PADM: A Physics-aware Diffusion Model for Attenuation Correction

TL;DR

This work tackles attenuation artifacts in cardiac SPECT MPI by introducing PADM, a CT-free attenuation correction diffusion model that embeds physics priors through a teacher–student distillation scheme. A CT-informed teacher guides a NAC-only student to reproduce high-fidelity attenuation-corrected reconstructions, enabling accurate AC outputs without CT input at inference. The authors also contribute CardiAC, a 424-study dataset with paired NAC/AC reconstructions and CT-based attenuation maps to benchmark CT-free attenuation correction. Across quantitative metrics and visual assessments, PADM outperforms existing generative baselines, demonstrating robust performance and clinical relevance for widespread, low-radiation attenuation correction in cardiac SPECT.

Abstract

Attenuation artifacts remain a significant challenge in cardiac Myocardial Perfusion Imaging (MPI) using Single-Photon Emission Computed Tomography (SPECT), often compromising diagnostic accuracy and reducing clinical interpretability. While hybrid SPECT/CT systems mitigate these artifacts through CT-derived attenuation maps, their high cost, limited accessibility, and added radiation exposure hinder widespread clinical adoption. In this study, we propose a novel CT-free solution to attenuation correction in cardiac SPECT. Specifically, we introduce Physics-aware Attenuation Correction Diffusion Model (PADM), a diffusion-based generative method that incorporates explicit physics priors via a teacher--student distillation mechanism. This approach enables attenuation artifact correction using only Non-Attenuation-Corrected (NAC) input, while still benefiting from physics-informed supervision during training. To support this work, we also introduce CardiAC, a comprehensive dataset comprising 424 patient studies with paired NAC and Attenuation-Corrected (AC) reconstructions, alongside high-resolution CT-based attenuation maps. Extensive experiments demonstrate that PADM outperforms state-of-the-art generative models, delivering superior reconstruction fidelity across both quantitative metrics and visual assessment.

Figures (3)

• Figure 1: Overview of the proposed PADM method. (a) Teacher–student framework: The teacher network (PADM-T) conditions the diffusion process on the Attenuation map via a cross-modality transformer, while the student network (PADM-S) learns from NAC images and is guided by feature distillation. (b) Physics-aware reconstruction: At each step of the diffusion process, the U-Net predicts projections $s_\phi$, mean $\mu$, and clean image $\tilde{x}_0$, which are refined through a physics-aware iterative update to produce final output.
• Figure 2: Qualitative comparison of reconstructed images across three standard views: horizontal long axis (top), short axis (middle), and vertical long axis (bottom).
• Figure 3: Comparison of PADM against the baseline diffusion models, i.e., BBDM and BBDM without VQGAN. PADM-T denotes the teacher model, and PADM-S denotes the student model. $\downarrow$ indicates lower is better, $\uparrow$ indicates higher is better. Diff. (%) is computed as the relative difference from the baseline score.