Quantification in respiratory-gated PET acquisition: can data-driven methods replace device-based systems?—a comparative and retrospective study

Background: Device-based respiratory gating improves diagnostic and quantification accuracy in positron emission tomography (PET), but requires additional time to setup the device and the failure rate can be significant. Our aim was to internally validate the quantification performance of data-driven respiratory-gated PET imaging against the gold standard, the device-based method, in clinical oncological practice. We retrospectively analysed [18F]FDG PET/CT scans of patients from our centre with at least one measurable [18F]FDG-avid malignant lesion. All PET/CT acquisitions were performed on a Siemens Biograph 64 Vision 600 system with respiratory gating by belt and also by adding the data-driven gating with OncoFreeze AI™. We recorded the SUVmax and SUVpeak for up to a maximum of 5 lesions per patient. We computed the mean absolute bias between the two gating methods and the 95% confidence intervals (CI) at the cohort level and in subgroups. Results: Of the 692 consecutive patients screened for inclusion, 196 patients were analysed, from whom 536 lesions were measured. The mean absolute biases in the SUVmax and SUVpeak of lesions in the whole cohort were 3.8% (CI 3.4–4.2) and 2.1% (CI 1.9–2.4), respectively. At patient-level, 21% of them had at least one lesion with a SUVmax bias above 10%, while for SUVpeak this proportion was 5%. In the subgroup analysis by PERCIST criteria, only 2% of patients had significant bias in the SUVmax, and 0.5% in SUVpeak. There was no clinically significant effect of lesion size or anatomical site on SUV measurements between the two respiratory gating methods. Conclusion: Quantitative comparison of data-driven and device-based respiratory-gated PET scans revealed negligible differences, proving that data-driven respiratory gating is a reliable and accurate alternative to the device-based gating method in routine [18F]FDG-PET/CT oncological evaluation.