MO-D-ValB-02: OpenGL Based 2D-3D Registration of a CT Image Dataset With OBI Images

B Choi,P Balter, L Dong, R Mohan,G Starkschall

MEDICAL PHYSICS(2006)

引用 0|浏览2
暂无评分
摘要
Introduction. A limiting factor in registration of a three‐dimensional (3D) computed tomography(CT) dataset with two‐dimensional (2D) x‐ray images has been the time‐consuming generation of digitally reconstructedradiographs(DRRs). This can be overcome using a commercial graphics card for DRR generation enabling fast, robust, and accurate automatic image‐based 2D‐3D registration. Methods and Materials. For the iterative registration process hundreds of DRRs are created using hardware rendering in OpenGL. Each DRR from a 512×512×100 CT volume is rendered in less than 0.1 seconds using an nVidia 7800GT graphics card. The registration is based on a publicly available implementation of Mattes Mutual Information (ITK, U.S. National Library of Medicine, Bethesda, MD). To improve speed, the registration is performed on a sub‐image. A two‐step registration strategy is adopted for robustness with the first pass using a larger margin around the sub‐image and a down‐sampled resolution. A thoracic phantom (Model 602, CIRS, Norfolk, VA) was imaged and setup according to our clinical protocol and then shifted from 0–1.5 cm along each of the major axes. Anterior‐Posterior and Lateral kV x‐ray images were acquired using a commercial patient imaging system (OBI, Varian Medical System, Palo Alto, CA). Results: The mean registration times were 8 and 16 seconds without and with rotations respectively. We observed a systematic 1.1 mm offset in the longitudinal direction that we believe results from 2.5 mm CT slice spacing and OBI calibration. With this removed the mean three‐dimensional distance of the registered positions from the phantom positions was 0.4 mm with the largest disagreement being 0.75 mm. The systems ability to calculate rotations was only tested numerically. Conclusions. The speed and accuracy of this system demonstrate that it could be a viable tool for reducing daily setup uncertainty by automating the analysis of setup images. Research sponsored by Phillips Medical Systems.
更多
查看译文
关键词
ct image dataset,opengl,registration,mo-d-valb
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要