Fig. 1.
The deep learning model architecture used for estimating full-dose PET images from low-dose (LD) ones. The filtered image (FI) is inputted to the estimator network (a), which tries to estimate the true full-dose (EFD) by predicting the residual image (R) from the true full-dose image (FD). The network also tries to trick the discriminator network (b) which tries to determine the ground truth full-dose images from the estimated ones