POUR-Net: A Population-Prior-Aided Over-Under-Representation Network for Low-Count PET Attenuation Map Generation

TL;DR

POUR-Net tackles the problem of radiation dose in PET by generating high-quality attenuation maps from low-count PET without CT. It combines an over-under-representation network (OUR-Net) with a population-prior generation machine (PPGM) in a cascade to iteratively refine $μ$-maps, using a large CT-derived dataset as population priors. The approach yields superior $μ$-map quality and improved PET attenuation correction under ultra-low-dose settings, surpassing existing DL baselines and enabling CT-free low-count PET. While computationally intensive, the method demonstrates a clear path toward reducing radiation exposure in clinical PET imaging with robust performance across low-count scenarios.

Abstract

Low-dose PET offers a valuable means of minimizing radiation exposure in PET imaging. However, the prevalent practice of employing additional CT scans for generating attenuation maps (u-map) for PET attenuation correction significantly elevates radiation doses. To address this concern and further mitigate radiation exposure in low-dose PET exams, we propose POUR-Net - an innovative population-prior-aided over-under-representation network that aims for high-quality attenuation map generation from low-dose PET. First, POUR-Net incorporates an over-under-representation network (OUR-Net) to facilitate efficient feature extraction, encompassing both low-resolution abstracted and fine-detail features, for assisting deep generation on the full-resolution level. Second, complementing OUR-Net, a population prior generation machine (PPGM) utilizing a comprehensive CT-derived u-map dataset, provides additional prior information to aid OUR-Net generation. The integration of OUR-Net and PPGM within a cascade framework enables iterative refinement of $μ$-map generation, resulting in the production of high-quality $μ$-maps. Experimental results underscore the effectiveness of POUR-Net, showing it as a promising solution for accurate CT-free low-count PET attenuation correction, which also surpasses the performance of previous baseline methods.

Figures (6)

• Figure 1: The overall pipeline of our population prior-aided over-under-representation network (POUR-Net) for attenuation map ($\mu$-map) generation. POUR-Net integrates the over-under-representation network (OUR-Net) and Population-Prior Generation Machine (PPGM) for iterative refinement of $\mu$-map generation (left part). The detailed steps of PPGM are illustrated on the right.
• Figure 2: The detailed network architecture of the over-under-representation network (OUR-Net) used in the cascade framework of POUR-Net (Figure \ref{['fig:framework']}). The network consists of an over-represented network (OvNet) branch at the top and an under-represented network (UnNet) branch at the bottom, for assisting the generation in the central full-resolution network (FuNet) branch.
• Figure 3: Visual comparison of $\mu$-map generation from different methods under 10% and 2.5% low-count PET settings. The coronal view and selected axial views (red and blue cuts) are shown. RMSE and PSNR values are calculated for each individual volume, with the CT-derived $\mu$ map as a reference (last column).
• Figure 4: Visual comparison of full-count PET reconstructions using $\mu$-map generated from low-count PET data. The non-attenuation-corrected PET reconstructions (bottom left) suffer from severe quantification errors, as compared to PET reconstructions with AC using CT-derived $\mu$-map (top left). Compared to prior $\mu$-map generation methods (central two columns), POUR-Net (last column) shows much better PET quantification accuracy.
• Figure 5: Examples of input, intermediate output, and final outputs of the proposed PPGM. Using the OUR-Net prediction (bottom right), PPGM searches for the most match $\mu$-map (bottom left) and performs registration (bottom middle).