Model-based Measurements in Digital Radiographs

Ph.D. Thesis, 2000, Department of Computing Science, Umeå University, Sweden.


This thesis is about measurements in radiographs from two medical applications; orthopaedics and prosthetic dentistry. However, some of the methods and theory also applies to other types of images.

The thesis is divided into seven parts; papers I-V and appendices A-B.

Papers I-IV and appendices are related to Radiostereometric Analysis (RSA), a stereographic measurement method used mainly in orthopaedics. RSA is today the method of choice for studies of micromotion and wear of orthopeadic implants in vivo. RSA is based on markers implanted into the patient. Classically, the measurements of marker coordinates in the X-ray films have been performed manually using a high-precision measurement table. Paper I describes a technique for performing the measurements in digital radiographs, where a non-linear model of the image intensity near a marker is least squares fitted to image data. Models of different scenarios are presented, including a model of a nearby interfering edge. The setup also facilitates selective exclusion of individual pixels. In Paper II, the measurement method in Paper I is shown to maintain or improve the clinical precision of the overall RSA method compared to manual measurements as measured by double examinations of knee and hip patients. These results are consolidated in Paper III, which concentrates on RSA examination using the uni-planar cage. The observer dependence is also studied. Paper IV presents an effort to model the photogrammetric calculations throughout the whole RSA method, including the reconstruction of the projection geometries and motion calculation.

Paper V presents a method for alignment of intra-oral radiographs for subtraction radiography, where the dental implant under study is utilized as a reference object. A simple geometric model of the projected thread outline is fitted to the image. A coordinate system is defined based on the model fit and used to provide a geometric correction between two exposures of the same implant. The physical size of the implant is used to normalize the coordinate system, making it possible to define regions of interest in physical units relative to the implant. The method is shown to be insensitive to a large variation of exposure conditions in vitro. Together with a standard method for contrast correction, the method is shown to be able to detect clinically invisible simulated bone density changes near the implant.

Appendix A is included as an example of what RSA is used for and how it is presented to the orthopaedic community.

Appendix B is included to indicate the programming effort behind the measurement program in RSA that is another result of the work behind this thesis.
Last modified: Tue Jul 25 13:38:58 MET DST 2000