Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diffusion-based Sinogram Interpolation for Limited Angle PET

Rüveyda Yilmaz, Julian Thull, Johannes Stegmaier, Volkmar Schulz

TL;DR

Limited-angle PET acquisitions create severely undersampled sinograms, leading to artifacts in conventional reconstructions. The authors introduce conditional diffusion models with ControlNet-style conditioning, using a Stable Diffusion backbone to interpolate missing sinogram bins guided by the observed LA data. They report a PSNR improvement from 47.93 dB to 55.08 dB and observe artifact-free, sinusoidal-consistent predictions in 2D data, indicating strong data priors and physically plausible completions. This approach enables more flexible, cost-effective PET geometries with potential extensions to full 3D acquisitions and broader generalization tests.

Abstract

Accurate PET imaging increasingly requires methods that support unconstrained detector layouts from walk-through designs to long-axial rings where gaps and open sides lead to severely undersampled sinograms. Instead of constraining the hardware to form complete cylinders, we propose treating the missing lines-of-responses as a learnable prior. Data-driven approaches, particularly generative models, offer a promising pathway to recover this missing information. In this work, we explore the use of conditional diffusion models to interpolate sparsely sampled sinograms, paving the way for novel, cost-efficient, and patient-friendly PET geometries in real clinical settings.

Diffusion-based Sinogram Interpolation for Limited Angle PET

TL;DR

Limited-angle PET acquisitions create severely undersampled sinograms, leading to artifacts in conventional reconstructions. The authors introduce conditional diffusion models with ControlNet-style conditioning, using a Stable Diffusion backbone to interpolate missing sinogram bins guided by the observed LA data. They report a PSNR improvement from 47.93 dB to 55.08 dB and observe artifact-free, sinusoidal-consistent predictions in 2D data, indicating strong data priors and physically plausible completions. This approach enables more flexible, cost-effective PET geometries with potential extensions to full 3D acquisitions and broader generalization tests.

Abstract

Accurate PET imaging increasingly requires methods that support unconstrained detector layouts from walk-through designs to long-axial rings where gaps and open sides lead to severely undersampled sinograms. Instead of constraining the hardware to form complete cylinders, we propose treating the missing lines-of-responses as a learnable prior. Data-driven approaches, particularly generative models, offer a promising pathway to recover this missing information. In this work, we explore the use of conditional diffusion models to interpolate sparsely sampled sinograms, paving the way for novel, cost-efficient, and patient-friendly PET geometries in real clinical settings.

Paper Structure

This paper contains 6 sections, 1 equation, 2 figures.

Table of Contents

  1. Introduction
  2. Materials & Methods
  3. Background and Proposed Approach
  4. Experimental Setup
  5. Results & Discussion
  6. Conclusion

Figures (2)

  • Figure 1: Sample LA, predicted, and GT sinograms (a), corresponding MLEM reconstructions (b), and the difference map between the GT and predicted sinograms normalized to the range [0,1] (c).
  • Figure 2: Sample absolute error (AE) maps, normalized to the range [0,1], are shown for reconstructions using MLEM from a direct LA sinogram and from our proposed method.