METHODS, SYSTEMS, AND COMPUTER READABLE MEDIA FOR REAL-TIME 2D/3D DEFORMABLE REGISTRATION USING METRIC LEARNING

摘要

Methods, systems, and computer readable media for real-time 2D/3D deformable registration using metric learning are disclosed. According to one aspect, a method for real-time 2D/3D deformable registration using metric learning includes creating a catalogue of simulated 2D projection images based on a reference 3D image and a shape space of 3D deformations, where each entry in the catalogue is created by: applying to the reference 3D image a set of deformation parameters from the shape space of deformations; simulating a 2D projection of the result; associating the simulated 2D projection image with the deformation parameters used to create the simulated 2D projection image; and storing the simulated 2D projection image and associated deformation parameters in the catalogue. The method also includes receiving a 2D image, and, in response to receiving the 2D image: calculating a value of distance between the received 2D image and a simulated 2D projection image for each of the simulated 2D projection images in the catalogue; using the calculated distances to calculate weighting factors to be applied to the deformation parameters of each of the simulated 2D projection images in the catalogue; and calculating deformation parameters for the received 2D image based on the weighted deformation parameters in the catalogue. The calculated deformation parameters are then used to deform a 3D volume of interest to produce a 3D volume that represents the 3D layout of the tissue at the time that the received 2D image was acquired.