Diffusion posterior sampling framework for joint image reconstruction and scatter estimation in cone‐beam CT
Medical Physics · 2026-11 · 53 (11)
DOI: 10.1002/mp.70755 ↗Abstrakt
Abstract Background X‐ray scatter can have a significant impact on image quality in cone beam CT (CBCT) including the introduction of severe artifacts and CT value inaccuracies. Classic solutions to the scatter problem often involve specialized hardware and data acquisition, or computationally expensive estimation of scatter via Monte Carlo (MC) methods. Purpose We explore a novel scatter correction approach that leverages the recent development of Diffusion Posterior Sampling (DPS) to jointly estimate CT attenuation and scattered radiation directly from single‐energy CT projection data. Methods We rewrite the nonlinear forward model for CT measurements to include an unknown additive scatter term which we jointly estimate with the image volume. Solution of this ill‐posed problem is enabled by diffusion models that capture the distribution of scatter‐free images of anatomy. The joint reconstruction and scatter estimation is evaluated using both simulation and phantom data and compared with a MC‐based scatter correction method. Results Qualitative and quantitative results obtained from simulated and phantom data shows that the proposed approach can effectively estimate the scatter contribution, significantly remove scatter artifacts, and restore the overall image contrast and attenuation values. Conclusions Joint estimation of scatter and reconstruction for CBCT has the potential to provide improved CBCT image quality and quantitation through direct estimation without the addition of specialized hardware or MC simulations.
EUVIMED