Beyond a Universal Cut-Off: [¹⁸F]FET-PET Metrics for Distinguishing High-Grade Glioma Recurrence from...

Background/Objectives: Differentiating tumour recurrence from treatment-related changes (TRC) after multimodal treatment of high-grade gliomas (HGGs) remains challenging because of the overlap in imaging findings on conventional magnetic resonance. This study compared the diagnostic performance of the main static semiquantitative O-(2-[¹⁸F]fluoroethyl)-L-tyrosine positron emission tomography ([¹⁸F]FET-PET) parameters and assessed the diagnostic contribution of qualitative tracer uptake. Methods: This retrospective single-centre study included 77 scans from 61 adult patients with previously resected and chemoradiotherapy-treated HGGs undergoing [¹⁸F]FET-PET for suspected recurrence. Standardised uptake values (SUVs), tumour-to-background ratios (TBRs), and metabolic tumour volume (MTV) were analysed. Recurrence was determined using a composite histopathological and clinical-radiological reference standard. Diagnostic performance was assessed using receiver operating characteristic analysis, Youden-derived cut-offs and DeLong comparisons. Internal validation was performed by using patient-grouped five-fold cross-validation, in which cut-offs were re-derived within each training partition and applied to the held-out partition. Results: Forty-nine scans were classified as recurrence and 28 as TRC. All semiquantitative parameters except MTV were significantly higher in recurrent disease after false-discovery-rate correction. TBR mean ratio showed the highest accuracy (AUC, 0.80; cut-off, 1.85; sensitivity, 77.6%; specificity, 75.0%), followed by SUV max (AUC, 0.793). Neither prespecified comparison among TBR parameters was significant. The presence of pathological uptake achieved 100% sensitivity, 42.9% specificity, 100% NPV, and 79.2% accuracy. MTV was non-discriminative (AUC, 0.499). SUV max differed significantly between scanners (p = 0.008). Conclusions: Static semiquantitative [¹⁸F]FET-PET parameters, except MTV, showed comparable moderate-to-good performance for distinguishing recurrence from TRC. TBRmean may represent a pragmatic primary marker due to its background normalisation reduction of scanner-related variability. Guideline-recommended thresholds should be regarded as reference values requiring local validation and, when appropriate, centre- and protocol-specific calibration.