RecoItTV

RecoItTV reconstructs limited-angle cone-beam computed tomography (CBCT) volumes using a GPU-accelerated Split Bregman iterative method to reduce artifacts from insufficient projections and enable rapid large-volume reconstruction.

Key Features:

• Split Bregman iterative solver: Implements an accelerated iterative reconstruction based on the Split Bregman formulation to incorporate prior information and mitigate artifacts from limited-angle data.
• GPU-accelerated kernels: Executes reconstruction kernels on GPUs to substantially reduce computation time and memory consumption compared with CPU-only approaches.
• Large-volume support: Supports reconstruction of volumetric datasets exceeding 1024^3 pixels via partitioning strategies.
• Partitioned forward/back-projection: Employs partitioning during forward- and back-projection operations to handle very large projections and volumes.
• Parallelization and memory management: Utilizes parallel processing techniques and efficient memory management strategies to lower runtime and memory footprint.
• Computational profiling and scaling: Notes that forward- and back-projection account for ~60% of computation time and that reconstruction time scales linearly with number of projections and quadratically with projection size while being relatively insensitive to angular span.
• Performance improvement: Demonstrated up to 48× speedup over traditional CPU-only methods on evaluated datasets.

Scientific Applications:

• Limited-angle CBCT reconstruction: Reconstruction of CBCT data acquired with restricted rotation (typically <180°), such as scans during surgical or other constrained-angle acquisitions.
• Large-volume high-resolution imaging: Reconstruction of very large volumetric datasets (>1024^3 pixels) for detailed anatomical and preclinical studies.
• Preclinical small-animal imaging: Applied and evaluated on small-animal projection data to accelerate iterative reconstructions.
• Accelerated iterative reconstruction for clinical and research workflows: Enables deployment of advanced iterative methods where rapid reconstruction is required.

Methodology:

Implements an accelerated Split Bregman iterative reconstruction using GPU-accelerated kernels, partitioned forward- and back-projection, parallel processing, and memory-management optimizations.