An Automatic Method For Physical-to-image/image-to-physical Geometric Coordination Registration

In oral implant surgery planning,surgeons usually perform trial operations on virtual or rapid prototype stereoscopic lithography models.The planning can now be done in the computer environment completely,the planned physical surgical stents can be fabricated directly with 3D printing.During the planning and post-operation assessment,multiple images which possibly acquired from various imaging modalities are registered.As all the imaging modality exhibits different level of error during image acquisition,image to image registration in pair does not give estimation of error to physical domain.A new geometric coordination registration for registering physical model and virtual image was proposed in recent year.This enables simple linear and coordinate measurement of physical and virtual domain.The idea is to directly register the 3D images to the physical domain based on some geometric structures fabricated on the surgical stent.The aim of this study is to develop a fully automated registration method based on some geometric structures,as fiducial markers,attached to surgical stent.The detected geometric markers are then used to register 3d images generated by scanners such as cone-beam computed tomography（CBCT）,intraoral scanner（IOS）,and extraoral scanner（EOS）to the reference coordinate axes.In the current study,a Cubic Corner marker（CC）and two Discs markers are adopted,which are easy to be fabricated and easy to be detected by using computer programs.The methodology of the proposed fully automated registration method is briefly summarized as follows:（1）Marker target recognition: Voxelize the input point cloud data for constructing local feature descriptors.The proposed descriptor is based on the distribution of vertices and normal vectors.It is used to locate the target surfaces,cube-corner edges,origin,the center of discs,etc of the geometric markers.（2）Registration based on marker recognition results: The feature points of the located markers are the used to derive the 3D transformation to register the input model to the Cartesian coordinate axes in the physical domain.To verify the performance of the method proposed in this study for clinical use,we designed the following experiments: On 30 maxillary resin models collected from dentate patients with dental implants inserted in the incisor area.3D images of each impression were acquired using CBCT,EOS and IOS scanners.The coordinates of both ends（apex end and neck end）of the model implant were measured by a coordinate measuring machine（CMM）in the physical domain as the ground truth value.Compare the target registration error（TRE）of the proposed method and the semi-automatic registration method.The results show that the average TRE of the method proposed in this thesis is 0.4mm,which is lower than the traditional semi-automatic registration method（TRE 0.52mm）,and it meets the requirements for clinical diagnosis.