Skip to Main Content
Positron emission tomography (PET) measurements with time-of-flight (TOF) information are often very sparse. As a result, direct reconstruction from raw list-mode data is an attractive strategy for dealing with the large dimension spanned by the measurements. However, even though sparse datasets are more efficiently processed in list mode than as sinograms, list-mode reconstruction remains computationally demanding and computer clusters are typically required for reconstructing clinical PET scans with TOF information. In this work, we demonstrate that off-the-shelf graphics processing units can be used as an alternative approach to accelerate line projections with TOF kernels.