LRIP-Net: Low-Resolution Image Prior-Based Network for Limited-Angle CT Reconstruction

Abstract

In the practical applications of computed tomography (CT) imaging, the projection data may be acquired within a limited-angle range and corrupted by noises due to the limitation of scanning conditions. The noisy incomplete projection data results in the ill-posedness of the inverse problems. Based on the observation that the low-resolution reconstruction problem has better numerical stability, we propose a novel low-resolution image prior-based CT reconstruction model for limited-angle reconstruction. More specifically, we build up a low-resolution reconstruction problem on the down-sampled projection data, and use the reconstructed low-resolution image as prior knowledge for the high-resolution limited-angle CT problem. The constrained minimization problem is then solved by the alternating direction method with all subminimization problems approximated by the convolutional neural networks. Numerical experiments demonstrate that our double-resolution network outperforms both the variational method and popular learning-based reconstruction methods on noisy limited-angle reconstruction problems.

Low-resolution prior equilibrium network for CT reconstruction

“…Although our downsampling imaging geometric modeling was studied for the fan-beam imaging systems, it can be also used for other modern CT settings. Similar works include [29] and [17], where the low-resolution image prior was incorporated into the reconstruction network. The prior information in [29] was obtained by halving the receiver signal, requiring a second scan of the patient, which is clinically prohibitive.…”

“…In contrast, our approach eliminates the need of re-scanning patients. On the other hand, Gao et al [17] also requires two distinct sets of projection data. Obviously, the low-resolution image obtained from the same scan can better maintain consistency with the high-resolution image, thereby preserving more structural and fine information.…”

“…The projection data obtained from both resolutions were combined to reconstruct an image with different levels of discretization, effectively incorporating information from both measurements to enhance the reconstructed image quality. Gao et al [17] proposed a low-resolution image prior image reconstruction model, which was realized in an end-to-end regime showing impressive ability in dealing with limited-angle reconstruction problems. He et al [29] introduced a downsampling imaging geometric modeling approach for the data acquisition process, which integrated the geometric modeling knowledge of the CT imaging system with prior knowledge obtained from a data-driven training process to achieve precise CT image reconstruction.…”

“…The loss function is the mean square error of the image domain and the Radon domain with the weight α being 1. • LRIP: the low-resolution image prior based Network in [17]. It is a low-resolution image prior image reconstruction model for the limited-angle reconstruction problems.…”

Prior frequency guided diffusion model for limited angle (LA)-CBCT reconstruction

“…These hybrid methods effectively leverage the power of neural networks in feeding prior information and serving as denoisers, and the proven records of physics-driven iterative reconstructions in solving model-based inverse problems. Different neural networks, including GAN, have been used as imaging priors to regularize model-based reconstructions, including LA-CBCT reconstruction (Yang et al 2020 , Barutcu et al 2021 , Zhou et al 2021 , Gao et al 2023 ). However, such methods may still fail, especially for very small scan angles, as the priors provided by these neural networks are insufficient or unstable.…”