Abstract

Background/Objectives: Differences among scanners and reconstruction methods may limit the comparability of quantitative metrics derived from fluorodeoxyglucose positron emission tomography (FDG PET)/computed tomography (CT). Although harmonization reduces inter-scanner variability in standardized uptake values (SUVs), its impact on the preservation of lesion-level ranking in real-world clinical datasets remains unclear. Here, we evaluated the robustness of PET quantitative metrics, particularly focusing on rank preservation after harmonization. Methods: Phantom and clinical data retrospectively acquired from three institutions using four PET/CT scanner types were analyzed. Harmonization parameters were derived from National Electrical Manufacturers Association IEC Body Phantom data, using the oldest scanner as the reference, and were directly applied to the clinical datasets. Clinical evaluation included head and neck malignant melanoma (HNMM; high FDG avidity) and adenoid cystic carcinoma (ACC; low FDG avidity). Rank preservation between pre- and post-harmonization values was assessed using Spearman’s rank correlation coefficient (ρ). Results: Phantom-based harmonization reduced inter-scanner differences and enabled consistent evaluation in clinical datasets (HNMM: 34 patients, 93 lesions; ACC: 18 patients, 38 lesions). SUVpeak demonstrated the highest rank preservation across tumor types and lesion sizes (ρ = 0.94–1.00). Metabolic tumor volume (MTV) showed a high rank correlation in HNMM with an absolute threshold (MTV2.5; ρ = 0.99), but robustness varied depending on threshold definition, tumor type, and lesion size. Tumor-to-liver ratios showed moderate rank preservation. Conclusions: Our results suggested that SUVpeak is the most robust preservation of lesion ranking across tumor types after harmonization, suggesting its suitability as a reliable imaging biomarker in multicenter studies. Meanwhile, careful standardization is needed when using MTV-based metrics for prognostic evaluation.

IPC Classification

Keywords