Generation of Whole-Body FDG Parametric <i>K</i> _i Images From Static PET Images Using Deep Learning

F-fluorodeoxyglucose parametric $K_{\mathrm{ i}}$ images show a great advantage over static standard uptake value (SUV) images, due to the higher contrast and better accuracy in tracer rate estimation. In this study, we explored feasibility of generating synthetic from SUV ratio (SUVR) using three configurations U-Nets with different sets input output image patches, which were single (SISO), multiple inputs (MISO), outputs (SIMO). SUVR generated by averaging 5-min dynamic frames starting at 60-min post-injection, then normalized mean values blood pool. The corresponding ground-truth derived Patlak graphical analysis functions measurement arterial samples. Even though not quantitatively accurate compared ground truth, linear regression joint histograms voxels body regions showed that notation="LaTeX">$R^{2}$ between U-Net prediction truth (0.596, 0.580, 0.576 SISO, MISO, SIMO), than (0.571). terms similarity metrics, closer (mean SSIM = 0.729, 0.704, 0.704 MISO) 0.691). Therefore, it is feasible use deep learning networks estimate surrogate map images.