Automated Vascular Segmentation in PET Imaging: From Technical Feasibility to Clinical Readiness, a Hands-On Scoping Review

Automated vascular segmentation could facilitate standardized quantitative PET/CT by reducing the burden of manual delineation, but clinical translation depends on robustness, quantitative validity, and workflow integration. We conducted a scoping review of the literature indexed in PubMed, Web of Science, and Scopus from 1 January 2000 to 1 September 2026 and identified 17 original studies spanning [18F]FDG, Na[18F]F, [68Ga]Ga-DOTATATE, [15O]H2O, [13N]NH3, and 82Rb PET applications. Methods ranged from PET-native clustering to CNN/U-Net, nnU-Net, SEQUOIA, TotalSegmentator, and MOOSE. The evidence was heterogeneous in population size, imaging protocol, reference standard, and degree of manual correction. The strongest direct evidence concerned aortic segmentation, including external validation in selected studies. Dynamic PET and multimodal studies showed that clinically relevant downstream measurements may remain valid despite imperfect geometric agreement, whereas small vessels and disease-altered anatomy remain challenging. To complement the review, we selected a segmentation tool, SEQUOIA, on the basis of its availability and the robustness of its published validation. We then performed an illustrative hands-on proof of SEQUOIA on two anonymized [18F]FDG PET/CT examinations as an internal test, in order to assess its readiness for use by clinical users in our institution, without validation purposes. Overall, automated vascular segmentation is technically mature for selected large-vessel applications, but current evidence favors expert-supervised, human-in-the-loop workflows over unrestricted autonomous use.