Intensity-Based 3D-2D Mesh-to-Image Registration Using Mesh-Based Digitally Reconstructed Radiography

Abstract

Intensity-based 3D-2D registration is a well-established technique shown to be effective for many clinical applications. However, it is valid mainly for 3D Computed Tomography (CT) volume to 2D X-ray image registration because the computation of volume-based Digitally Reconstructed Radiography (DRR) relies on the linear relationship between CT’s intensity and the attenuation coefficient of the underlying structure for X-ray. This paper introduces a mesh-based DRR renderer that simulates realistic-looking X-ray images from 3D meshes, which can be used to replace conventional volume-based DRR in intensity-based 3D-2D registration for 3D volumes from various image modalities. The proposed renderer calculates the travel distance of a given ray within the mesh, and computes X-ray attenuation based on the travel distance and the object’s attenuation property. The proposed method also uses a novel ray-casting strategy that takes GPU architecture into consideration for high computational efficiency. Validation results show that the proposed mesh-based DRR simulates X-ray images with a high fidelity, and intensity-based 3D-2D registration using the resulting mesh-based DRR achieves satisfactory results on clinical data.