Rapid 3-D cone-beam reconstruction with the simultaneous algebraic reconstruction technique (SART) using 2-D texture...

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Science

Background:

• Algebraic reconstruction techniques (ART) like SART offer advantages over Filtered Backprojection for 3-D cone-beam reconstruction.
• The iterative nature of ART methods leads to slow computation, limiting clinical use.

Purpose of the Study:

• To accelerate 3-D algebraic reconstruction using readily available 2-D texture mapping graphics hardware.
• To investigate the impact of framebuffer precision on reconstruction quality and propose a solution.

Main Methods:

• Utilized 2-D texture mapping graphics hardware to accelerate 3-D algebraic reconstruction.
• Implemented a scheme to extend framebuffer precision by 4 bits using color channels.
• Compared reconstruction quality with 8-bit and 12-bit framebuffers for varying tissue contrasts.

Main Results:

• Achieved speed-ups exceeding 50x compared to CPU-only implementations, enabling near-interactive 3-D cone-beam reconstruction.
• Demonstrated that reconstruction quality is sensitive to framebuffer resolution.
• Showed that a 12-bit framebuffer with extended precision yields useful reconstructions at 0.5% tissue contrast, compared to 4% for an 8-bit framebuffer.

Conclusions:

• Graphics hardware can significantly accelerate 3-D algebraic reconstruction, making it viable for clinical applications.
• Extending framebuffer precision is crucial for maintaining reconstruction quality, especially for low-contrast targets.
• SART is well-suited for hardware acceleration due to its volume correction granularity.