Automatic generation of surface meshes for right ventricle with 1-to-1 correspondence from cine-MR images

Abstract

We develop an automatic method to generate a set of 4D 1-to-1 corresponding surface meshes of the right ventricle (RV) which are motion registered over the whole cardiac cycle. The inputs are a set of 3D RV surface meshes at different phases of the cardiac cycle, each reconstructed independently from border-delineated MR images. To generate point correspondence, a template mesh is matched to the actual shape of the RV meshes in each of the time phases. This is done via a coarse matching phase and a fine matching phase. In the former, an initial rough matching between the template and the target is achieved using a radial basis function (RBF) morphing process. The feature points on the template and target meshes are automatically identified using a parameterization method. In the latter, a progressive mesh projection process is used to conform the rough estimate to fit the exact shape of the target. In addition, an optimization-based smoothing process is used to achieve superior mesh quality and continuous point motion. Eight healthy volunteers were recruited for MRI scanning and the algorithm was tested on the acquired data. It was observed that the algorithm took an average of approximately 70 seconds to complete. The maximum absolute deviation of the matched model from the original geometry was 0.187mm.