An exploration of the prognostic utility of shortened dynamic imaging protocols for PET-FDG scans

Standard whole-body clinical fluoro-deoxyglucose (FDG)-PET scans typically involve imaging for around 15 minutes about 60 minutes after tracer injection. The scan duration is often the critical constraint limiting patient through-put. Scans taken long after tracer injection restrict the ability to assess vascular and perfusion information that might be revealed by the early pattern of tracer uptake. On the other hand, early scanning may compromise the recovery of the late time uptake (SUV) which in many contexts has well established prognostic value. In this study, we explore the potential for short-duration dynamic scans, acquired immediately after tracer injection, to recover information that can predict late-stage uptake of FDG. The work involves re-analysis of existing series of dynamic brain and breast tumour imaging data to simulate the type of information that would arise from early and late scanning. Using a collection of machine learning techniques (including random forests, neural networks, gradient boosting), we find that short-duration clinical protocols, soon after the tracer injection, show significant potential to recover the late stage FDG flux information.