Fast DRR generation for intensity-based 2D/3D image registration in radiotherapy

Conformal radiotherapy requires accurate patient positioning according to a reference given by an initial 3D CT image. Patient setup is controlled with the help of portal images (PI), acquired just before patient treatment. To date, the comparison with the reference by physicians is mostly visual. Several automatic methods have been proposed, generally based on segmentation procedures. However, PI are of very low contrast, leading to segmentation inaccuracies. In the present study, we propose an intensity-based method, with no segmentation, associating two portal images and a 3D CT scan to estimate patient positioning. The process is a 3D optimization of a similarity measure in the space of rigid transformations. To avoid time-consuming DRRs (Digitally Reconstructed Radiographs) at each iteration, we used 2D transformations and sets of DRRs pre-generated from specific angles. Moreover, we propose a method for computing intensity-based similarity measures obtained from several couples of images. We used correlation coecient, mutual information, pattern intensity and correlation ratio. Preliminary experiments, performed with simulated and real PIs, show good results with the correlation ratio and correlation coecient (lower than 0.5 mm median RMS for tests with simulated PI and 1.8 mm with real PI).