Adaptive Diffusion Models for Sparse-View Motion-Corrected Head Cone-beam CT

TL;DR

This work tackles motion artifacts in sparse-view head CBCT by introducing JRM-ADM, a framework that jointly reconstructs the 3D attenuation volume and estimates rigid motion while incorporating an adaptive diffusion prior. By performing diffusion posterior sampling in the wavelet domain and coupling it with a cubic-B-spline motion model, the method delivers data-consistent reconstructions that preserve fine anatomical details under highly undersampled acquisitions. The approach demonstrates consistent PSNR and SSIM gains over traditional and learning-based baselines, and shows improved fracture visibility and artifact suppression in both fracture-free and fractured head scenarios. While computationally intensive, the method offers a principled route to motion-robust, low-dose CBCT with potential clinical impact, contingent on validation with real projection data and speed-ups via differentiable rendering and scalable architectures.

Abstract

Cone-beam computed tomography (CBCT) is an imaging modality widely used in head and neck diagnostics due to its accessibility and lower radiation dose. However, its relatively long acquisition times make it susceptible to patient motion, especially under sparse-view settings used to reduce dose, which can result in severe image artifacts. In this work, we propose a novel framework combining joint reconstruction and motion estimation (JRM) with an adaptive diffusion model (ADM) that simultaneously addresses motion compensation and sparse-view reconstruction in head CBCT. Leveraging recent advances in diffusion-based generative models, our method integrates a wavelet-domain diffusion prior into an iterative reconstruction pipeline to guide the solution toward anatomically plausible volumes while estimating rigid motion parameters in a blind fashion. We evaluate our method on simulated motion-affected CBCT data derived from real clinical computed tomography (CT) volumes. Experimental results demonstrate that JRM- ADM achieves consistent quantitative improvements over both traditional and learning-based baselines. In highly undersampled cases, JRM-ADM improves peak signal-to-noise ratio (PSNR) by more than 4 dB and structural similarity index measure (SSIM) by 0.10 compared to the baseline motion-corrected (MC) reconstruction method. These results highlight the potential of our approach to enable motion-robust, low-dose CBCT imaging, paving the way for improved clinical viability. The project page is available at https://antoinedepaepe.github.io/jrm-adm-io/.

Figures (10)

• Figure 1: Reconstructions under sparse-view setting, patient motion setting, and a combination of both.
• Figure 2: Overview of the proposed JRM-ADM approach.
• Figure 3: GT and reconstructed volumes (axial, coronal and sagittal planes) in the 60-view CBCT setting for one patient.
• Figure 4: GT and reconstructed volumes (axial, coronal and sagittal planes) in the 20-view CBCT setting for one patient.
• Figure 5: GT and reconstructed volumes with non-MC DPS approach for different values of $\gamma$, in the 60-view setting. When $\gamma$ is low, the algorithm tends to ignore the prior and instead focuses on the data which are affected by motion, thus resulting in motion artifacts. On the contrary, when $\gamma$ is high, the algorithm tends to favor the prior, which produces a motion-free image that deviates from the GT.