Registration of Tomographic Images to X-ray Projections for Use in Image Guided Interventions

Student thesis: Doctoral ThesisDoctor of Philosophy


This thesis contributes to knowledge by describing a new method to allow
information from a pre-operative 3D modality to be used during an intervention which is guided using 2D fluoroscopy images.

An algorithm has been designed to obtain the pose of a CT volume with respect to a single fluoroscopy image. The registration algorithm is based on the production of digitally reconstructed radiographs from the CT volume, which are compared to the fluoroscopy image using a similarity measure.
The novelty of the work described in this thesis is in both the design of
the registration algorithm and also in investigating the specific requirements placed upon a similarity measure when attempting to register a pre-operative CT volume to an interventional fluoroscopy image. Seven similarity measures were investigated. Experiments were carried out to calculate the accuracy and robustness of the registration algorithm using each similarity measure. Initially fluoroscopy and CT images of a lumbar spine phantom were used. The accuracy of the registration algorithm was calculated by comparing the final registration positions with a ``gold-standard'' registration calculated using fiducial markers. More realistic datasets were simulated using the phantom fluoroscopy image with clinical image features overlaid.
Results show that the introduction of soft-tissue structures and interventional instruments into the phantom image can have a large effect on the performance of some similarity measures previously applied to 2D-3D image registration. The similarity measures were also tested on clinical
data from aortic stenting procedures, where k-fold cross validation was
used to obtain an estimate of the registration accuracy. The results from these experiments showed that two measures were able to
register accurately (RMS rotational error of 0.76 degrees and RMS in-plane translational error of 0.85mm) and robustly (10% failure rate) even when soft-tissue structures and interventional instruments were present as differences between the images. These two measures were pattern intensity and gradient difference.

Finally the thesis describes a novel combination of the 2D-3D registration
algorithm with a deformation algorithm. The registration algorithm was used
to obtain information on the relative movement of the vertebrae between
the pre-operative CT image and interventional fluoroscopy image. This
information was then used to warp the pre-operative modality so that it more
accurately represented the intra-operative scene.
Date of Award2000
Original languageEnglish
Awarding Institution
  • King's College London
SponsorsPhilips Healthcare
SupervisorDavid Hawkes (Supervisor) & Derek Hill (Supervisor)

Cite this