Robust and Fast 2D/3D Image Registration using Regression Learning

In computer vision and image analysis, image registration between 2D projections and a 3D image while obtaining high accuracy and real-time computation is challenging. In this paper, we propose a novel method that can speedily detect the object’s 3D rigid motion or deformation from a small set of its 2D projection images. The method consists of two stages: registration and pre-registration learning. In the registration stage, it iteratively estimates the motion/deformation parameters based on the current intensity residue between the target projection and the projection of the estimated 3D image using learned linear operators. The linear operators are learned in the pre-registration learning stage: First, it builds a low-order parametric model of the image region’s motion/deformation shape space from its prior 3D images. Second, using learning-time samples produced from the 3D images, it formulates the relationships between the model parameters and the co-varying 2D projection intensity residues by multi-scale linear regressions. The calculated multi-scale regression matrices give the coarse to fine linear operators used in estimating the model parameters from the 2D projection intensity residues in the registration. The method’s application to Image-guided Radiation Therapy (IGRT ), called CLARET (Correction via Limited-Angle Residues in External Beam Therapy), requires only a few seconds and has given good results in localizing a tumor under rigid motion in the head and neck and under respiratory deformation in the lung using a small set of treatment-time imaging 2D projections.