Answers are grounded in this paper and cite their sources.
Incomplete angular measurements in computed tomography (CT) cause severe directional ambiguity and streak artifacts, limiting clinical applicability. Existing prior-guided implicit neural representation methods typically use a single network to preserve prior information and reconstruct the target image, which may compromise reliable prior structures. We propose a prior-guided residual implicit neural representation framework for limited-angle CT reconstruction. The target image is modeled as the sum of a frozen prior representation and a trainable residual representation. The frozen prior network preserves stable anatomical structures learned from a previous scan, while the residual network recovers target-specific changes and compensates for artifacts using measured limited-angle projections. A periodically updated missing-view projection loss is further introduced to constrain the unmeasured angular range. The residual network is optimized using measured-view data fidelity, estimated missing-view consistency, and residual sparsity regularization. Under the 120 ° setting, it achieved the highest PSNR and SSIM on all three simulated slices and improved mean PSNR by approximately 1.29 dB over NeRP. This advantage remained as the acquisition range decreased from 150 ° to 60 ° . The experimental results demonstrate that the proposed method effectively suppresses limited-angle artifacts and preserves reliable details compared with competing algorithms.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: