Efficiently computing and registering digitally reconstructed radiographs

Generation of digitally reconstructed radiographs (DRRs) is computationally expensive and has been cited as the rate-limiting step in the execution time of intensity-based two-dimensional/three-dimensional registration algorithms. This paper considers the problem of generating DRRs by conventional ray tracing. Experiments confirm that good quality reconstructions can be obtained using this approach in a few seconds. We evaluate the approach for automatic patient setup prior to radiotherapy treatment by performing intensity based 2D-3D registration using normalized cross correlation. Preliminary results using a pelvic CT data set show the method is accurate to about ±2 pixels (i.e. ±0.3 mm).