Reconstruction matters: influence of EARL standards and BSREM in [18F]FET PET for distinguishing glioma progression from treatment-related changes Authors: J. de Jong, M. Segbers, A. A. Harteveld, W. J. Bron, R. Valkema, R. Boellaard, S. E. M. Veldhuijzen van Zanten Published: 2026 ABSTRACT BackgroundDifferentiating progressive disease (PD) from treatment-related changes (TRC) in glioma remains challenging using magnetic resonance imaging (MRI) alone. Amino acid positron-emission tomography (PET) with O-(2-[18F]fluoroethyl)-L-tyrosine ([¹⁸F]FET) is increasingly applied in clinical practice, and the first guideline for PET-based Response Assessment in Neuro-Oncology (PET-RANO) has recently been developed. However, quantitative PET metrics are sensitive to image reconstruction, which may directly influence response classification and clinical interpretation. This study aimed to investigate how different reconstruction standards affect quantitative [¹⁸F]FET PET metrics and to what extent these differences impact clinical interpretation.ResultsThirty-three patients with 42 lesions were analysed. In lesion‑based comparisons, both EANM Research Ltd. harmonization standard version 2.0 (EARL2) and block-sequential regularized expectation maximization using a β-value of 150 (BSREM150) yielded significantly higher standardized uptake values (SUV), tumour‑to‑background ratios (TBR), and biological tumour volumes (BTV) than EANM Research Ltd. harmonization standard version 1.0 (EARL1), and identified a greater number of lesions as measurable disease (MD). BTV differed across reconstructions; adapting the BSREM150 segmentation threshold to SUV ≥ 1.7 x mean background activity improved BTV concordance while preserving superior MD detection. Excellent discrimination between PD and TRC was observed for TBRmax across all reconstructions (area under the curve ≥ 0.94), although optimal thresholds were reconstruction‑specific (EARL1: 1.67; EARL2: 1.84; BSREM150: 2.39). At the patient level, EARL1 failed to identify three PD cases, whereas BSREM150 misclassified one TRC case as PD.ConclusionReconstruction standards substantially influence quantitative [¹⁸F]FET PET metrics and clinical interpretation. Applying universal thresholds across reconstructions risks misclassification, with EARL1 potentially underestimating tumour extent and BSREM150 increasing false-positive findings. Reconstruction-specific thresholds and neuro-oncology-specific harmonization frameworks are therefore essential for reliable [18F]FET PET assessment in routine clinical practice and PET-RANO implementation in multicentre trials.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13550-026-01458-0. Background Gliomas are the most common primary brain tumours [1] and remain incurable despite multimodal treatment [2–4]. Following surgery and chemoradiotherapy, new or enlarging contrast‑enhancing and/or T2‑weighted fluid‑attenuated inversion recovery (FLAIR) hyperintense magnetic resonance imaging (MRI) lesions frequently emerge, complicating the distinction between progressive disease (PD) and treatment-related changes (TRC), such as pseudoprogression. Accurate differentiation is critical to avoid premature therapy changes or unnecessary surgical interventions, yet structural MRI alone frequently fails to provide clarity [5]. Dynamic Susceptibility Contrast perfusion MRI, utilizing relative cerebral blood volume, can improve diagnostic accuracy by detecting neovascularity as an indirect marker of malignancy [6], although its interpretation is limited by heterogeneous thresholds and reduced accuracy near large vessels or cortical structures [7]. Amino-acid positron emission tomography (PET) using O-(2-[¹⁸F]fluoroethyl)-L-tyrosine ([18F]FET) offers complementary metabolic information and has become integral for resolving equivocal MRI findings [8, 9]. Its quantitative metrics, such as standardized uptake value (SUV), tumour-to-background ratio (TBR), and biological tumour volume (BTV), are central to image interpretation and form the basis of the recent PET-based Response Assessment in Neuro-Oncology (PET-RANO) criteria [8]. However, these metrics depend on image reconstruction standards [10, 11]. Although harmonization frameworks, such as EANM Research Ltd harmonization standard version 1.0 (EARL1) and version 2.0 (EARL2) [12, 13], have improved inter-centre consistency, newer reconstruction algorithms such as block-sequential regularized expectation maximization (BSREM) [14] offer higher resolution but challenge the transferability of thresholds derived under earlier standards. Because PET-RANO relies on quantitative thresholds such as TBR, understanding reconstruction-driven variability is essential to prevent misclassification in both clinical trials and patient care. Recognizing how reconstruction standards shape quantitative metrics is therefore essential for [18F]FET PET interpretation. This study evaluates how EARL1, EARL2, and BSREM150 reconstruction standards influence quantitative [¹⁸F]FET PET metrics and clinical interpretation in suspected glioma progression. We hypothesize that optimal TBRmax thresholds are reconstruction-specific, and that applying universal thresholds increases the risk of misclassification between PD and TRC. Methods Study design and patients This retrospective single-centre study was approved by the Erasmus Medical Centre Institutional Review Board and conducted in accordance with the Declaration of Helsinki. All patients provided written informed consent for the use of their imaging and clinical data for research purposes. Eligible patients met the following criteria: (i) histopathological diagnosis of glioma, (ii) age ≥ 18 years, (iii) prior radiotherapy, (iv) equivocal MRI findings after treatment, and (v) referral for hybrid [18F]FET PET-MRI to evaluate suspected PD. Patients scanned for other indications (e.g., treatment planning or response monitoring) were excluded. Image acquisition Hybrid PET-MRI was performed on a 3.0T PET-MR scanner (SIGNA; GE Healthcare, Chicago, IL). After fasting ≥ 4 h, ~ 200 MBq [18F]FET (median; range: 183–242) was administered intravenously, followed by a 40-minute list-mode acquisition. Static PET images corresponded to the 20–40 min post-injection interval. [18F]FET was commercially acquired from both BV Cyclotron VU (Amsterdam, the Netherlands) and Theranostics Erasmus MC BV (Rotterdam, the Netherlands). MRI was acquired simultaneously with PET in the setting of diagnostic uncertainty. During further patient follow-up, MRI was performed as a stand-alone examination on a 1.5 or 3.0T MRI scanner (GE Healthcare, Chicago, IL). MRI attenuation correction included a Dixon-based 4-tissue (air, lung, water, fat) sequence and a zero-echo-time sequence for bone inclusion. MRI acquisition was performed in accordance with the European Organisation for Research and Treatment of Cancer (EORTC) brain tumour imaging protocol [15]. Reconstruction algorithms All PET reconstructions were generated from the same raw list-mode acquisition to ensure direct comparability. Data were reconstructed according to EARL1 and EARL2 standards, and with a BSREM (Q.Clear; GE Healthcare, Chicago, IL) algorithm. All images were reconstructed into a 256 × 256 matrix with a 30-cm dual field of view. For both EARL-compliant reconstructions, a 3D ordered-subset expectation-maximization algorithm incorporating time-of-flight (TOF) was used. Point-spread function (PSF) was applied for EARL2, but not in EARL1. EARL1 reconstructions employed 3 iterations and 8 subsets with a Gaussian filter of 10.0 mm full width at half maximum (FWHM); EARL2 reconstructions used 4 iterations, 28 subsets, and a 7.0 mm Gaussian filter. BSREM reconstructions inherently incorporate both TOF and PSF and were generated using a β-value of 150 (BSREM150). All reconstructions were corrected for attenuation, scatter, and random coincidences. Quantitative analysis Image analysis followed international consensus guidelines [8]. Quantitative analysis was performed using Hermes Hybrid 3D viewer (Hermes Medical Solutions AB, Stockholm, Sweden) and in-house Python software based on the SimpleITK framework [16–18]. Background activity was measured using a 25 mm crescent-shaped volume of interest contralateral to the tumour, avoiding regions of physiologic uptake [19]. Lesions were semi-automatically segmented using a segmentation threshold of SUV ≥ 1.6 × mean background activity, with manual refinement to exclude physiological high-uptake regions. The resulting segmented volume was defined as the BTV. PET-positive disease was defined as any lesion with a SUV of ≥ 1.6 x the mean background activity. Measurable disease (MD) was defined as lesions with a BTV > 0.5 mL [8]. Extracted PET metrics included lesional SUVmean, SUVmax, SUVpeak, TBRmax, TBRmean, TBRpeak, BTV, and background SUVmean. In the lesion-based analysis, lesions were classified as PD, TRC, or borderline using a TBRmax threshold of > 2.3, < 1.9, and 1.9–2.3, respectively, consistent with current international consensus guidelines [20]. In the patient-based analysis, patients were classified as PD if they had at least one PET-positive lesion with a TBRmax > 2.3; as TRC if they had no PET-positive lesions or PET-positive lesions with a TBRmax < 1.9; and as borderline if they had a PET-positive lesion with TBRmax between 1.9 and 2.3. To investigate whether segmentation and the resulting BTV concordance between EARL1, EARL2, and BSREM150 could be improved, we performed an exploratory analysis in which EARL2 and BSREM150 were re-segmented using segmentation thresholds ranging from SUV ≥ 1.7 to ≥ 2.0 × mean background activity. Segmentation concordance was quantified using Dice coefficients and Jaccard indices. We further explored higher thresholds based on the hypothesis that higher-resolution reconstructions may require more stringent SUV thresholds to achieve spatial agreement with EARL1. In all comparisons, EARL1 segmentations obtained using a segmentation threshold of SUV ≥ 1.6 × mean background activity served as the reference standard, as this value aligns most closely with current guideline recommendations [20]. Reference standard The reference standard was based on histopathologic confirmation and/or clinico-radiological follow-up. PD was defined by (i) histopathological evidence of PD obtained via biopsy or (re-)resection, or (ii) multidisciplinary tumour board (MTB) consensus of PD supported by serial MRI (minimum follow-up ≥ 6 months after [¹⁸F]FET PET-MRI), or (iii) cancer-associated death or a decision to initiate best supportive care. Statistical analysis Lesion-based differences in quantitative PET metrics across reconstructions were assessed using the Friedman test with Wilcoxon signed-rank post hoc comparisons. Proportions of lesions classified as MD or FET-positive were compared using Cochran’s Q test with McNemar post hoc analysis. For lesion classification (PD/borderline/TRC) Friedman testing with Wilcoxon post-hoc analysis was applied. Agreement between reconstructions was evaluated using Cohen’s unweighted kappa (κ). Diagnostic performance was determined by receiver operating characteristic (ROC) analysis, and optimal cut-off values were identified by Youden’s index. Segmentation concordance between reconstructions with consecutive SUV-based segmentation thresholds was assessed using Dice similarity coefficients (DICE) and Jaccard scores. All statistical analyses were performed in IBM SPSS Statistics version 27.0 (IBM Corp., Armonk, NY). Statistical significance was defined as < 0.05, with multiple comparisons adjusted using the Dunn-Bonferroni post hoc correction. Results Patient characteristics Between 2022 and 2025, 33 patients (16 oligodendroglioma, 15 astrocytoma, and 2 glioblastoma) were included. Baseline characteristics are summarized in Table 1, and full patient characteristics can be found in Supplemental Table 1. All patients had previously undergone radiotherapy, with most also receiving chemotherapy (94%, 31/33). Twenty-eight patients presented with 42 measurable lesions, occurring at a median of 34 months (range: 4–187) after radiotherapy. Eight patients had multiple measurable lesions. Five patients showed no measurable lesions on [18F]FET and were therefore not included in the lesion-based analysis. Table 1Baseline demographics and disease characteristics of the study cohortParameterN = 33Sex Male (%)21 (64) Female (%)12 (36)Age at diagnosis (years) Median, range 50 (27-74)Glioma subtype Oligodendroglioma (%)16 (48) Astrocytoma (%)15 (46) Glioblastoma (%)2 (6)WHO grade Grade 2 (%)15 (46) Grade 3 (%)11 (33) Grade 4 (%)7 (21)Previous treatment Resection + radiotherapy (%)33 (100) Resection + radiotherapy + chemotherapy (%)31 (94)Time between PET and last radiotherapy fraction (months) Median, range34 (4–187) Table 1Baseline demographics and disease characteristics of the study cohortParameterN = 33Sex Male (%)21 (64) Female (%)12 (36)Age at diagnosis (years) Median, range 50 (27-74)Glioma subtype Oligodendroglioma (%)16 (48) Astrocytoma (%)15 (46) Glioblastoma (%)2 (6)WHO grade Grade 2 (%)15 (46) Grade 3 (%)11 (33) Grade 4 (%)7 (21)Previous treatment Resection + radiotherapy (%)33 (100) Resection + radiotherapy + chemotherapy (%)31 (94)Time between PET and last radiotherapy fraction (months) Median, range34 (4–187) Baseline demographics and disease characteristics of the study cohort Lesion-based analysis Results of the comparison between quantitative PET metrics across EARL, EARL2, and BSREM150 can be found in Table 2. All quantitative PET metrics increased progressively from EARL1 to EARL2 to BSREM150 (p < 0.001 for all). Median BTV was lowest with EARL1 (5.99 mL) and highest with BSREM150 (10.89 mL). Significant differences in BTV were observed between EARL1 and both higher-resolution reconstructions, while no significant difference was found between EARL2 and BSREM150. Table 2Quantitative [¹⁸F]FET PET parameters across reconstruction standardsSUV meanEARL1EARL2BSREM150p-value1.90 (1.21–3.53)a, b2.06 (1.40–3.75)a, c2.41 (2.15–10.45)b, c< 0.001SUVmax2.58 (1.21–7.17)a, b3.16 (1.62–8.27)a, c4.55 (2.15–10.45)b, c< 0.001SUVpeak2.18 (1.22–6.39)a, b2.56 (1.21–7.30)a, c2.99 (1.34–7.67)b, c< 0.001TBRmean1.92 (1.61–2.97)a, b2.03 (1.57–3.35)a, c2.25 (1.75–3.62)b, c< 0.001TBRmax2.56 (1.62–6.89)a, b3.17 (1.77–7.95)a, c3.80 (2.09–8.04)b, c< 0.001TBRpeak2.15 (1.36–6.14)a, b2,39 (1.56–7.02)a, c2.60 (1.76–6.73)b, c< 0.001BTV (mL)5.99 (0.01–220.25)a, b7.05 (0.39–205.24)a10.89 (0.64–195.86)b< 0.001SUVmean BG0.99 (0.70–1.45)a, b1.02 (0.69–1.50)a, c1.14 (0.70–1.52)b, c< 0.001a, b, c Indicates significant difference at p < 0.05 between pairs with the same superscript letter (Dunn-Bonferroni correctedpost-hoc test after Friedman test) Table 2Quantitative [¹⁸F]FET PET parameters across reconstruction standardsSUV meanEARL1EARL2BSREM150p-value1.90 (1.21–3.53)a, b2.06 (1.40–3.75)a, c2.41 (2.15–10.45)b, c< 0.001SUVmax2.58 (1.21–7.17)a, b3.16 (1.62–8.27)a, c4.55 (2.15–10.45)b, c< 0.001SUVpeak2.18 (1.22–6.39)a, b2.56 (1.21–7.30)a, c2.99 (1.34–7.67)b, c< 0.001TBRmean1.92 (1.61–2.97)a, b2.03 (1.57–3.35)a, c2.25 (1.75–3.62)b, c< 0.001TBRmax2.56 (1.62–6.89)a, b3.17 (1.77–7.95)a, c3.80 (2.09–8.04)b, c< 0.001TBRpeak2.15 (1.36–6.14)a, b2,39 (1.56–7.02)a, c2.60 (1.76–6.73)b, c< 0.001BTV (mL)5.99 (0.01–220.25)a, b7.05 (0.39–205.24)a10.89 (0.64–195.86)b< 0.001SUVmean BG0.99 (0.70–1.45)a, b1.02 (0.69–1.50)a, c1.14 (0.70–1.52)b, c< 0.001a, b, c Indicates significant difference at p < 0.05 between pairs with the same superscript letter (Dunn-Bonferroni correctedpost-hoc test after Friedman test) Quantitative [¹⁸F]FET PET parameters across reconstruction standards a, b, c Indicates significant difference at p < 0.05 between pairs with the same superscript letter (Dunn-Bonferroni correctedpost-hoc test after Friedman test) Both EARL2 and BSREM150 identified significantly more lesions as MD than EARL1 (40 and 42 vs. 32; p < 0.001; Supplemental Table 2). BSREM150 also detected significantly more PET-positive lesions compared with EARL1, whereas no significant differences were observed between EARL1 vs. EARL2 or EARL2 vs. BSREM150. In addition, BSREM150 classified significantly more lesions as PD than both EARL1 and EARL2 (41 vs. 22 and 30; p < 0.001). The mean BTV in lesions missed as PD by EARL1 was 0.05 mL (range 0–2.40). Patient-based analysis For PET-based classification at the patient level, EARL1 classified 21/33 patients (64%) as PD, 8/33 (24%) as TRC and, 4/33 (12%) as borderline; EARL2 classified 24/33 (73%) as PD, 5/33 (15%) as TRC and, 4/33 as borderline (12%); BSREM150 classified 29/33 (88%) as PD and 4/33 (12%) as TRC (Table 3). κ indicated moderate inter-method agreement between EARL1 vs. BSREM150 (κ = 0.41, SE 0.15) and EARL2 vs. BSREM150 (0.56, SE 0.16), and substantial agreement between EARL1 vs. EARL2 (κ = 0.68, SE 0.12). Table 3Patient-level classification of progressive disease (PD) versus treatment-related changes (TRC)Cross-tabulation comparing the diagnostic performance of EARL1 and EARL2EARL2EARL1TotalPDTRCBorderlinePD211224 (73%)TRC0505 (15%)Borderline0224 (12%)Total21 (64%)8 (24%)4 (12%)33Cross-tabulation comparing the diagnostic performance of EARL1 and BSREM150BSREM150EARL1TotalPDTRCBorderlinePD214429 (88%)TRC0404 (12%)Borderline0000Total21 (64%)8 (24%)4 (12%)33Cross-tabulation comparing the diagnostic performance of EARL2 and BSREM150BSREM150EARL2TotalPDTRCBorderlinePD241429 (88%)TRC0404 (12%)Borderline0000Total24 (73%)5 (15%)4 (12%)33 Table 3Patient-level classification of progressive disease (PD) versus treatment-related changes (TRC)Cross-tabulation comparing the diagnostic performance of EARL1 and EARL2EARL2EARL1TotalPDTRCBorderlinePD211224 (73%)TRC0505 (15%)Borderline0224 (12%)Total21 (64%)8 (24%)4 (12%)33Cross-tabulation comparing the diagnostic performance of EARL1 and BSREM150BSREM150EARL1TotalPDTRCBorderlinePD214429 (88%)TRC0404 (12%)Borderline0000Total21 (64%)8 (24%)4 (12%)33Cross-tabulation comparing the diagnostic performance of EARL2 and BSREM150BSREM150EARL2TotalPDTRCBorderlinePD241429 (88%)TRC0404 (12%)Borderline0000Total24 (73%)5 (15%)4 (12%)33 Patient-level classification of progressive disease (PD) versus treatment-related changes (TRC) After a median follow-up of 19 months (range: 4-41) post PET-MRI, 28 patients were classified as PD and five as TRC. PD was confirmed through re-resection in seven, MTB consensus supported by 2–6 follow-up MRI examinations in nine (median follow-up 23 months, range: 8–35), cancer-associated death in eleven (median follow-up 10 months, range: 4–25) and the initiation of best supportive care in one. In one case (Supplemental Table 1, #32), a patient with a recurrent glioblastoma 4 months after resection followed by chemoradiation showed elevated [18F]FET uptake (EARL1 TBRmax = 2.96; EARL2 TBRmax = 3.34; BSREM150 TBRmax = 4.21), and histopathological analysis demonstrated a mixture of PD and features of radionecrosis. When PET-MRI findings were compared with the reference standard (Supplemental Table 3), EARL1 missed three PD cases (Fig. 1), whereas BSREM misclassified one TRC case as PD (Fig. 2). Of the six cases classified as borderline by both EARL1 and EARL2, five were deemed PD by the MTB and were started on lomustine. The remaining patient was managed with a watch-and-wait approach and demonstrated progression of [18F]FET uptake and contrast-enhancing abnormalities on PET-MRI after 5 months, with further progression of contrast-enhancing abnormalities on MRI after 9 months. Fig. 1Representative false-negative [18F]FET PET-MRI in a patient with isocitrate dehydrogenase-mutant grade 4 astrocytoma. (a) EARL1 shows no measurable lesion (TBRmax = 1.59), (b) EARL2 identifies a measurable lesion that is classified as borderline findings (BTV = 3.07 mL, TBRmax = 2.07), (c) BSREM150 further enhances delineation and classifies the lesion as PD (BTV = 6.09 mL, TBRmax = 2.60); (d-f) simultaneously acquired MRI (contrasted enhanced T1-weighted, T2-weighted, FLAIR) demonstrates partly linear and partly nodular contrast-enhancement at the border of the resection cavity within a T2‑weighted FLAIR hyperintense region of the left frontal lobe. It was decided by the MTB to treat the lesion as PD with lomustine. Follow-up MRI imaging after 6 months showed further tumour growth under lomustine. Representative false-negative [18F]FET PET-MRI in a patient with isocitrate dehydrogenase-mutant grade 4 astrocytoma. (a) EARL1 shows no measurable lesion (TBRmax = 1.59), (b) EARL2 identifies a measurable lesion that is classified as borderline findings (BTV = 3.07 mL, TBRmax = 2.07), (c) BSREM150 further enhances delineation and classifies the lesion as PD (BTV = 6.09 mL, TBRmax = 2.60); (d-f) simultaneously acquired MRI (contrasted enhanced T1-weighted, T2-weighted, FLAIR) demonstrates partly linear and partly nodular contrast-enhancement at the border of the resection cavity within a T2‑weighted FLAIR hyperintense region of the left frontal lobe. It was decided by the MTB to treat the lesion as PD with lomustine. Follow-up MRI imaging after 6 months showed further tumour growth under lomustine. Fig. 2Representative false-positive [18F]FET PET-MRI in a patient with isocitrate dehydrogenase-mutant grade 4 astrocytoma. (a) EARL1 shows no measurable lesion (TBRmax = 1.54), (b) EARL2 depicts a lesion classified as TRC (BTV = 0.54 mL, TBRmax = 1.80), (c) BSREM further delineates a lesion suggestive of PD (BTV = 1.42, TBRmax = 2.32), (d-f) simultaneously acquired MRI (contrast-enhanced T1, T2, FLAIR) shows rim enhancement within the area of T2‑weighted FLAIR hyperintensity, (g-i) Follow-up MRI after 4 months reveals a marked reduction in enhancement and a mild decrease in or T2‑weighted FLAIR signal, consistent with TRC rather than PD. Representative false-positive [18F]FET PET-MRI in a patient with isocitrate dehydrogenase-mutant grade 4 astrocytoma. (a) EARL1 shows no measurable lesion (TBRmax = 1.54), (b) EARL2 depicts a lesion classified as TRC (BTV = 0.54 mL, TBRmax = 1.80), (c) BSREM further delineates a lesion suggestive of PD (BTV = 1.42, TBRmax = 2.32), (d-f) simultaneously acquired MRI (contrast-enhanced T1, T2, FLAIR) shows rim enhancement within the area of T2‑weighted FLAIR hyperintensity, (g-i) Follow-up MRI after 4 months reveals a marked reduction in enhancement and a mild decrease in or T2‑weighted FLAIR signal, consistent with TRC rather than PD. Diagnostic performance Lesion-based ROC analysis of SUVpeak, TBRmax, and TBRpeak for differentiating PD from TRC yielded excellent diagnostic performance across all reconstruction algorithms (Table 4; Fig. 3). SUVmax resulted in good diagnostic performance for BSREM150 (area under the curve (AUC) 0.83) and excellent diagnostic performance for EARL1 (AUC 0.97) and EARL2 (AUC 0.93). TBRmax achieved AUC ≥ 0.94 for all reconstructions, with optimal thresholds differing per reconstruction: EARL1 = 1.67 (AUC 0.96; sensitivity 85%, specificity 100%), EARL2 = 1.84 (AUC 0.94; sensitivity 95%, specificity 83%), and BSREM150 = 2.39 (AUC 0.98; sensitivity 95%, specificity 100%). Table 4Diagnostic performance of SUVmax, SUVpeak and TBRmax for distinguishing progressive disease (PD) from treatment-related changes (TRC)Cut-offSensitivity (%)Specificity (%)AUCp-valueEARL1SUVmax1.65901000.97< 0.001SUVpeak1.53881000.97< 0.001TBRmax1.67851000.96< 0.001TBRpeak1.45931000.96< 0.001EARL2SUVmax1.5798830.93< 0.001SUVpeak1.66931000.96< 0.001TBRmax1.8495830.94< 0.001TBRpeak1.73831000.96< 0.001BSREM150SUVmax2.6688830.83< 0.001SUVpeak1.7695830.91< 0.001TBRmax2.39951000.98< 0.001TBRpeak2.09761000.93< 0.001 Table 4Diagnostic performance of SUVmax, SUVpeak and TBRmax for distinguishing progressive disease (PD) from treatment-related changes (TRC)Cut-offSensitivity (%)Specificity (%)AUCp-valueEARL1SUVmax1.65901000.97< 0.001SUVpeak1.53881000.97< 0.001TBRmax1.67851000.96< 0.001TBRpeak1.45931000.96< 0.001EARL2SUVmax1.5798830.93< 0.001SUVpeak1.66931000.96< 0.001TBRmax1.8495830.94< 0.001TBRpeak1.73831000.96< 0.001BSREM150SUVmax2.6688830.83< 0.001SUVpeak1.7695830.91< 0.001TBRmax2.39951000.98< 0.001TBRpeak2.09761000.93< 0.001 Diagnostic performance of SUVmax, SUVpeak and TBRmax for distinguishing progressive disease (PD) from treatment-related changes (TRC) Fig. 3Receiver operating charecteristics (ROC) curves of SUVmax, SUVpeak, TBRmax, and TBRpeak derived from EARL1, EARL2, and BSREM150 reconstructions in distinguishing progressive disease (PD) from treatment-related changes (TRC) Receiver operating charecteristics (ROC) curves of SUVmax, SUVpeak, TBRmax, and TBRpeak derived from EARL1, EARL2, and BSREM150 reconstructions in distinguishing progressive disease (PD) from treatment-related changes (TRC) In a subset of 30 lesions in 21 consecutive patients, the highest segmentation concordance was observed between EARL1 and EARL2 at a SUV threshold of 1.6 x mean background activity (DICE = 0.869; Jaccard = 0.769), and between EARL1 and BSREM150 at 1.7 (DICE = 0.765; Jaccard = 0.619) (Supplemental Table 4). In this subset, the median BTV for BSREM150 at a harmonized threshold of SUV ≥ 1.7 x mean background activity was no longer significantly different from EARL1 derived volumes. Notably, lesions classified as non-measurable with EARL1 were still identified as MD using BSREM150 at this adapted threshold. Figure 4 illustrates the variation in BTV segmentations across reconstructions. Fig. 4Comparative visualization of a brain lesion using T1-weighted gadolinium-enhanced MRI (T1gd) and [18F]FET PET imaging reconstructed with EARL1, EARL2, and BSREM150. The T1gd image (left) highlights the anatomical location of the lesion, which is further delineated in PET images (middle) using SUV thresholds of > 1.6 x mean background activity for all reconstructions, and a SUV threshold of > 1.7 x mean background activity for BSREM150. The right column shows magnified views of the lesion with segmentation overlays corresponding to EARL1 (orange), EARL2 (cyan), BSREM150_TBR 1.6 (yellow), and BSREM150_TBR 1.7 (magenta). The figure illustrates the impact of reconstruction standards on lesion delineation, emphasizing differences in spatial resolution and quantitative PET metrics across reconstructions Comparative visualization of a brain lesion using T1-weighted gadolinium-enhanced MRI (T1gd) and [18F]FET PET imaging reconstructed with EARL1, EARL2, and BSREM150. The T1gd image (left) highlights the anatomical location of the lesion, which is further delineated in PET images (middle) using SUV thresholds of > 1.6 x mean background activity for all reconstructions, and a SUV threshold of > 1.7 x mean background activity for BSREM150. The right column shows magnified views of the lesion with segmentation overlays corresponding to EARL1 (orange), EARL2 (cyan), BSREM150_TBR 1.6 (yellow), and BSREM150_TBR 1.7 (magenta). The figure illustrates the impact of reconstruction standards on lesion delineation, emphasizing differences in spatial resolution and quantitative PET metrics across reconstructions Discussion This study systematically compared EARL1, EARL2, and BSREM150 reconstructions for [18F]FET PET in patients with glioma and suspected PD and directly quantified their effect on quantitative metrics and clinical interpretation. We found that quantitative PET metrics (SUVmean, SUVmax, SUVpeak, TBRmean, TBRmax, TBRpeak) and segmentation volumes (BTV) increased significantly when applying higher-resolution reconstructions. While all reconstructions achieved excellent diagnostic accuracy, applying a universal TBRmax threshold across reconstruction standards led to variable disease classification at both the lesion- and patient-level. EARL1 underestimated lesion extent and thus missed small PD lesions, whereas BSREM150 tended to overclassify PD. These findings are consistent with observations in other tumour types, where BSREM reconstructions increased lesion visibility but occasionally at the cost of false positives [21–25]. EARL2 resulted in borderline findings in four cases but overall demonstrated a balanced performance, suggesting it may represent the most clinically robust standard for amino acid PET harmonization at present. All cases classified as borderline by both EARL1 and EARL2 were ultimately assessed as PD by the MTB. In this cohort, which largely consisted of patients in the late pseudoprogression window, a TBRmax threshold of > 1.9 appeared sufficient to discriminate PD from TRC. These observations have direct implications for estimating tumour extent using semi-automatic segmentation thresholds and ensuing TBR-based thresholds for disease classification. BTV, a key metric for PET-RANO response assessment [8] and an emerging parameter for treatment planning [26, 27], varied substantially across reconstructions. EARL1 underestimated BTV in smaller lesions due to partial-volume effects. Conversely, BSREM150 required threshold adjustment to SUV ≥ 1.7 x mean background uptake to yield optimal segmentation concordance while still identifying more MD. These results emphasize that segmentation thresholds established under earlier reconstruction standards cannot be universally transferred to higher-resolution reconstructions without recalibration. The observed variability in SUV, TBR, and BTV across reconstructions reflects the known dependence of PET quantification on image-processing parameters. Our findings confirm that applying a universal threshold (e.g., TBRmax > 2.3) across reconstructions can lead to misclassification, consistent with prior meta-analyses reporting TBRmax thresholds ranging from 1.9 to 3.4 [28, 29]. Importantly, much of the published literature does not specify the reconstruction standard used, limiting reproducibility and comparability across studies. In addition, the current guideline by Law et al. [20] predates the 2021 WHO classification of CNS tumours [30], meaning that the thresholds derived from that guideline may not reflect current clinical practice. The recent PET-RANO guidelines provide a valuable update on the diagnostic work-up of amino acid PET but focus primarily on response assessment, leaving an updated framework for distinguishing PD from TRC still needed. In current clinical practice, amino acid PET assessment should explicitly account for the applied reconstruction standard, with preference for EARL1 or EARL2 reconstructions. When these reconstructions are unavailable, classification of PD based solely on TBR-based thresholds should be interpreted with caution, and combined visual assessment alongside MRI findings should be incorporated into clinical decision‑making. Both the RANO working group [8] and others, including Langen et al. [31], have recently cautioned against universal thresholds. We add that quantitative interpretation should always be evaluated in the context of reconstruction standards, spatial resolution, lesion size, and clinical context. Approaches integrating probabilistic or confidence-based metrics may provide more robust classification of infiltrative glioma regions, where malignant and treatment induced alterations often coexist [32]. Future harmonization efforts should integrate biopsy-validated, reconstruction-specific thresholds and reference phantoms optimized for neuro-oncology to ensure reproducible assessment across centres. Broad validation, such as in EORTC-Brain Tumour Group sites, will be essential to maximize reproducibility and facilitate adoption in routine clinical practice [9]. The dedicated EARL brain PET accreditation, based on signal recovery in a Hoffman brain phantom and targeting ~ 5 mm FWHM resolution [33], primarily harmonizes grey matter contrast and is well suited to neurodegenerative imaging. However, it may not fully reflect the focal lesion–to–background uptake patterns typical of brain tumours. For brain tumour imaging, oncological EARL1 and EARL2 standards using phantoms with fillable spheres, such as the NEMA Image Quality phantom, more closely represent tumour uptake and size-dependent recovery [12, 13]. Beyond reconstruction, combining PET with advanced MRI biomarkers and radiomics offers promising avenues for further improving differentiation of PD from TRC [29]. Multiparametric models incorporating [18F]FET PET, diffusion, and perfusion imaging, or artificial intelligence-based fusion analyses, may further enhance diagnostic accuracy and guide personalized therapy planning. Limitations of this study include the retrospective design, modest sample size, and heterogeneous glioma subtypes. Because follow-up outcome classification was derived from retrospectively collected data, final classification of PD frequently depended on MTB consensus rather than on histopathological validation. This limitation may have introduced a degree of incorporation bias, whereby the PET‑MRI findings influenced both therapeutic decisions and outcome classification. Nevertheless, our cohort reflects real-world clinical use of [18F]FET PET as a problem-solving tool in patients with equivocal MRI findings. Conclusions Reconstruction standards profoundly affect quantitative [18F]FET PET metrics and clinical interpretation. Universal thresholds across reconstructions lead to inconsistent tumour segmentation and variable disease classification. EARL1 may underestimate tumour extent and miss progressive disease, whereas BSREM150 can increase false-positive findings. EARL2 currently offers a balanced compromise between harmonization and resolution, but reconstruction-specific calibration remains essential for reliable [18F]FET PET interpretation and PET-RANO implementation. Future harmonization initiatives should incorporate reconstruction-specific thresholds and standardized reference phantoms to ensure reproducible, clinically reliable application of [18F]FET PET in routine neuro-oncologic care and PET-RANO in clinical trials. Supplementary Information Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3