Efficient digitally reconstructed radiographs with middle-out rendering and convex hull-based early stopping

Abstract

Kinematic tracking of native anatomy in dynamic radiography facilitates understanding of in vivo movement. Tracking is performed using model-image registration, which involves estimating 6 degree-of-freedom poses of anatomic structures by comparing captured radiographs to dynamically rendered images of digital models. This procedure can produce accurate pose estimates of native anatomy, but computational efficiency remains a concern. A key source of latency in model-image registration is the iterative rendering of digitally reconstructed radiographs, required for evaluation of candidate poses. This technical note introduces an efficient algorithm for rendering digitally reconstructed radiographs. The proposed method is shown to accelerate rendering speeds by up to 50% compared to a baseline method which is commonly used in registration frameworks. Efficient rendering of digitally reconstructed radiographs may enhance the overall viability of model-image registration for use in kinematic tracking, and may contribute to bringing this technology closer to adoption in clinical settings.