Deep unrolled primal dual network for TOF-PET list-mode image reconstruction

TL;DR

The results demonstrate the potential application of deep unrolled methods for TOF-PET list-mode data and show better performance than current mainstream TOF-PET list-mode reconstruction algorithms, providing new insights for the application of deep learning methods in TOF list-mode data.

Abstract

Time-of-flight (TOF) information provides more accurate location data for annihilation photons, thereby enhancing the quality of PET reconstruction images and reducing noise. List-mode reconstruction has a significant advantage in handling TOF information. However, current advanced TOF PET list-mode reconstruction algorithms still require improvements when dealing with low-count data. Deep learning algorithms have shown promising results in PET image reconstruction. Nevertheless, the incorporation of TOF information poses significant challenges related to the storage space required by deep learning methods, particularly for the advanced deep unrolled methods. In this study, we propose a deep unrolled primal dual network for TOF-PET list-mode reconstruction. The network is unrolled into multiple phases, with each phase comprising a dual network for list-mode domain updates and a primal network for image domain updates. We utilize CUDA for parallel acceleration and computation of the system matrix for TOF list-mode data, and we adopt a dynamic access strategy to mitigate memory consumption. Reconstructed images of different TOF resolutions and different count levels show that the proposed method outperforms the LM-OSEM, LM-EMTV, LM-SPDHG,LM-SPDHG-TV and FastPET method in both visually and quantitative analysis. These results demonstrate the potential application of deep unrolled methods for TOF-PET list-mode data and show better performance than current mainstream TOF-PET list-mode reconstruction algorithms, providing new insights for the application of deep learning methods in TOF list-mode data. The codes for this work are available at https://github.com/RickHH/LMPDnet

Figures (11)

• Figure 1: The reconstruction scheme of the proposed LMPDnet. The reconstruction process is unrolled with $n$ phases, each phase contains one dual module for dual variable of list-mode domain updating and one primal module for primal variable of image domain updating. The teal arrows indicate the path of the list-mode data, and the lime green arrows indicate the path of the image domain data.
• Figure 2: The ground truth images and the reconstructed images with 3e5 counts and TOF resolution of 200 ps by different reconstruction methods. From left to right: Truth, LM-OSEM, LM-EM-TV, LM-SPDHG, LM-SPDHG-TV, FastPET-2D and proposed LMPDnet.
• Figure 3: The illustration of the calculation of TOF PET list-mode projection matrix.
• Figure 4: The CRC-STD and Bias-STD trade-off curves between contrast and noise in the simulation study for different methods. (a) CRC vs. STD curve; (b)Bias vs. STD curve. For OSEM and EM-TV, markers were plotted every 3 iterations. For SPDHG and SPDHG-TV, markers were plotted every 1 iterations. For FastPET-2D, markers were plotted every 400 epochs and for proposed LMPDnet, markers were plotted every 2 phases.
• Figure 5: The High count clinical images reconstructed by EM and the reconstructed images with 3e5 counts and TOF resolution of 200 ps by different reconstruction methods. From left to right: High count EM, OSEM, EM-TV, SPDHG, SPDHG-TV, FastPET-2D and proposed LMPDnet.