Thesis Details

Model-based 2D-3D Registration Methods for Analysis of Conventional Radiographs

Ph.D. Thesis Student: Klíma Ondřej Academic Year: 2021/2022 Supervisor: Zemčík Pavel, prof. Dr. Ing.
Czech title
Metody 2D-3D registrace modelů pro analýzu konvenčních rentgenových snímků

Conventional radiography plays a key role within bone fracture diagnoses, preoperative planning, and postoperative evaluations of orthopedic interventions. The thesis is focused on methods enabling 3D analysis of conventional radiographs, based on the registration of digital models into pairs of calibrated X-ray images.The thesis presents a method for reconstruction of complete intact models of long bones suffering displaced diaphyseal fractures from conventional radiographs of individual fragments. The reconstruction is achieved using a 2D-3D registration of a statistical shape model into the fragments with simultaneously performed fracture reduction. The reduction is enabled by accurate estimation of the bone length, which is achieved by an automatic division of the statistical shape model into individual fragments. The proposed approach is involved in a software application intended for preoperative planning of diaphyseal fractures reduction with a focus on the identification of the best fitting intramedullary nail or bone plate.The reconstruction adopts a proposed intensity-based revision of nonoverlapping area registration procedure, intended for involvement in model-based radiostereometry. In contrast with the original contour-based formulation, the revised method is able to handle occlusions or unreliable parts of involved digital models. Moreover, performed evaluations reveal an order of magnitude higher accuracy in comparison with the contour-based approach, reaching similar results as the state-of-the-art feature-based approaches, while allowing the involvement of highly detailed 3D models and straightforward acceleration.In addition to the non-overlapping area approach, also a density-based registration pipeline using statistical shape and intensity models was created for reconstruction of patient-specific bone models. Although reaching slightly lower surface reconstruction accuracy, the density-based registration is able to estimate the internal bone structures such as spongy and compact tissues, potentially providing more information for the planning. For practical aspects, different optimisation methods and possibilities of statistical shape and intensity models lossy compression were investigated.


Model-based 2D-3D registration, radiostereometric analysis, preoperative osteosynthesis planning

Degree Programme
Back to top