CORONARY ARTERY MOTION MODELING FROM 3D CARDIAC CT SEQUENCES USING TEMPLATE MATCHING AND GRAPH SEARCH Dong Ping Zhang a* , Laurent Risser a,b , Coert Metz c , Lisan Neefjes d Nico Mollet d , Wiro Niessen c , Daniel Rueckert a a Department of Computing, Imperial College London, London, UK b Institute for Mathematical Science, Imperial College London, London, UK c Dept. of Medical Informatics and Radiology, Erasmus MC, University Medical Center, Rotterdam, NL d Dept. of Radiology and Cardiology, Erasmus MC, University Medical Center, Rotterdam, NL ABSTRACT In this paper we present a method for coronary artery motion track- ing in 4D cardiac CT data sets. The algorithm allows the automatic construction of a 4D coronary motion model from pre-operative CT which can be used for guiding totally-endoscopic coronary artery bypass surgery (TECAB). The proposed approach is based on two steps: In the first step, the coronary arteries are extracted in the end-diastolic time frame using a minimal cost path approach. To achieve this, the start and end points of the coronaries are identi- fied interactively and the minimal cost path between the start and end points is computed using the A* graph algorithm. In the sec- ond stage the coronaries are tracked automatically through all other phases of the cardiac cycle. This is achieved by automatically iden- tifying the start and end points in subsequent time points through a non-rigid template-tracking algorithm. Once the start and end points have been located, the minimal cost path is constructed in every time frame. We compare the proposed approach to two alternative ap- proaches: The first one is based on a semi-automatic extraction of the coronaries with start and end points manually supplied in each time frame and the second approach is based on propagating the ex- tracted coronaries from the end-diastolic time frame to other time frames using non-rigid registration. Our results show that the pro- posed approach performs significantly better than non-rigid registra- tion based method and that the resulting motion model is comparable to the motion model constructed from semi-automatic extractions of the coronaries. Index Terms— Image Registration, Motion Detection and Tracking, Image Guided Surgery, Cardiovascular Image Analysis 1. INTRODUCTION As one of the leading causes of death worldwide, coronary artery disease occurs due to the failure of blood circulation to supply ade- quate oxygen and nutrition to cardiac tissues. It is typically caused by the excessive accumulation of atheromatous plaques and fatty de- posits within certain regions of the arteries which restricts the blood flow. To treat this disease, arteries or veins grafted from the pa- tient’s body are used to bypass the blockages and restore the sup- ply to the heart muscle. Based on image-guided robotic surgical system, totally endoscopic coronary artery bypass (TECAB) surgery has been developed to allow clinicians to perform the bypass surgery off-pump with three pin-hole incisions in the chest cavity, through which two robotic arms and one stereo endoscopic camera are in- serted. However, 20-30% conversion rates from TECAB surgery to the conventional invasive surgical approach [1, 2] have been reported due to the vessel misidentification and mis-localization caused by the restricted field of view of the stereo endoscopic images. The goal of our work is to construct a patient-specific 4D coro- nary artery motion model from preoperative cardiac CT sequences. By temporally and spatially aligning this model with the intraop- erative endoscopic views of the patient’s beating heart, we expect to assist the surgeon to identify and locate the correct coronaries during the TECAB procedures [3, 4]. In previous work, Shechter et al. [5, 6] tracked coronary artery motion in a temporal sequence of biplane X-ray angiography im- ages. In their approach, a 3D coronary model is reconstructed from extracted 2D centrelines in end-diastolic angiography images. The deformation throughout the cardiac cycle is then recovered by a registration-based motion tracking algorithm. The disadvantage is that 3D reconstruction of the coronary is required. An alternative approach for the extraction of the coronaries from cardiac CT has been proposed by Metz et al. [7]: Here the coronaries are man- ually or semi-automatically identified at one time frame and then tracked throughout the cardiac cycle using non-rigid registration of the multi-phase cardiac CT images. The restriction of this approach is that highly localized motion of the coronaries can not be fully recovered by the motion tracking of the entire heart. In this paper, we present an approach for coronary motion track- ing in cardiac CT images which significantly improves the accuracy of motion tracking and reduces the manual interaction. The pro- posed approach is based on an template fitting and tracking algo- rithm which automatically identifies the start and end points of each vessel in every time frame. Once the start and end points have been identified the vessel is extracted as the minimal cost path between both points. The proposed approach is compared to a registration based approach similar to the one presented by Metz et al. [7] and to manual tracking of the coronaries. This simplifies the 4D motion modeling of the coronaries significantly. 2. METHOD We first use contrast limited adaptive histogram equalization to im- prove the image contrast. Due to the ECG pulsing windows applied in the acquisition and reduced radiation dose [8], the signal-to-noise ration is varying in the multiple-phase 4D data sets. To improve the image quality, anisotropic filtering is used to reduce this noise and preserve the cardiac chamber boundaries and vessel structures. Using Euclidean distance as the heuristic term, A* graph search 1053 978-1-4244-4126-6/10/$25.00 ©2010 IEEE ISBI 2010