High Performance 2D/3D-Registration for Patient Positioning in Radiotherapy

Abstract

The precise and repeatable positioning of patients is one of the key elements of successful radiation therapy. Our aim is to develop an automated system, which determines the exact position of the patient. We use volume rendering based on 3D-textures to generate a 3D-model of the region of interest from pre-treatment CT-data. The idea is to generate before each treatment session additional x-ray images from orthogonal positions. These x-ray images represent the current position of the patient. With the help of projection images of the 3D-model, so-called digitally reconstructed radiographs (DRRs), we determine the rigid transformation between the 2D-x-ray images and the 3D CT data via an intensity based 2D/3D registration process. This represents the misalignment between the current position and the position in the treatment plan. For our simulation results we used as a first step ray-casting images instead of real x-ray images. Because of the high performance of our approach, we were able to investigate, which optimization methods and which number of orthogonal x-ray tubes give best results.