We demonstrate an improvement to cone-beam tomographic imaging by using a prior anatomical model. A protocol
for scanning and reconstruction has been designed and implemented for a conventional mobile C-arm:
a 9 inch image-intensifier OEC-9600. Due to the narrow field of view (FOV), the reconstructed image contains
strong truncation artifacts. We propose to improve the reconstructed images by fusing the observed x-ray
data with computed projections of a prior 3D anatomical model, derived from a subject-specific CT or from a
statistical database (atlas), and co-registered (3D/2D) to the x-rays.
The prior model contains a description of geometry and radiodensity as a tetrahedral mesh shape and density
polynomials, respectively. A CT-based model can be created by segmentation, meshing and polynomial fitting of
the object's CT study. The statistical atlas is created through principal component analysis (PCA) of a collection
of mesh instances deformably-registered (3D/3D) to patient datasets.
The 3D/2D registration method optimizes a pixel-based similarity score (mutual information) between the
observed x-rays and the prior. The transformation involves translation, rotation and shape deformation based on
the atlas. After registration, the image intensities of observed and prior projections are matched and adjusted,
and the two information sources are blended as inputs to a reconstruction algorithm.
We demonstrate recostruction results of three cadaveric specimens, and the effect of fusing prior data to
compensate for truncation. Further uses of hybrid reconstruction, such as compensation for the scan's limited
arc length, are suggested for future research.