Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fast Voxel-Wise Kinetic Modeling in Dynamic PET using a Physics-Informed CycleGAN

Christian Salomonsen, Samuel Kuttner, Michael Kampffmeyer, Robert Jenssen, Kristoffer Wickstrøm, Jong Chul Ye, Elisabeth Wetzer

TL;DR

This work tackles the challenge of fast, voxel-wise kinetic imaging in dynamic PET by removing dependence on invasive arterial input function estimation. It introduces a physics-informed CycleGAN that maps dPET time series to both kinetic parameter maps and the arterial input function using a forward $2TCM$ relationship, trained on unpaired data. On a dataset of 70 mouse [18F]FDG dPET scans, the method achieves competitive AIF estimation compared with state-of-the-art DL-based approaches and yields parametric maps with substantial fidelity to a reference pipeline (SSIM $0.7425$, PSNR $34.49$ dB). The approach enables noninvasive, voxel-wise kinetic imaging with potential clinical impact, though normalization effects on absolute rate scales and the need for uncertainty quantification must be addressed for translation.

Abstract

Tracer kinetic modeling serves a vital role in diagnosis, treatment planning, tracer development and oncology, but burdens practitioners with complex and invasive arterial input function estimation (AIF). We adopt a physics-informed CycleGAN showing promise in DCE-MRI quantification to dynamic PET quantification. Our experiments demonstrate sound AIF predictions and parameter maps closely resembling the reference.

Fast Voxel-Wise Kinetic Modeling in Dynamic PET using a Physics-Informed CycleGAN

TL;DR

This work tackles the challenge of fast, voxel-wise kinetic imaging in dynamic PET by removing dependence on invasive arterial input function estimation. It introduces a physics-informed CycleGAN that maps dPET time series to both kinetic parameter maps and the arterial input function using a forward relationship, trained on unpaired data. On a dataset of 70 mouse [18F]FDG dPET scans, the method achieves competitive AIF estimation compared with state-of-the-art DL-based approaches and yields parametric maps with substantial fidelity to a reference pipeline (SSIM , PSNR dB). The approach enables noninvasive, voxel-wise kinetic imaging with potential clinical impact, though normalization effects on absolute rate scales and the need for uncertainty quantification must be addressed for translation.

Abstract

Tracer kinetic modeling serves a vital role in diagnosis, treatment planning, tracer development and oncology, but burdens practitioners with complex and invasive arterial input function estimation (AIF). We adopt a physics-informed CycleGAN showing promise in DCE-MRI quantification to dynamic PET quantification. Our experiments demonstrate sound AIF predictions and parameter maps closely resembling the reference.
Paper Structure (7 sections, 1 equation, 1 figure)

This paper contains 7 sections, 1 equation, 1 figure.

Table of Contents

  1. Introduction
  2. Methods
  3. Data
  4. Implementation Details
  5. Experiments and Results
  6. Discussion and Conclusion
  7. Potential Negative Societal Impact

Figures (1)

  • Figure 1: Comparison of (a) the ground truth and estimated kinetic parameter maps for $K_1$ and $K_i$, and (b) the estimated AIF from FC-DLIF salomonsen2025DLIF, PI-DLIF salomonsen2025physicsinformed and the proposed CycleGAN approach. In (a), the first axis contains the measured AIF values from blood sampling, while the second axis contains the estimated AIF values from the three methods. The dashed line reports the identity line, i.e., perfect estimation.