Deformable Registration using Spring Mass System with Cross-section Correction

When applying 2D deformable registration to crosssectional images from 3D data, it is assumed the images represent the same object. However, in reality, they are acquired at different axial positions and with different coronal and sagittal rotations. Therefore, if the crosssection's position and orientation can be corrected, it should offer better registration accuracy. The paper describes a "2.5D" method for correcting position and orientation of the cross-section under the in-plane 2D deformable registration. For a given source image, the algorithm estimates the optimal position and orientation of the cross-section generating target image as well as displacement field of the source image. For in-plane 2D deformable registration, a previously proposed method is being used. It models the deformations using a spring mass system, which contains a number of sparse masses interconnected by springs. The external forces calculated from the image contents are used to iteratively guide the system to the "good" matching configuration under the constraints imposed by the internal forces. The performance of the method is tested using real radiation therapy planning CT (RTPCT) and cone beam CT (CBCT) as well as the simulated data. The results indicate that the newly proposed method improves registration accuracy.