tofu

tofu performs high-throughput image processing and tomographic reconstruction of computed tomography (CT) data to produce accurate 3D reconstructions across multiple imaging geometries.

Key Features:

• Workflow management: Organizes complex image processing tasks into structured workflows that manage individual processing steps.
• Supported geometries: Supports parallel beam, cone beam, tomography, and laminography imaging geometries.
• Pre- and post-processing algorithms: Implements phase retrieval, ring removal, and de-noising algorithms for improved reconstruction quality.
• GPU optimization: Optimized to run on stand-alone GPU workstations and to efficiently utilize all available GPUs in a system.
• Code generation: Generates 3D reconstruction code with minimal instructions based on the input geometry to optimize runtime performance.
• High-performance reconstruction: Achieves reconstruction speeds comparable to those of CPU clusters.
• Batch processing: Enables batch processing of datasets using pre-configured workflows.

Scientific Applications:

• Computed tomography (CT) reconstruction: 3D reconstruction for parallel-beam and cone-beam CT datasets.
• Laminography: Tomographic reconstruction for laminography experiments.
• Phase-contrast imaging: Phase retrieval for phase-contrast tomography to enhance image contrast.
• Artifact correction and denoising: Ring removal and de-noising for improved image fidelity prior to analysis.
• Medical imaging and bioinformatics research: Processing and reconstruction of CT data for research applications across custom-built and off-the-shelf hardware.

Methodology:

Implements structured workflows and applies phase retrieval, ring removal, and de-noising; supports reconstruction for parallel-beam, cone-beam, tomography, and laminography geometries; generates GPU-optimized 3D reconstruction code and executes across available GPUs.