Matthias Baer Dipl Phys, Presenter: Nothing to Disclose

Stefan Sawall DIPLENG, Abstract Co-Author: Nothing to Disclose

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

Marc Kachelrieß PhD, Abstract Co-Author: Nothing to Disclose

To provide images free of cone-beam artifacts from spiral cone-beam CT scans for clinical, interventional, and preclinical CT scanners in real-time.

Today’s clinical CT scanners simultaneously acquire between 64 and 320 slices, flat detector-based systems such as C-arm scanners or micro-CT scanners acquire up to 2048 slices. Due to the increased cone-angle significant cone-beam artifacts may become apparent if the reconstruction algorithm is not designed properly. Recently, new approaches based on Katsevich’s inversion formula were developed. These algorithms are exact and free of cone-beam artifacts but more complicated than conventional approaches. To improve reconstruction speed we designed a high-performance implementation of an exact cone-beam CT algorithm based on rebinning to K-lines. Our implementation uses a data-level and compute-level parallelization during forward rebinning, convolution and backprojection. Due to a sophisticated spiral backprojection algorithm we can store the complete system matrix. Since the backprojection allows to precompute and store all coordinate look-up tables we further use an optimized voxel-specific weight that smoothly fades data at the edge of the Tam window on a voxel-by-voxel basis. Our implementation is evaluated using patient data acquired with a clinical 128-slice dual source CT scanner and with a preclinical 1024-slice dual source micro-CT scanner.

Our implementation reconstructs about 50 images per second (512² pixels, 512 projections per 360°) on a standard two-socked quad-core CPU running at 3 GHz. 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 a conventional reconstruction while cone-beam artifacts are completely removed.

Due to the high parallelization level of our approach, exact spiral cone-beam CT image reconstruction allows to provide images free of cone-beam artifacts in real time.

Fast image reconstruction has the potential to enhance the clinical throughput and to improve image quality when being used as a component of iterative image reconstruction.

Baer, M, Sawall, S, Knaup, M, Kachelrieß, M, High Performance Exact Spiral Cone Beam CT Image Reconstruction.  Radiological Society of North America 2010 Scientific Assembly and Annual Meeting, November 28 - December 3, 2010 ,Chicago IL. http://archive.rsna.org/2010/9005140.html