Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PoCGM: Poisson-Conditioned Generative Model for Sparse-View CT Reconstruction

Changsheng Fang, Yongtong Liu, Bahareh Morovati, Shuo Han, Li Zhou, Hengyong Yu

TL;DR

This work introduces PoCGM, a Poisson-Conditioned Generative Model that reformulates the Poisson Flow Generative Model (PFGM++) as a conditional posterior sampler for sparse-view CT. By conditioning on sparse-view measurements, PoCGM models the posterior $p(x|x_{sparse})$, enabling efficient stepwise sampling along Poisson flow trajectories and substantially reducing streak artifacts while preserving fine details. The method extends PFGM++ with a conditional objective and uses a 16-step ODE sampling strategy, demonstrating superior performance over baselines on the Mayo Clinic AAPM low-dose CT dataset in terms of PSNR, SSIM, and LPIPS. This approach offers a practical, dose-conscious solution for time-critical CT reconstruction, integrating learned priors with observed measurements to improve image fidelity under extreme sparsity.

Abstract

In computed tomography (CT), reducing the number of projection views is an effective strategy to lower radiation exposure and/or improve temporal resolution. However, this often results in severe aliasing artifacts and loss of structural details in reconstructed images, posing significant challenges for clinical applications. Inspired by the success of the Poisson Flow Generative Model (PFGM++) in natural image generation, we propose a PoCGM (Poisson-Conditioned Generative Model) to address the challenges of sparse-view CT reconstruction. Since PFGM++ was originally designed for unconditional generation, it lacks direct applicability to medical imaging tasks that require integrating conditional inputs. To overcome this limitation, the PoCGM reformulates PFGM++ into a conditional generative framework by incorporating sparse-view data as guidance during both training and sampling phases. By modeling the posterior distribution of full-view reconstructions conditioned on sparse observations, PoCGM effectively suppresses artifacts while preserving fine structural details. Qualitative and quantitative evaluations demonstrate that PoCGM outperforms the baselines, achieving improved artifact suppression, enhanced detail preservation, and reliable performance in dose-sensitive and time-critical imaging scenarios.

PoCGM: Poisson-Conditioned Generative Model for Sparse-View CT Reconstruction

TL;DR

This work introduces PoCGM, a Poisson-Conditioned Generative Model that reformulates the Poisson Flow Generative Model (PFGM++) as a conditional posterior sampler for sparse-view CT. By conditioning on sparse-view measurements, PoCGM models the posterior , enabling efficient stepwise sampling along Poisson flow trajectories and substantially reducing streak artifacts while preserving fine details. The method extends PFGM++ with a conditional objective and uses a 16-step ODE sampling strategy, demonstrating superior performance over baselines on the Mayo Clinic AAPM low-dose CT dataset in terms of PSNR, SSIM, and LPIPS. This approach offers a practical, dose-conscious solution for time-critical CT reconstruction, integrating learned priors with observed measurements to improve image fidelity under extreme sparsity.

Abstract

In computed tomography (CT), reducing the number of projection views is an effective strategy to lower radiation exposure and/or improve temporal resolution. However, this often results in severe aliasing artifacts and loss of structural details in reconstructed images, posing significant challenges for clinical applications. Inspired by the success of the Poisson Flow Generative Model (PFGM++) in natural image generation, we propose a PoCGM (Poisson-Conditioned Generative Model) to address the challenges of sparse-view CT reconstruction. Since PFGM++ was originally designed for unconditional generation, it lacks direct applicability to medical imaging tasks that require integrating conditional inputs. To overcome this limitation, the PoCGM reformulates PFGM++ into a conditional generative framework by incorporating sparse-view data as guidance during both training and sampling phases. By modeling the posterior distribution of full-view reconstructions conditioned on sparse observations, PoCGM effectively suppresses artifacts while preserving fine structural details. Qualitative and quantitative evaluations demonstrate that PoCGM outperforms the baselines, achieving improved artifact suppression, enhanced detail preservation, and reliable performance in dose-sensitive and time-critical imaging scenarios.

Paper Structure

This paper contains 10 sections, 9 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Problem definition
  4. PFGM++ framework
  5. Posterior guidance Poisson flow
  6. Experiments and Results
  7. Dataset
  8. Experimental settings
  9. Results
  10. Discussion and Conclusion

Figures (3)

  • Figure 1: Data acquisition process of sparse-view CT
  • Figure 2: Illustration of 3D Poisson field trajectories for a 2D CT image distribution. The evolvements of a distribution or an augmented sample by the forward/backward ODEs pertains to the Poisson field.
  • Figure 3: Representative reconstruction results from 128 views using different methods. The $2^{nd}$ and $4^{th}$ row are the ROIs of the $1^{st}$ and $3^{rd}$ rows. The display window is [-160,240] HU.