Intensity-based 2D/3D rigid registration of fluoroscopic X-ray to CT

Dotan Knaan

Masters of Science Thesis, 2003

School of Engineering and Computer Science
The Hebrew University of Jerusalem


Registration is an essential step in most computer-aided surgery systems.
It consists of finding a transformation from the coordinate system of one
data set to another so that all features that appear in both are aligned.
Registration is necessary to match information from different data
modalities obtained at different times.  One of the most sought after
methods is anatomy image-based rigid registration between preoperative CT
and intraoperative X-ray images. This type of registration enables surgeons
to use preoperative spatial data and plans in the operating room for
image-guided navigation and robot positioning with no need for fiducial
markers or contact with the anatomy.

This thesis presents an intensity-based algorithm for anatomy image-based
rigid registration between preoperative CT and intraoperative X-ray
images. Intensity-based algorithms iteratively compute the transformation
by matching the X-ray image intensities to those of simulated X-ray images,
called Digitally Reconstructed Radiographs (DRRs), generated from the CT at
estimated viewpoints, and reducing their differences.  The algorithm
proposed in this thesis improves upon existing methods and presents new
techniques to overcome common robustness, accuracy, and computation time
problems.  For speed, it generates Digitally Reconstructed Radiographs on
small, dynamically selected regions of interest from precomputed ray gray
levels in expected viewing directions, and uses a multiresolution hierarchy
of fluoroscopic X-ray images.  For robustness and accuracy, it uses a
two-step comparison measure: Normalized Cross Correlation followed by
Variance Weighted Sum of Local Normalized Correlation.  To avoid local
minima, it uses a genetic search method.

The effectiveness of the algorithm is demonstrated by incorporating it in
the registration process and validating it with simulated, in-vitro, and
cadaver studies.  The experimental results show that anatomy image-based
rigid registration between fluoroscopic X-ray and CT with an overall mean
Target Registration Error of 1mm (max 2mm), 95\% of the time, with three
X-ray images in less than two minutes in a realistic setup, is practically
feasible.