Estimating 3D respiratory motion from orbiting views

Abstract

This paper describes a method for estimating 3D respiratory motion so as to characterize tumor motion. This method uses two sets of measurements. One is a reference thorax volume obtained from a conventional fast CT scanner under breath-hold condition. The other is a sequence of projection views of the same patient (acquired at treatment time) using a slowly rotating cone-beam system (1 minute per rotation) during free breathing. We named this method deformation from orbiting views (DOV). Breathing motion over the entire acquisition period is estimated by deforming the reference volume through time so that its projections best match the measured projection views. The nonrigid breathing motion is described by a B-spline based deformation model. The parameters of this model are estimated by minimizing a regularized squared error cost function, using a conjugate gradient descent algorithm. Performance of this approach was evaluated by simulation. Results showed good agreement between the estimated and synthesized motion, with a mean absolute error of 1.63 mm. Relatively larger errors tended to occur in uniform regions, which would not have significant effects on generating deformed volumes based on the estimated motion. The results indicate that it is feasible to estimate realistic nonrigid motion from a sequence of slowly rotating cone beam projection views.