Dynamic PET

Standardized uptake value ratio (SUVR) derived from a static tau scan (~20 min) is currently the most used semi-quantitative metric of tau binding. One drawback of SUVR is its sensitivity to the measurement time window and changes in blood flow. Compared to SUVR, distribution volume ratio (DVR) obtained from dynamic PET is more quantitative. Furthermore, recent studies showed that the delivery rate R1 derived from a dynamic tau scan can act as a proxy of cerebral blood flow and glucose metabolism for MCI/AD characterization. Though promising, wider adoption of dynamic tau imaging is impeded by the high complexity of current protocols (e.g., 2-hour 18F-MK-6240 scan at MGH). Additionally, supervised deep learning is difficult to be applied to dynamic PET imaging as the scan time is already long. We are working on medical physics -informed unsupervised deep learning to improve dynamic PET image quality. Below is one network structure designed which combines the U-net and the kinetic model process into one network structure.

Related Publications:

  1. Gong K, Cheng-Liao J, Wang G, Chen KT, Catana C, Qi J. Direct Patlak reconstruction from dynamic PET data using kernel method with MRI information based on structural similarity. IEEE transactions on medical imaging. 2018 Apr 1;37(4):955-65.
  2. Xie N*, Gong K*, Guo N, Qin Z, Wu Z, Liu H, Li Q. Rapid High-Quality PET Patlak Parametric Image Generation based on Direct Reconstruction and Temporal Nonlocal Neural Network, NeuroImage. 2021 Oct 15;240:118380.
  3. Gong K, Catana C, Qi J, Li Q. Direct Reconstruction of Linear Parametric Images from Dynamic PET Using Nonlocal Deep Image Prior, IEEE Transactions on Medical Imaging, 2021 Oct 15;41(3):680-9.
  4. Cui J*, Gong K*, N Guo, K Kim, Liu H, Li Q. Unsupervised PET Logan Parametric Image Estimation Using Conditional Deep Image Prior.  Medical Image Analysis, 2022 Aug 1;80:102519.