A Fast DRR Generation Scheme for 3D-2D Image Registration Based on the Block Projection Method

Zhiping Mu; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2016, pp. 169-177

In three-dimensional to two-dimensional (3D-2D) image registration, DRR (digitally reconstructed radiograph) generation is often a bottleneck in computation. In this article, a novel fast DRR generation scheme is proposed based on the recently introduced Block Projection method and Slab algorithm that reuse building blocks of DRRs previously generated for known poses. The scheme is flexible as exemplified in pose grid design and slab binding, and upper bounds in projection error exist and can be estimated. Experiments were conducted to evaluate DRR quality and sensitivity to pose difference; computing time and error bounds were reported. The results showed that on a conventional computer the proposed scheme generated high quality, pose-preserving DRRs of size 512x512 in 6 ms with slab binding, demonstrating its potential to be a viable solution to fast, high quality DRR generation for 3D-2D image registration.