GPDM: Generation-Prior Diffusion Model for Accelerated Direct Attenuation and Scatter Correction of Whole-body 18F-FDG PET

TL;DR

The paper addresses the need for accurate attenuation and scatter correction (ASC) in PET without relying on CT/MR-based priors. It introduces the Generation-Prior Diffusion Model (GPDM), a two-stage framework that first generates an Intermediate ASC PET from NASC-PET and then refines it with a diffusion model using a Generation-Prior to produce final ASC-PET. GPDM demonstrates superior quantitative performance (lower MAE and NMSE, higher PSNR) and reduced sampling steps (about 200 vs ~1000 in standard DDPM) compared with GAN-based and traditional diffusion baselines, while maintaining robustness across scanners and TOF resolutions. The approach enables CT-less ASC-PET suitable for standalone PET and has potential to streamline clinical workflows, though it remains sensitive to 3D spatial effects and requires further cross-scanner validation and larger datasets.

Abstract

Accurate attenuation and scatter corrections are crucial in positron emission tomography (PET) imaging for accurate visual interpretation and quantitative analysis. Traditional methods relying on computed tomography (CT) or magnetic resonance imaging (MRI) have limitations in accuracy, radiation exposure, and applicability. Deep neural networks provide potential approaches to estimating attenuation and scatter-corrected (ASC) PET from non-attenuation and non-scatter-corrected (NASC) PET images based on VAE or CycleGAN. However, the limitations inherent to conventional GAN-based methods, such as unstable training and mode collapse, need further advancements. To address these limitations and achieve more accurate attenuation and scatter corrections, we propose a novel framework for generating high-quality ASC PET images from NASC PET images: Generation-Prior Diffusion Model (GPDM). Our GPDM framework is based on the Denoising Diffusion Probabilistic Model (DDPM), but instead of starting sampling from an entirely different image distribution, it begins from a distribution similar to the target images we aim to generate. This similar distribution is referred to as the Generation-Prior. By leveraging this Generation-Prior, the GPDM framework effectively reduces the number of sampling steps and generates more refined ASC PET images. Our experimental results demonstrate that GPDM outperforms existing methods in generating ASC PET images, achieving superior accuracy while significantly reducing sampling time. These findings highlight the potential of GPDM to address the limitations of conventional methods and establish a new standard for efficient and accurate attenuation and scatter correction in PET imaging.

Figures (10)

• Figure 1: Overall Framework for Training and Sampling in GPDM
• Figure 2: Attenuation and scatter corrected PET images of two representative patients using (A) non-ASC and (B) CT-based ASC, (C) GPDM, and (C) their difference. The unit of the color bar is SUV.
• Figure 3: Cranial–caudal profiles of four representative patients, obtained by plotting the summed activity concentration across each axial (z-direction) slice for NASC PET, CT-ASC PET, and GPDM-ASC PET images.
• Figure 4: Comparison of ASC PET images generated through cGAN, Cycle GAN, cDDPM and GPDM for one patient. The error maps relative to CT-ASC PET are shown in the second row. The unit of the color bar is SUV.
• Figure 5: . Comparison of ASC PET images generated through cGAN, Cycle GAN, cDDPM, and GPDM for a different patient. The error maps relative to CT-ASC PET are shown in the second row. The unit of the color bar is SUV.