O-(2-[¹⁸F]fluoroethyl)-l-tyrosine PET in gliomas: influence of data processing in different centres

PET using the amino acid O-(2-¹⁸F-fluoroethyl)-l-tyrosine (¹⁸F-FET) has received increasing attention for brain tumour diagnostics due to logistic advantages of ¹⁸F labelling (half-life, 109.8 min) compared with l-[methyl-¹¹C]-methionine (¹¹C-MET) PET, efficient radiosynthesis and high in-vivo stability [1‐6]. Multiple studies have demonstrated the clinical potential of ¹⁸F-FET PET to determine the extent of cerebral gliomas for biopsy guidance, treatment planning, detection of tumour recurrence, estimation of prognosis in newly diagnosed and untreated gliomas and treatment monitoring [7‐12]. ¹⁸F-FET uptake in the tumour is usually expressed by mean and maximum tumour-to-brain ratios (TBR_mean, TBR_max) within a scan period between 20 and 40 min after injection. Furthermore, the time-activity curves (TACs) of ¹⁸F-FET uptake typically show differences in high-grade and low-grade gliomas or nonneoplastic lesions which provide valuable additional information for tumour grading or differential diagnosis [13‐15]. Thus, continuously increasing ¹⁸F-FET uptake is more frequently observed in low-grade gliomas and nonneoplastic lesions, while kinetics with an early peak of ¹⁸F-FET uptake in the first 10–20 min after injection followed by a decreasing TAC is a common finding in more aggressive tumours like high-grade glioma or brain metastases [10, 16, 17]. TBR and dynamic parameters of ¹⁸F-FET uptake may be influenced on the one hand by the spatial resolution of the PET scans which is dependent on the scanner type, reconstruction algorithms and data filtering and on the other hand by the definition of the region of interest (ROI) in the tumour and the brain. There is some controversy about the diagnostic accuracy of TAC analysis for tumour grading raising the question whether discrepant results in different centres may be caused by differences in the methodology, patient population or other reasons such as neuropathological interpretation.

In the present study, methodological differences between two large centres in Germany (A = Ludwig-Maximilians-University Munich, Germany; B = Forschungszentrum Jülich, Germany) with a high number of ¹⁸F-FET PET investigations (> 500/year) and multiple publications in the field were identified. Based on a balanced group of 20 patients with high-grade glioma of WHO grade III and IV (HGG) and 20 patients with low-grade glioma of WHO grade II (LGG) [18] who were investigated with the same scanner used in both centres, the influence of methodological differences between the two centres on common parameters for tumour characterization with ¹⁸F-FET PET was evaluated. This analysis may be helpful to develop a more standardized approach in order to make the data between different centres more comparable.

PET using radiolabelled amino acids is gaining increasing interest for the diagnostics of brain tumours because conventional MRI is limited in differentiating tumour tissue from nonspecific tissue changes, especially after therapy [21]. Recently, the Response Assessment in Neuro-Oncology (RANO) working group—an international effort to develop new standardized response criteria for clinical trials in brain tumours—has recommended the additional use of amino acid PET imaging for brain tumour management [5]. The longest established amino acid tracer ¹¹C-MET has been replaced in many neuro-oncology centres by the more convenient ¹⁸F-FET, and several thousand ¹⁸F-FET PET scans have been performed in some centres in recent years [22]. The broad clinical use of ¹⁸F-FET PET requires comparable quantitative parameters, but as yet, the reported cut-off values of different parameters like TBR_mean, TBR_max, TTP, slope and T_vol for tumour grading or differentiation of recurrent tumour from treatment related changes appear to vary among different centres. It remains unclear whether diverging results are caused by the composition of the study population, differences in the technical equipment or differences in data processing or a combination thereof.

In this study, we have identified methodological differences between two centres in Germany which have a high frequency of ¹⁸F-FET PET investigations and contributed to the clinical evaluation of ¹⁸F-FET PET through numerous publications (for review, see [3, 5]). Since in both centres, the previous publications on ¹⁸F-FET PET studies were performed with the same PET scanner, differences in the technical equipment between the centres can be excluded. A comparison of data processing, however, identified methodological differences between the centres concerning the framing of PET data, data reconstruction, tumour delineation and ROI definition to evaluate tracer kinetics in the tumour. The effects of these methodological differences on quantitative parameters of FET PET were examined in a collective of 40 brain tumour patients.

One frequently used parameter for brain tumour characterization is the TBR_max which showed approximately 15% lower values when generated with the methodology of centre A than with that of centre B. In agreement with these findings, we observed lower peak values in a phantom study when data were reconstructed according to the method of centre A. In accordance with these findings, the reported average TBR_max values in primary brain tumours are lower in centre A than in centre B (HGG: A/TBR_max 3.3 ± 1.2 versus B/TBR_max 3.6 ± 1.4, LGG: A/TBR_max 2.1 ± 1.0 versus B/TBR_max 2.4 ± 1.0) [23, 24]. Nevertheless, the cut-off values of the TBR_max for differentiating between HGG and LGG in those studies were similar (cut-off: A/TBR_max 2.7, B/TBR_max 2.5), and the accuracy of differentiating between HGG and LGG based on the TBR_max appears to be similar to both centres. For the differentiation of recurrent tumours from treatment-related changes which is a frequent clinical question, the reported cut-off value for TBR_max was 2.0 in centre A and 2.3 in centre B which might also reflect the difference in data processing in the two centres [12, 25]. TBR_mean values generated with the methodology of centres A and B showed a small but significant difference in the paired t test (A/TBR_mean 2.20 ± 0.41 versus and B/TBR_mean 2.16 ± 0.41, p < 0.04), but this difference would no longer be significant if a correction for multiple testing were applied.

A major difference between both centres was observed for the definition of the biological tumour volume. The approach of centre A led to approximately 25% smaller tumour volumes than that of centre B which is of substantial importance since ¹⁸F-FET PET is increasingly used when planning therapeutic interventions such as surgery and radiotherapy. It has to be considered that the approach of centre A is based on clinical experience while the approach of centre B was developed on the basis of a biopsy-controlled study which from a scientific point of view has to be regarded as more reliable [8]. On the other hand, that study is based on 52 biopsies in 31 patients only, which appears not sufficient to establish a clinical standard. Furthermore, only patients with newly diagnosed brain tumours were included, and the optimal threshold might be different for pretreated patients who can present slight unspecific tracer uptake at the primary tumour site. Therefore, further biopsy-controlled studies are needed to clarify this clinically important aspect.

The evaluation of dynamic ¹⁸F-FET PET by the approach of centre A yielded shorter TTP values and lower values for the slope in the late phase of the TAC. A detailed analysis showed that the different framing procedures in centres A and B did not influence the dynamic parameters (Fig. 7) and also the ROI definition by 90% isocontour instead of TBR > 1.6 caused only minor differences. The decisive difference was caused by the more extensive search for tumour areas with negative slope in the entire tumour volume in centre A. This approach appears to be advantageous but is technically more challenging. On the other hand, we did not detect a significant difference between the two approaches for the differentiation between HGG and LGG by ROC analysis in this group of patients.

An important reason for the differences in TBR_max and T_vol observed in both centres is the different spatial resolution of the PET scans which is caused by the different reconstruction methods. It appears that the use of FBP in centre A is a major cause of the observed differences which is also evident from the phantom study (Fig. 2). This observation is in line with previous studies comparing the difference of FBP to OSEM on quantification of glucose metabolism which all report higher SUVs for OSEM as compared to FBP [26‐28]. Since in newer PET systems, data processing is generally based on iterative reconstruction differences between centres which may become smaller in the future. In any case, the spatial resolution of PET scans should be comparable when applying threshold values for the definition of ROIs in the tumour area with FET PET. It is therefore necessary to adapt the reconstruction parameters in order to achieve a similar spatial resolution in various PET systems. Kinetic parameters are less strongly influenced by spatial resolution but are altered by the selection of specific tumour regions.

It needs to be considered that owing to the lack of molecular data for histological analysis, tumour classification was based on the WHO classification from 2007 instead of the new one from 2016, which includes molecular parameters [18, 29]. Therefore, the AUC values obtained for tumour grading in this study might not be directly applicable to the current classification, and an influence on the comparison of the AUC values for centres A and B cannot be excluded.

The differences in data processing between the two centres lead to considerable differences especially for TBR_max, slope, TTP and T_vol. Although there was a strong linear correlation between TBR_max and T_vol values of both centres, the absolute values cannot directly be compared. Absolute TBR_mean values, by contrast, seem comparable between both centres. However, concerning the evaluation of dynamic ¹⁸F-FET PET, both centres achieved comparable high accuracy to discriminate between HGG and LGG. A standardization of data processing and protocols for ¹⁸F-FET-PET is needed in order to make clinical results comparable.