Fusion of deformable anatomical structures of human torso.

摘要

In order to construct a personalized tetrahedron mesh model of human trunk including spinal column, rib cage, soft tissue and external trunk skin surface from clinical datasets via less-invasive 3D acquisition techniques, many steps of work are needed.; The specific objective of this project is to align and fuse the external and internal human trunk surface models. We have investigated and compared several landmark-based registration methods including rigid, affine, Thin-Plate Spline (TPS) interpolation and approximation. We use the Bidirectional Point-to-Surface Mean Distance (BPTSMD) metric to estimate the deformation errors from the difference of postures. Lots of comparisons have been done and the results show that the local elastic deformation (TPS interpolation and approximation) methods are much better than the global transformation (rigid, affine) and suitable for registering the flexible external and internal models of the human trunk. And we also have used Brent Minimization algorithm to find the optimal trade off parameter for the TPS approximation to account for the landmark localization errors. (Abstract shortened by UMI.)