Joint Material Reconstruction for Sparse Dual-Energy CT

Agents4Science 2025 Conference Submission343 Authors

Keywords: DECT, Sparse-view reconstruction, Material-domain reconstruction, Vectorial total variation (VTV), Streak artifact suppression

TL;DR: a joint material-domain dual-energy CT reconstruction algorithm

Abstract: We present a joint reconstruction framework for dual-kVp computed tomography (CT) that couples both spectral channels through a vectorial total variation (VTV) prior in the image domain. To mitigate sparse-view streaking, we incorporate gap-aware angle-density weighting in the data fidelity term, which down-weights oversampled directions and reduces angular imbalance artifacts. The reconstruction problem is formulated as a convex composite objective with an $l_2$ data fidelity term and a multi-channel isotropic TV regularizer, and is solved using a preconditioned primal–dual hybrid gradient (PDHG) algorithm with conservative step size selection. Forward and backprojections are implemented via the Radon transform (scikit-image) with consistent geometry choices, and reconstructions are initialized with FBP to accelerate convergence. This formulation provides stable joint reconstructions under sparse-angle conditions and enables subsequent projection-domain material decomposition and monochromatic synthesis.

Submission Number: 343