High performance cone-beam spiral backprojection with voxel-specific weighting

Area of Science:

• Medical imaging physics
• Computational algorithms
• Image reconstruction

Background:

• Cone-beam spiral backprojection is computationally intensive, posing a challenge for efficient image reconstruction.
• Traditional methods struggle with voxel-specific weights and massive data storage requirements for spiral scans.
• Existing algorithms like Feldkamp are limited by computational demands and data handling.

Purpose of the Study:

• To develop a novel, highly efficient algorithm for cone-beam spiral backprojection.
• To overcome the computational and storage limitations of existing backprojection methods.
• To significantly accelerate image reconstruction for spiral CT scans.

Main Methods:

• Proposed a new algorithm that exploits spiral symmetry by backprojecting into rotated slices.
• Utilizes 64-bit operating systems for precomputing and storing large amounts of voxel-specific weights.
• Employs data-level parallelism and high vectorization for optimized performance.

Main Results:

• Achieved up to 17 giga voxel updates per second on a multi-CPU system.
• Demonstrated an order of magnitude speed improvement over highly optimized non-spiral-aware codes.
• Reconstructed 344 images per second under specific slice and projection parameters.

Conclusions:

• The novel spiral backprojection algorithm offers a significant speed-up for image reconstruction.
• The method's efficiency makes it a versatile component for future image reconstruction pipelines.
• Potential for further development into ray-driven versions and high-performance forward projectors.