Sven Steckmann, Abstract Co-Author: Nothing to Disclose

Olivier Bockenbach, Abstract Co-Author: Employee, Mercury Computer Systems, Inc

Michael Knaup PhD, Abstract Co-Author: Nothing to Disclose

Marc Kachelriess PhD, Presenter: Nothing to Disclose

To reduce the computational requirements for spiral CT image reconstruction.

Cone-beam spiral CT image reconstruction is computationally highly demanding — most computations are required for the backprojection step. At first sight, its requirements appear similar to cone-beam backprojection from circular scans such as in the widely used Feldkamp algorithm. However, there is an additional complication: the illumination of each voxel, i.e. the range of angles the voxel is seen by the x-ray cone is a complex function of the voxel position. A voxel-specific weight w(x, y, z, α) needs to be multiplied to each voxel (x, y, z) at each projection angle α. The weight function has no analytically closed form and must be numerically determined. Storage of the weights is prohibitive due to the large number of entries. We propose a new algorithm that combines the spiral symmetry with the ability of today's 64 bit operating systems to store large amounts of precomputed weights. Our approach uses slices that rotate around the z-axis in the same way as the spiral trajectory. This symmetry removes z-dependence from the weight function and it further allows to store the detector channel and row number in a look-up table rather than computing it from scratch. In a final step the images are rotated back into their desired horizontal position. To compensate for the blurring introduced in this final step the convolution kernel is adapted to enhance higher frequencies. Image quality is evaluated using simulated projection data and patient data acquired with a clinical CT scanner.  

Our backprojection processes 57 images per second (512² pixels, 512 projections) on a standard two-socked quad core CPU. Thereby it is seven times faster than our optimized reference code that does not make use of the spiral symmetry. Spatial resolution and image noise are identical to the slower reconstruction which does not need the final rotation step. Visual inspection of the patient images does not show any differences.

Fast spiral cone-beam CT image reconstruction is possible with standard PCs using our approach. The speed-up can also be used to enhance iterative image reconstruction where a forward projection step is involved.

Faster image reconstruction allows to enhance the clinical throughput or to improve the image quality and dose usage by applying more complex algorithms.

Steckmann, S, Bockenbach, O, Knaup, M, Kachelriess, M, Fast Cone-Beam Spiral CT Image Reconstruction.  Radiological Society of North America 2008 Scientific Assembly and Annual Meeting, February 18 - February 20, 2008 ,Chicago IL. http://archive.rsna.org/2008/6010131.html