Abstract
A tomosynthesis-based prostate brachytherapy seed localization method is described. Gaussian-blurred images are computed from a limited number of X-ray images, and a 3-D volume is reconstructed by backprojection. Candidate seed locations are extracted from the reconstructed volume and false positive seeds are removed by optimizing a local cost function. In case where the estimated pose error is large, a self-calibration process corrects the estimation error of the intrinsic camera parameters and the translation of the pose in order to improve the reconstruction. Simulation and phantom experiment results imply that the implanted seed locations can be estimated from four or five images depending on the number of seeds. The algorithm was also validated using patient data, successfully localizing the implanted seeds.