Deep learning for accelerated scatter correction in quantitative PET/MRI

To obtain quantitatively accurate images in positron emission tomography (PET), corrections have to be applied for gamma photons that were scattered due to interactions inside the patient or scanner hardware components. The aim of this study was to investigate the feasibility and potential benefit of using real patient data for training deep learning neural networks for scatter correction in PET. We have successfully developed a pipeline which facilitates the generation of a gold standard and the preprocessing of the data for the training of such deep learning neural networks. Our conclusion is that the use of real patient data for improving the training of deep learning neural networks for scatter estimation in PET is feasible and has the potential to improve the accuracy of reconstructed PET images in clinically feasible computation times.