LegoPET: Hierarchical Feature Guided Conditional Diffusion for PET Image Reconstruction

TL;DR

LegoPET is introduced, a hierarchicaL feature guided conditional diffusion model for high-perceptual quality PET image reconstruction from sinograms that improves the performance of cDPMs but also surpasses recent DL-based PET image reconstruction techniques in terms of visual quality and pixel-level PSNR/SSIM metrics.

Abstract

Positron emission tomography (PET) is widely utilized for cancer detection due to its ability to visualize functional and biological processes in vivo. PET images are usually reconstructed from histogrammed raw data (sinograms) using traditional iterative techniques (e.g., OSEM, MLEM). Recently, deep learning (DL) methods have shown promise by directly mapping raw sinogram data to PET images. However, DL approaches that are regression-based or GAN-based often produce overly smoothed images or introduce various artifacts respectively. Image-conditioned diffusion probabilistic models (cDPMs) are another class of likelihood-based DL techniques capable of generating highly realistic and controllable images. While cDPMs have notable strengths, they still face challenges such as maintain correspondence and consistency between input and output images when they are from different domains (e.g., sinogram vs. image domain) as well as slow convergence rates. To address these limitations, we introduce LegoPET, a hierarchical feature guided conditional diffusion model for high-perceptual quality PET image reconstruction from sinograms. We conducted several experiments demonstrating that LegoPET not only improves the performance of cDPMs but also surpasses recent DL-based PET image reconstruction techniques in terms of visual quality and pixel-level PSNR/SSIM metrics. Our code is available at https://github.com/yransun/LegoPET.

Figures (3)

• Figure 1: Overview of LegoPET. LegoPET includes two parts: a learned PnPNet for hierarchical feature maps extraction (right), and a standard 2D sinogram-conditioned DPM for PET image reconstruction (left). Two lists of structured layered feature maps, $b_d$ and $b_m$, from the pre-trained PnPNet are added to the cDPM as biases.
• Figure 2: Comparison of LegoPET with Four Baselines on Two Example Reconstructed Slices. The first column shows the input sinogram images, and the second column shows the reference images reconstructed using OSEM algorithm. The third to sixth columns correspond to the four baselines (labeled above each image), and the final column shows the reconstructed PET image using proposed LegoPET method. PSNR/SSIM values are reported below each slice, and squared error maps between each method and the reference image are also displayed (second and fourth rows). We prove that LegoPET generates PET images with the highest perceptual quality, and also improves the performance of cDPM through feature guidance.
• Figure 3: Effectiveness of Hierarchical Feature Guidance. We compare the performance of LegoPET and cDPM within 500 epochs in terms of PSNR and SSIM. We regard cDPM as "LegoPET w/o guidance", and LegoPET as "cDPM w/ guidance".