Search papers, labs, and topics across Lattice.
This paper introduces the Shared-Structure 4D Spectral Gaussian Representation (4D-SG) for reconstructing energy-resolved attenuation volumes in sparse-view spectral computed tomography (CT). By leveraging a Gaussian-wise Spectral Density Curve Network (GSC-Net), the method effectively separates spatial structure from spectral attenuation variations, allowing for improved reconstruction from limited projection views. Experimental results show that 4D-SG outperforms existing Gaussian baselines, achieving significant improvements in PSNR, SSIM, and LPIPS metrics across various datasets.
Achieving a 1.05 dB increase in PSNR and enhanced structural similarity, 4D-SG redefines the capabilities of sparse-view spectral CT reconstruction.
Sparse-view spectral computed tomography (CT) reconstructs energy-resolved attenuation volumes from limited projection views, requiring simultaneous handling of angular undersampling and spectral coupling. We propose a SharedStructure 4D Spectral Gaussian Representation (4D-SG) that learns shared Gaussian geometry from full spectrum structural projections and uses a Gaussian-wise Spectral Density Curve Network (GSC-Net) to predict Gaussian raw density transformations. This factorization separates shared spatial structure from spectral attenuation variation, avoids independent channel geometry optimization, and establishes a continuous 4D-SG representation from discrete spectral measurements for unobserved spectral channel queries. Experiments on six synthesized, simulated projection, and real projection datasets with 50 views demonstrate the best average performance. Compared with the strongest Gaussian baseline, 4D-SG improves PSNR from 35.56 dB to 36.61 dB, increases SSIM from 0.909 to 0.914, and reduces LPIPS from 0.208 to 0.194, demonstrating its effectiveness for sparse-view spectral CT reconstruction.
