Skip to main content
. 2022 Mar 21;49(9):3098–3118. doi: 10.1007/s00259-022-05746-4

Fig. 2.

Fig. 2

Description of deep learning-based PET image reconstruction methods. A End-to-end methods are fully data-driven and do not require an instrument-based system matrix. B Regularisation-based methods utilise a neural network in combination with data consistency, retaining the system matrix. C Unrolling iterative algorithms into a sequence of processing steps makes it feasible to train iteration-specific regularisation terms