FAST AND HIGH PRECISION DIGITALLY RECONSTRUCTED RADIOGRAPH GENERATION ON GRAPHIC PROCESSING UNIT

Abstract

Digitally reconstructed radiograph (DRR) from CT volumetric data has been used in numerous medical applications such as 3D treatment planning and CT-to-fluoroscopic alignment. The poor efficiency of the DRR generation is the main problem in such applications. Many researches have been attempted to accelerate the DRR calculation. However, the performance and precision cannot be achieved without the sacrifice of one or the other. In this study, a fast and high precision DRR generation method is proposed on a consumer PC platform. Instead of using CPU, the method takes the advantages of the powerful parallel computation and flexible programming capability of the graphic processing unit (GPU) to reach almost interactive rendering rate while maintaining 12-bit precision of the original CT data. This method can generate DRR images at 4.6 frames per second using 512 × 512 × 261 dataset in the 512 × 512 view port, and its precision is compatible to that generated by the CPU-based method. Besides, in order to simulate clinical radiograph images, a compensation filter is implemented in the DRR generation to compensate varying thickness of bone structures. The additional compensation filter can achieve a DRR image with more uniform optical density but takes no obvious performance overhead.