Diffusion Models Reduce Truncation Artifacts in Cone-Beam CT Research
Published in Artificial Intelligence, Journals, Scanners and Imaging, Tools.
A study published in BMC Medical Imaging on September 29, 2026 investigates conditional diffusion models for reducing truncation artifacts in cone-beam CT. The paper received media coverage on October 4.
Jing Xie and colleagues use generative models to complete missing portions of projection data before reconstructing images. Compared with RTK padding, the study reports average reconstructed-volume improvements of 9 dB in peak signal-to-noise ratio and 0.10 in structural similarity, together with better image uniformity and visibility of low-contrast structures.
These measurements assess reconstruction fidelity; they are not direct measures of better diagnosis or patient benefit. The study was retrospective, and the publisher identifies the article as a peer-reviewed early version subject to further edits.
The approach warrants further testing across scanners, anatomy and patient populations. Because missing information is generated, clinical evaluation must examine whether important structures are preserved reliably, including in cases outside the model’s training experience.
Source: Xie et al., BMC Medical Imaging, September 29, 2026. DOI: 10.1186/s12880-026-02818-x.