method In this navigation system, 3D CT data set is used to visualize the structure of bones and optical marker (POLARIS, NDI) is used to detect the position of femur, tibia and surgical tools. We used 3D Slicer [4] for displaying these bones and tools on computer screen. For initializing the relative position between the marker and bone, 2D/3D registration method between CT model and X-ray images was developed. In the registration, we used similarity measure which sum up the length of perpendicular lines between contour lines in X-ray image and 2D image of CT model because of the highcontrast nature of X-ray image [5]. Simplex method was used in the optimization procedure. The optical markers has to be detected also in X-ray images. We designed the marker to be observed as a set of eight disconnected spots in X-ray images by creating it from a combination of resin (PEEK) and metal (brass); Fig.1 shows a photo of the marker. The positions of these spots are sensitive along any rotational movement, which enables us to calculate accurate position and orientation of the marker in each image. A Fig. 1. A: Photo of designed marker both for registration and localization, B: same marker in X-ray image. Since ACL reconstruction requires precision, three X-ray images taken from different viewpoint (front, side and top) are used in the registration. We used three images B
because a registration only with a single image causes low-sensitive similarity measure along some axes such as depth variation. Considering the difficulty in accurate positioning of X-ray camera and collision between X-ray camera and other devices, we designed the registration algorithm so that these three images need not be taken from exact perpendicular view directions. Relative positions of these X-ray images are computed from markers in X-ray image. Using the markers in each image, registration takes following procedures: 1. Perform registration in front image from marker to X-ray image and from bone to X-ray image (Fig.2 A). 2. In top image, perform registration from marker to X-ray image and place the bone according to the result of procedure 1 (Fig.2 B). 3. Also in top image, optimize the position along depth variation of first image and the orientation whose similarity measure is sensitive in side image. Front-view coordinates is used in this optimisation. By doing so, this procedure does not affect the registration result in the front image (Fig.2 C). 4. Repeat procedure 2. and 3. in top image. This time, the optimization is performed only around one axis (Fig.2 D). A B C Fig. 2. Multi-viewpoint registration method. Black arrow refers to the registration along all six axes and white arrow along one axis. 3. Experimental Results D
Experimental registrations with a sheep's femur and tibia were conducted. We reconstructed 3D models of these bones from CT slice data with a thickness of 0.6 mm, image resolution of 512x512. After fixing the marker to the tibia with 3mm diameter fixation screw, we took sets of three X-ray images from roughly perpendicular view direction considering the limitation of camera viewpoint in surgery. 3D model of the marker from CAD data and bone models were superimposed to X-ray images in these three different views as shown in Fig.3. Whole registration process was automatic except for the setting of approximate initial positions. Using 512x512 X-ray images, registration was performed in less than two minutes with standard PC (Intel Pentium4 2.6GHz, main memory 512MB). We measured the registration errors as Fiducial Registration Error (FRE, the root mean distance between corresponding points) and it was 1.2mm, regardless of view direction. A B C D E F Fig. 3. X-ray images taken from three different viewpoint (A, B, C) and the 3D models of tibia and marker superimposed on each X-ray image (D, E, F). 4. Discussion As a result of multi-viewpoint X-ray registration method, high precision registration was achieved and there was no variation in error between X-ray images and 3D models according to the view direction. Due to the simple and round shape of femur and tibia, the third image also contributed to the precision. No improvement in precision was observed even when we used more than three X-ray images, because with three images, similarity measure is sharp and sensitive enough along all six axes. In terms of usability, semi-automatic process of this registration can be performed without professional skill. Although the manual setting of approximate initial position
was required currently, we can omit this process by taking the X-ray images from roughly the same viewpoints each time. Also, the position of camera in three images does not need to be measured, enabling operators to take X-ray images easily. After acquiring the position and orientation of femur, tibia and surgical tool, this information can be used not only for displaying internal structure to surgeons but also for robot control. For future work, we are planning to develop drilling robot based on this navigation system for making accurate ACL graft tunnels. The navigation system together with this drilling robot, error variation of insertion tunnel will be minimized and surgeons will be able to expect successful surgical outcome regardless of their skills. Since we are aiming the overall system error (including robot and navigation error) to be less than 1mm, the 1.2mm error only with registration is not small enough. In order to achieve higher precision, 1) accurate modeling of X-ray camera parameters such as focal length and view angle, 2) registration with higher resolution images by using pixel interpolation, are necessary. Also as a future task, we have to determine the optimal position and size of the optical marker since there is a tradeoff between precision and workspace: while higher precision can be expected as the marker is larger and closer to graft insertion point, the workspace for surgeon becomes smaller. In conclusion, we have proposed a high-precision registration algorithm for ACL reconstruction navigation using multi-view point X-ray images and confirmed the accuracy to be 1.2mm with experiments. References [1] K.Yasuda, J.Tsujino, Y.Tanabe, K.Kaneda, Effects of initial graft tension on clinical outcome after anterior cruciate ligament reconstruction, American Journal of Sports Medicine, vol.25, No.1, pp.99-106, 1997. [2] P.J.Besel, N.D.McKay, A method for registration of 3-D shapes, IEEE Trans. Pattern Anal. Machine Intell, vol.14, pp.239-256, 1992. [3] H. Fuchs, et al., Augmented reality visualization for laparoscopic surgery, MICCAI 98, pp.934-943, 1998. [4] D. Gering, A. Nabavi, R. Kikinis, N. Hata, L. Odonnell, W. Eric L. et al., An integrated visualization system for surgical planning and guidance using image fusion and an open MR, Journal of Magnetic Resonance Imaging, vol. 13, pp. 967-975, June, 2001. [5] A.Liu, E.Bullitt, S.M.Pizer, 3D/2D registration via skeletal near projective invariance in tubular objects, MICCAI 98 pp.952-963, 1998.