Efficient Diffusion Posterior Sampling for Dose Reduced CT Reconstruction
Keywords: CT reconstruction, Diffusion model, Posterior sampling
Abstract: The clinical efficacy of Computed Tomography (CT) is well-established, yet concerns regarding its radiation exposure persist. To mitigate this risk, a reduction in X-ray photon count or projection views is typically pursued, albeit at the expense of image quality. In this study, we introduce an innovative diffusion posterior sampling approach for CT image reconstruction at reduced radiation doses. This method initiates with a predictive step, leveraging data enhancement on the posterior approximation derived from a pre-trained diffusion model and the measurement data. Subsequently, a forward sampling phase ensues, which maps the output to a noisy timestep, followed by a diffusion estimation process. Additionally, we propose an acceleration strategy that employs superior initialization to significantly curtail the sampling steps required. Our experimental findings indicate that this method not only enhances the quality of reconstructed images by an average of 3.5 db but also accelerates the process to over ten times faster than existing diffusion-based techniques. These outcomes underscore the method's potential in clinical settings.
Primary Area: generative models
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 13495
Loading