tomoCAM: fast model-based iterative reconstruction via GPU acceleration and non-uniform fast Fourier transforms

Area of Science:

• Materials Science
• Physics
• Computer Science

Background:

• X-ray computed tomography (CT) is crucial for 3D structure determination.
• Synchrotron advancements enable faster, higher-resolution CT.
• New experimental needs challenge traditional CT reconstruction methods.

Purpose of the Study:

• To develop an efficient implementation of model-based iterative reconstruction (MBIR) for challenging CT scenarios.
• To overcome computational limitations of existing MBIR algorithms.
• To provide accessible, high-quality 3D reconstructions for materials science.

Main Methods:

• Introduced tomoCAM, a GPU-accelerated MBIR implementation.
• Utilized non-uniform fast Fourier transforms for efficient Radon and back-projection.
• Employed asynchronous memory transfers for maximized GPU throughput.

Main Results:

• tomoCAM significantly outperforms traditional MBIR codes in speed.
• Achieved high-quality reconstructions despite limited projection data.
• Demonstrated MBIR benefits with affordable computing resources.

Conclusions:

• tomoCAM offers a fast and accessible solution for advanced CT image reconstruction.
• Enables in situ and in operando studies with complex sample geometries.
• Python front-end facilitates integration into synchrotron workflows.