Prediction of survival in patients with IDH-wildtype astrocytic gliomas using dynamic O-(2-[¹⁸F]-fluoroethyl)-l-tyrosine PET

We retrospectively assessed data from 60 adult glioma patients (median age, 55 years; age range, 21–78 years; 25 women and 35 men) of two university centres with neuropathologically confirmed newly diagnosed malignant astrocytic glioma (GBM, n = 45; anaplastic astrocytoma (AA), n = 15) who underwent dynamic FET PET imaging at the Research Centre Juelich, Germany, at initial diagnosis, i.e. before biopsy, resection, or any other kind of treatment. The data had been acquired from 2009 to 2017. Clinical data of all patients, including first-line treatment, are listed in Table 1. The local ethics committees approved the retrospective analysis of the data. There was no conflict with the Declaration of Helsinki. Before PET imaging, all patients had given written informed consent for the PET investigation and the use of the data for scientific purposes.

As described previously, the amino acid FET was produced via nucleophilic ¹⁸F-fluorination with a radiochemical purity of greater than 98%, specific radioactivity greater than 200 GBq/μmol, and a radiochemical yield of about 60% [28]. According to national and international guidelines for brain tumour imaging using labelled amino acid analogues [29, 30], all patients fasted for at least 4 h before the PET measurements. All patients underwent a dynamic PET scan from 0 to 50 min post-injection of 3 MBq of FET per kg of body weight. PET imaging was performed either on an ECAT Exact HR+PET scanner (n = 37 patients) in 3-dimensional mode (Siemens, Erlangen, Germany) (axial field-of-view, 15.5 cm) or simultaneously with 3 T MR imaging using a BrainPET insert (n = 23 patients) (Siemens, Erlangen, Germany). The BrainPET is a compact cylinder that fits into the bore of the Magnetom Trio MR scanner (axial field of view, 19.2 cm) [31].

Iterative reconstruction parameters were 16 subsets, 6 iterations using the OSEM algorithm for the ECAT HR+PET scanner and two subsets, and 32 iterations using the OP-OSEM algorithm for the BrainPET. Data were corrected for random, scattered coincidences, dead time, and motion for both systems. Attenuation correction for the ECAT HR+PET scan was based on a transmission scan, and for the BrainPET scan, a template-based approach [31]. The reconstructed dynamic data sets consisted of 16 time frames (5 × 1 min; 5 × 3 min; 6 × 5 min) for both scanners.

To optimize comparability of the results related to the influence of the two different PET scanners, reconstruction parameters, and post-processing steps, a 2.5-mm 3D Gaussian filter was applied to the BrainPET data before further processing, resulting in an image resolution of approximately 4 mm full width at half maximum (FWHM) (image resolution of the ECAT HR+PET scanner, approximately 6-mm FWHM). In phantom experiments using spheres of different sizes to simulate lesions, this filter kernel demonstrated the best comparability between PET data obtained from the ECAT HR+PET and the BrainPET scanner [32].

FET PET data analysis was performed as described previously [33]. In brief, for the evaluation of FET data, summed PET images over 20–40 min post-injection were used. A larger crescent shaped reference ROI was placed in the semioval centre of the contralateral unaffected hemisphere including white and grey matter [30]. Mean amino acid uptake in the tumour was determined by a 2-dimensional auto-contouring process in the transversal slice containing the voxel with the maximum tumour uptake using a tumour-to-brain ratio (TBR) of 1.6, as described previously [34, 35]. For the calculation of the maximal amino acid uptake, a circular region-of-interest (ROI) with a diameter of 1.6 cm was centred on the voxel with the maximum tumour uptake [33]. Maximum and mean TBRs (TBR_max, TBR_mean) were calculated by dividing the mean SUV of the tumour ROIs by the mean SUV of healthy brain tissue. The FET metabolic tumour volume (MTV) was determined by a 3-dimensional auto-contouring process using a TBR of 1.6 or more using the software PMOD (Version 3.9, PMOD Technologies Ltd.).

As described previously [33], time-activity curves (TACs) of FET uptake (mean SUV) in the tumour were generated by the application of a spherical volume-of-interest (VOI) with a volume of 2 mL centred on the voxel with the maximum tumour uptake and the reference ROI as described above to the entire dynamic dataset. A reference TAC was generated by placing a reference ROI in the unaffected brain tissue as reported [33]. For TAC evaluation, the time-to-peak (TTP; time in minutes from the beginning of the dynamic acquisition up to the maximum SUV of the lesion) was determined. In cases with steadily increasing FET uptake without identifiable peak uptake, we defined the end of the dynamic PET acquisition as TTP. Furthermore, the slope of the TAC in the late phase of FET uptake was assessed by fitting a linear regression line to the late phase of the curve (20–50 min post-injection). The slope was expressed as the change of the SUV per hour. This procedure allows for a more objective evaluation of kinetic data compared with an assignment of TACs to earlier reported patterns of FET uptake during dynamic acquisition [33].

The ROIs and VOIs are selected, positioned and validated by two experienced, board-certified specialists in nuclear medicine (G.S., K-J.L.) with more than 10 years of experience in the analysis of FET PET images.

All tumours were neuropathologically re-classified according to the WHO Classification of Tumours of the Central Nervous System of 2016 [7]. For molecular biomarker analysis, tumour DNA was extracted from formalin-fixed and paraffin-embedded tissue samples with a histologically estimated tumour cell content of 80% or more. For assessment of the IDH mutation status, the presence of an IDH1-R132H mutation was evaluated by immunohistochemistry using a mutation-specific antibody in a standard immunohistochemical staining procedure as reported before [36, 37]. When immunostaining for IDH1-R132H remained negative, the mutational hot-spots at codon 132 of IDH1 and codon 172 of IDH2 were directly sequenced as reported [10, 38]. The 1p/19q co-deletion status was determined by PCR-based microsatellite analysis as reported [39]. The O⁶-methylguanine-DNA-methyltransferase (MGMT) promoter methylation status was assessed by methylation-specific PCR as described elsewhere [38].

Progression-free survival (PFS) was defined as the time in months between the neuropathologically confirmed glioma diagnosis and tumour progression according to the response assessment in neuro-oncology (RANO) working group [40]. The overall survival (OS) was defined as the time in months between the neuropathologically confirmed glioma diagnosis and death.

Descriptive statistics are provided as mean and standard deviation or median and range. The Student t test for independent samples was used to compare two different groups. The Mann-Whitney rank-sum test was used when variables were not normally distributed. The diagnostic performance of FET uptake, as determined by TBR_max, TBR_mean, TTP, and slope, was assessed by receiver operating characteristic (ROC) curve analyses using a favourable OS of ≥ 24 months as reference (the median OS in GBM patients is in the range of 15–20 months [41‐43], and the 2-year survival rate is 30% [44]; therefore, an OS of ≥ 24 months was considered as favourable). The decision cutoff was considered optimal when the product of paired values for sensitivity and specificity reached its maximum. As a measure of the diagnostic quality of the test, we determined the area under the ROC curve (AUC), its standard error, and the level of significance. Univariate survival analyses were performed using the log-rank test. Multivariate Cox proportional hazards models were constructed to test the relationship between static and dynamic FET PET parameters and other predictors of survival. Parameters that were significant in univariate analyses were included in multivariate models. A P value of < 0.05 was considered significant. Statistical analyses were performed using SigmaStat software (SigmaPlot for Windows 11.0, Chicago, IL) and SPSS Statistics software (Release 24.0, SPSS Inc., Chicago, IL, USA).

In the whole cohort (n = 60), the median PFS was 12 months (range, 0–75 months), and the median OS was 17 months (range, 0–75 months). In the subgroup of patients with IDH-wildtype gliomas (n = 45), the median PFS was 9 months (range, 0–52 months), and the median OS was 15 months (range, 0–52 months). Following biopsy or resection, the vast majority of patients underwent radiotherapy plus an adjuvant treatment option (n = 52; 87%) (Table 1).

An overview of the results of the optimal thresholds derived from FET PET is provided in Table 2. Specifically, ROC analysis revealed that the dynamic FET PET parameter slope predicts a favourable OS of ≥ 24 months with a sensitivity of 70% and a specificity of 90% (AUC, 0.77 ± 0.09; threshold, − 0.103 SUV/h; P = 0.010). The most significant OS prediction of ≥ 24 months could be obtained by the dynamic FET PET parameter TTP (threshold, 25 min; sensitivity 90%; specificity, 87%; AUC, 0.90 ± 0.07; P < 0.001) (Figs. 1, 2, and 3). In contrast, the static FET PET parameters TBR_max, TBR_mean, and MTV were not prognostic for a favourable OS of ≥ 24 months (AUC, 0.63, 0.69, and 0.56, respectively).

In the whole cohort (n = 60), the univariate survival analysis revealed that glioma patients with IDH-mutant tumours had significantly longer PFS (33 vs. 9 months) and OS (54 vs. 13 months) than patients with IDH-wildtype gliomas (both P < 0.001). Similarly, glioma patients with MGMT promoter-methylated tumours had a significantly longer PFS (18 vs. 7 months) and OS (29 vs. 12 months) than glioma patients whose tumours lacked MGMT promoter methylation (both P < 0.001). Patients with a KPS of 100% had a significantly longer OS than patients with a KPS between 60 and 90% (29 vs. 13 months; P = 0.015). Furthermore, patients with complete resection of the glioma at initial diagnosis had a significantly longer PFS compared with incompletely resected patients or patients who received a biopsy only (13 vs. 10 months; P = 0.038), but not a significantly longer OS (P = 0.121). Patients aged < 70 years at initial diagnosis did not show longer PFS and OS compared with patients ≥ 70 years (P = 0.552 and P = 0.108, respectively) (Table 3).

In patients diagnosed with an IDH-wildtype glioma at initial diagnosis (n = 45), the dynamic FET PET parameter TTP (threshold, 25 min) predicted both a significantly longer PFS (13 vs. 7 months) and OS (29 vs. 12 months) (P = 0.005 and P < 0.001, respectively) (Figs. 4 and 5). Furthermore, the dynamic FET PET parameter slope (threshold, − 0.103 SUV/h) also predicted a significantly longer OS (17 vs. 9 months; P = 0.021), but not a significantly longer PFS (P = 0.065). Compared with TTP, TBR_max, TBR_mean, and MTV were only prognostic for a longer OS at a lower level of significance (P = 0.004, P = 0.038, and P = 0.048, respectively) (Table 3).

Besides MGMT promoter methylation and complete resection at initial diagnosis, the dynamic FET PET parameter TTP remained statistically significant in the multivariate survival analysis (P = 0.008; HR, 0.941; 95% CI, 0.900–0.984), indicating an independent predictor for OS. The parameters KPS, TBR_max, TBR_mean, slope, and MTV were not statistically significant in the multivariate survival analysis (Table 4).