Geometrical transformation approximation for 2D/3D intensity-based registration of portal images and CT scan

Abstract

Conformal radiotherapy treatments need accurate patient positioning in order to spare normal tissues. Patient pose can be evaluated by registering portal images (PI) with Digitally Reconstructed Radiographs (DRR). Several methods involve segmentation which is known to be a difficult task for noisy PI. In this paper, we study another approach by using a fully 3D intensity-based registration method, without segmentation. Our approach uses the correlation ratio as similarity measure and replace DRR generation with a treatment on pre-computed DRR. A specific geometrical transformation is applied to approximate a given projection by the composition of out-of-plane rotations and inplane transformation. Some preliminary experiments on both simulated and real portal images, lead to good results (RMS error lower than 2 mm).