[Paper Review] Continuity-driven Synergistic Diffusion with Neural Priors for Ultra-Sparse-View CBCT Reconstruction

Junlin Wang, Jiancheng Fang+3 more|arXiv (Cornell University)|Feb 8, 2026

Medical Imaging Techniques and Applications0 citations

TL;DR

Introduces a continuity-driven diffusion framework with neural priors (CSDN) to reconstruct high-quality CBCT from ultra-sparse angular views by synergistically refining sinograms and inter-slice consistency, followed by adaptive fusion for coherent 3D volume reconstruction.

ABSTRACT

The clinical application of cone-beam computed tomography (CBCT) is constrained by the inherent trade-off between radiation exposure and image quality. Ultra-sparse angular sampling, employed to reduce dose, introduces severe undersampling artifacts and inter-slice inconsistencies, compromising diagnostic reliability. Existing reconstruction methods often struggle to balance angular continuity with spatial detail fidelity. To address these challenges, we propose a Continuity-driven Synergistic Diffusion with Neural priors (CSDN) for ultra-sparse-view CBCT reconstruction. Neural priors are introduced as a structural foundation to encode a continuous threedimensional attenuation representation, enabling the synthesis of physically consistent dense projections from ultra-sparse measurements. Building upon this neural-prior-based initialization, a synergistic diffusion strategy is developed, consisting of two collaborative refinement paths: a Sinogram Refinement Diffusion (Sino-RD) process that restores angular continuity and a Digital Radiography Refinement Diffusion (DR-RD) process that enforces inter-slice consistency from the projection image perspective. The outputs of the two diffusion paths are adaptively fused by the Dual-Projection Reconstruction Fusion (DPRF) module to achieve coherent volumetric reconstruction. Extensive experiments demonstrate that the proposed CSDN effectively suppresses artifacts and recovers fine textures under ultra-sparse-view conditions, outperforming existing state-of-the-art techniques.

Motivation & Objective

• Motivate ultra-sparse-view CBCT reconstruction with reduced radiation dose while maintaining diagnostic image quality.
• Incorporate neural priors to encode a continuous 3D attenuation representation for dense projection synthesis from ultra-sparse data.
• Develop a dual diffusion refinement framework that enforces angular continuity and inter-slice consistency.

Proposed method

• Use neural priors to establish a continuous 3D attenuation representation for synthesizing dense projections from ultra-sparse measurements.
• Develop Sinogram Refinement Diffusion (Sino-RD) to restore angular continuity in projection data.
• Develop Digital Radiography Refinement Diffusion (DR-RD) to enforce inter-slice consistency from projection images.
• Introduce Dual-Projection Reconstruction Fusion (DPRF) to adaptively fuse outputs from Sino-RD and DR-RD for coherent volumetric reconstruction.
• Base the approach on a diffusion-based refinement pipeline that iteratively improves the reconstructed volume under ultra-sparse-view conditions.

Experimental results

Research questions

• RQ1 Can neural priors provide a robust continuous 3D attenuation representation to synthesize dense projections from ultra-sparse CBCT measurements?
• RQ2 Can a dual diffusion refinement strategy jointly restore angular continuity and inter-slice consistency to improve ultra-sparse-view CBCT reconstructions?
• RQ3 Does adaptive fusion of Sino-RD and DR-RD outputs yield more coherent 3D reconstructions than single-path refinements?
• RQ4 How does the proposed method perform relative to state-of-the-art ultra-sparse-view CBCT reconstruction techniques in suppressing artifacts and preserving textures?

Key findings

• The proposed CSDN method effectively suppresses artifacts under ultra-sparse-view CBCT conditions.
• Neural priors enable synthesis of physically consistent dense projections from sparse measurements.
• Two collaboration diffusion paths (Sino-RD and DR-RD) restore angular continuity and inter-slice consistency, respectively.
• Adaptive fusion via the DPRF module produces coherent volumetric reconstructions by combining the two refinement paths.