Automated atlas-based pelvimetry using hybrid registration

摘要

An automated atlas-based pelvimetry framework is proposed using hybrid registration, with the aim of examining the relationship between pelvic inlet area and the difficulty of laparoscopic operations for prostate cancer. The atlas is built with a 3D pelvis CT scan with a corresponding manual segmentation of pelvic inlet contour. The transformation between the atlas image and the target images is automatically built by applying a hybrid registration incorporating affine registraion with a statistical deformation model constrained non-rigid registration. The inlet contour from the atlas can be propagated to the target individual to estimate the inlet area. Then, Eigenmode analysis is introduced to define the inlet plane by the first two principal axes of the transformed points and all points are projected onto this plane. A triangulation of the planar points is used to calculate the pelvic inlet area. The result shows that our method works well for automated pelvic inlet area measurement with average relative error 2.82% ± 0.0167 compared with the gold standard defined by an expert.