MCR Toolkit

MCR Toolkit reconstructs multi-channel x-ray computed tomography (CT) data using GPU-accelerated analytical and iterative algorithms to enable spectral and time-resolved imaging for preclinical and clinical studies.

Key Features:

• Multi-Energy and Dynamic Imaging: Supports reconstruction of multi-energy and time-resolved x-ray CT data enabling low-dose coronary angiography and multi-contrast imaging.
• Implementation and Acceleration: Implemented in C/C++ with NVIDIA CUDA GPU programming interface and provides scripting support via MATLAB and Python.
• Analytical Reconstruction: Includes filtered backprojection (FBP) for circular cone-beam CT (CBCT), weighted FBP (WFBP) for helical CBCT, and cone-parallel projection rebinning followed by WFBP for 3rd generation cylindrical multi-detector row CT (MDCT).
• Iterative Multi-Channel Reconstruction: Supports arbitrary combinations of energy and temporal channels under a generalized multi-channel signal model solved algebraically.
• Optimization and Solvers: Utilizes the split Bregman optimization method and the BiCGSTAB(l) linear solver.
• Regularization Techniques: Applies rank-sparse kernel regression (RSKR) for energy-dimension regularization and patch-based singular value thresholding (pSVT) for time-dimension regularization to denoise and support post-reconstruction material decomposition.
• Automatic Parameter Estimation: Regularization parameters are automatically estimated from input data under a Gaussian noise model.
• Performance and Scalability: Benchmarking on NVIDIA RTX 8000 GPU hardware demonstrates 61%–99% efficiency when scaling computation across one to four GPUs.

Scientific Applications:

• Photon-Counting Spectral Imaging: Applicable to photon counting spectral CT experiments and analysis.
• Dynamic Cardiac and Perfusion Imaging: Enables time-resolved imaging for cardiac motion and perfusion studies.
• Low-Dose Coronary Angiography and Multi-Contrast Imaging: Supports reconstruction workflows for low-dose angiography and multi-contrast protocols.
• Preclinical and Clinical Translation: Facilitates translation of reconstruction methods between preclinical research and clinical scenarios.
• Phantom and In Vivo Validation: Demonstrated with the digital MOBY mouse phantom (cardiac motion), in vivo mouse model of atherosclerosis, XCAT phantom, and DukeSim CT simulator.

Methodology:

Analytical reconstruction via FBP for circular CBCT, WFBP for helical CBCT, and cone-parallel projection rebinning followed by WFBP for 3rd generation cylindrical MDCT; iterative reconstruction using a generalized multi-channel signal model solved with split Bregman optimization and the BiCGSTAB(l) linear solver; energy regularization by rank-sparse kernel regression (RSKR) and temporal regularization by patch-based singular value thresholding (pSVT); automatic estimation of regularization parameters from input data under a Gaussian noise model.