PET/MR in recurrent glioblastoma patients treated with regorafenib: [18F]FET and DWI-ADC for response assessment and survival prediction

Objective: The use of regorafenib in recurrent glioblastoma patients has been recently approved by the Italian Medicines Agency (AIFA) and added to the National Comprehensive Cancer Network (NCCN) 2020 guidelines as a preferred regimen. Given its complex effects at the molecular level, the most appropriate imaging tools to assess early response to treatment is still a matter of debate. Diffusion-weighted imaging and O-(2-18F-fluoroethyl)-L-tyrosine positron emission tomography ([18F]FET PET) are promising methodologies providing additional information to the currently used RANO criteria. The aim of this study was to evaluate the variations in diffusion-weighted imaging/apparent diffusion coefficient (ADC) and [18F]FET PET-derived parameters in patients who underwent PET/MR at both baseline and after starting regorafenib. Methods: We retrospectively reviewed 16 consecutive GBM patients who underwent [18F]FET PET/MR before and after two cycles of regorafenib. Patients were sorted into stable (SD) or progressive disease (PD) categories in accordance with RANO criteria. We were also able to analyze four SD patients who underwent a third PET/MR after another four cycles of regorafenib. [18F]FET uptake greater than 1.6 times the mean background activity was used to define an area to be superimposed on an ADC map at baseline and after treatment. Several metrics were then derived and compared. Log-rank test was applied for overall survival analysis. Results: Percentage difference in FET volumes correlates with the corresponding percentage difference in ADC (R = 0.54). Patients with a twofold increase in FET after regorafenib showed a significantly higher increase in ADC pathological volume than the remaining subjects (p = 0.0023). Kaplan–Meier analysis, performed to compare the performance in overall survival prediction, revealed that the percentage variations of FET- and ADC-derived metrics performed at least as well as RANO criteria (p = 0.02, p = 0.024 and p = 0.04 respectively) and in some cases even better. TBR Max and TBR mean are not able to accurately predict overall survival. Conclusion In recurrent glioblastoma patients treated with regorafenib, [18F]FET and ADC metrics, are able to predict overall survival and being obtained from completely different measures as compared to RANO, could serve as semi-quantitative independent biomarkers of response to treatment. Advances in knowledge Simultaneous evaluation of [18F]FET and ADC metrics using PET/MR allows an early and reliable identification of response to treatment and predict overall survival.


INTRODUCTION
Glioblastoma multiforme (GBM), the most common primary malignant brain tumor in adults, still carries a dismal prognosis, with a median overall survival of less than 24 months, even after maximal safe resection, concomitant chemoradiotherapy and adjuvant temozolomide. [1][2][3] In the setting of disease relapse, the use of regorafenib has been recently approved by the Italian Medicines Agency (AIFA) and added to the new National Comprehensive Cancer Network (NCCN) 2020 guidelines as a preferred regimen, based on promising results from a multicenter Phase II trial (REGOMA) comparing this new drug with the standard lomustine regimen. 4 Regorafenib is an orally available multi kinase inhibitor with several molecular targets involved in angiogenesis (VEGFR1-3 and TIE2), oncogenesis (KIT, RET, RAF1, and BRAF) and maintenance of the tumoral microenvironment (PDGFR and FGFR). [5][6][7] Given the complexity of its effects at the molecular level, the choice of the most appropriate imaging parameters to be used with patients treated with regorafenib is still a matter of debate. Currently, the recommendations of the Response Assessment in Neuro-Oncology (RANO) Study Group are widely used in both clinical practice and research settings and were also implemented in the REGOMA trial. 4,8 The RANO criteria are based on measurement of areas of contrastenhancement on post-gadolinium T 1 weighted sequences and of non-enhancing disease captured on T 2 weighted/fluid attenuated inversion recovery (FLAIR) images. This approach, however, has already been shown to have several limitations and shortcomings in patients treated with anti angiogenetic drugs, such as bevacizumab, given the normalization of vascular permeability and the related decrease in contrast enhancement induced by these agents. [8][9][10] In fact, up to 40% of patients treated with bevacizumab show seemingly stable contrast-enhancing disease with an increase in T 2 weighted/FLAIR signal abnormalities, indicating disease progression. 11 Moreover, the lack of a quantifiable measure of non-enhancing disease progression, and the confounding effect of radiation therapy, ischemic injury, and post-operative changes on FLAIR images further complicate the issue.
Diffusion-weighted imaging (DWI) is a promising methodology that could improve the assessment of treatment response in GBM, thereby extending the existing RANO criteria. 12 It is based on measuring the Brownian motion of water molecules and the various constraints that hamper this physical phenomenon in live tissues. Moreover, DWI-derived apparent diffusion coefficient (ADC) maps offer quantitative information related to tumor cellularity and have already been used in glioma patients to detect the presence of neoplastic tissue in the peritumoral edema. 13,14 Since necrosis, ischemia and inflammation are known to influence water diffusion, heterogeneous ADC values are usually evident in tumoral areas, especially after treatment. 15,16 Consequently, the mean ADC values of one area can fail to depict the spatial heterogeneity of brain tumors, although histogram analysis has already been successfully used as a possible workaround. 16,17 O-(2-18 F-fluoroethyl)-L-tyrosine ( 18 F-FET), an amino acid tracer used in positron emission tomography (PET), is another important tool routinely used for therapy assessment during anti-angiogenetic treatment. 18,19 Even though several studies have already demonstrated the additional value of amino acid PET over conventional MR-based assessment in this setting, 20,21 the interplay between ADC and [ 18 F]FET PET in patients treated with regorafenib has been explored so far only in a small case series comprising five cases. 22 The aim of this study was to evaluate the variations in DWI/ ADC-and [ 18 F]FET PET-derived parameters in recurrent patients undergoing PET/MR both at baseline and after beginning regorafenib. Furthermore, we analyzed the performance in survival prediction of RANO criteria compared to DWI/ADCand [ 18 F]FET PET-derived parameters.

METHODS AND MATERIALS
This was a single-center, retrospective, observational study conducted in accordance with the Declaration of Helsinki and after formal approval by our local Ethics Committee (protocol number: AOP1673 -4831/AO/20 In cases of assumed CR or PR at the post-regorafenib time point, a follow-up MR scan was performed at least 4 weeks later and reviewed for confirmation. PD was defined (according to RANO) as the fulfillment of one or more of the following conditions: (1) ≥ 25% increase in the sum of the products of the perpendicular diameters of the enhancing lesions compared with the smallest tumor measurement at baseline; (2) appearance of any new contrast-enhancing lesion; (3) significant increase in T2/FLAIR non-enhancing lesion.
Patients fell into the SD category if they did not meet the conditions for CR, PR, or PD, and were administered the same or a lower dose of corticosteroids.

Image data processing
The images were imported into PMOD (PMOD ® Technologies LLC, Zurich, Switzerland) for volume of interest (VOI) delineation.
The mean standardized uptake value of a crescent-shaped VOI (BG FET ), manually drawn in the hemisphere contralateral to the tumor, was used as the [ 18 F]FET background. 27 The pathological FET volume (FET vol/pat ) was segmented through a 3D semiautomatic contouring process, excluding areas with an [ 18 F]FET uptake less than 1.6 times the background mean activity. This threshold was based on an [ 18 F]FET biopsy-controlled study, where it was proven to accurately differentiate between tumoral and non-tumoral tissue. 28 The chosen cut-off has been subsequently used successfully in a number of publications presenting histopathological confirmation and/or MR comparisons. 29,30 The derived segmented volume was visually refined to exclude areas of non-specific [ 18 F]FET spillover (major blood vessels, cranial bones, meninges etc.) using the aligned MDC images as the morphological reference.
FET vol/pat was then superimposed onto the ADC images ( Figure 1) to obtain the corresponding ADC volume (ADC vol ). The details are as follows: (1) FET vol/pat was imported into the aligned ADC image.
(2) Areas with non-specific high ADC values were subtracted (with the aim also to correct for anatomical distortions induced by metal implants and air filled cavities) from the original volume, pinpointing the ADC values in the cerebrospinal fluid of the lateral ventricles. (3) Areas of the original volume located outside the brain parenchyma were analogously subtracted.
A standard spherical volume (radius = 5 mm) was then placed on the ADC images in the hemisphere contralateral to the tumor, carefully avoiding lateral ventricles and major vessels, in order to derive the mean ADC value of the normal brain parenchyma (BG ADC ). This method was chosen in view of the stability of the ADC values in the "healthy" brain parenchyma during treatment with antiangiogenetic agents. 12 A qualitative assessment of the high resolution DWI and ADC derived maps was performed in every patient and revealed no significant distortions or misregistrations affecting the selected tumor area or background area.
The quality of alignment and segmentation was finally checked by an experienced nuclear medicine physician (with more than 10 years' experience in the field of neuro-oncology).

Data analysis
A pixel dump of FET vol/pat , BG FET , ADC vol , and BG ADC was imported into the R software 31 for further analyses. The mean ADC value of the BG ADC was used as a threshold for ADC vol .

of 1birpublications.org/bjr
Statistical analysis All statistical analyses were performed using the R Software. The Shapiro-Wilk normality test was performed on the distribution of all the parameters. Where normal distributions could not be assumed, non-parametric tests were performed. The percentage changes in ADC vol/pat and FET vol/pat before and after regorafenib were plotted and the Pearson correlation coefficient (PCC) calculated, assuming a linear correlation between the two variables. The differences in the percentage changes in FET vol/pat and ADC vol/pat between the response groups determined according to RANO criteria 8 were compared using the Wilcoxon signedrank test for repeated measures. The significance level (α) was set at 0.05. Log-rank test was applied for overall survival analysis. A p-value < 0.05 was considered significant.

Patients
Our study population consisted of 15 IDH-wt and one glioblastoma NOS patients (6 females, 10    In two of the study subjects at baseline and in one post-regorafenib no pixels remained after ADC normalization. The RANO criteria were used to sort the patients into response categories. Br J Radiol;95:20211018 7 of 14 birpublications.org/bjr absolute and percentage variations after treatment are presented in Table 4. Although the absolute and percentage increases in FET vol/pat were on average higher in PD than SD patients (21,605 mm 3 and 168% vs −1160 mm 3 and 70%), the differences between the two groups were not statistically significant (p = 0.17). Similarly, the average absolute and percentage increases in ADC pathological volume were also higher in PD than SD subjects (501 mm 3 and 554% vs 33 mm 3 and 297%), and also failed to reach statistical significance (p = 0.53). The percentage variations in mean ADC, FET vol/pat , TBR max and TBR mean did not differ significantly between SD and PD patients (Tables 3 and  4, Figure 2). When the percentage difference in FET pathological volumes was plotted against the corresponding percentage difference in ADC pathological volumes, a linear regression model revealed a correlation between the two variables (R = 0.54) (Figure 3). We found no evident correlation between the percentage variation in mean ADC values and the corresponding percentage variation in FET vol/pat (R = 0.04). Patients with at least a twofold increase in FET pathological volume ( Figure 4) after regorafenib showed a significantly higher increase in ADC pathological volume than the remaining subjects (p = 0.0023). In 2/9 subjects classified as progressive (according to RANO) after two cycles of regorafenib, the FET pathological volume decreased by 76 and 31%, respectively. Consistent with this, a decrease in the ADC pathological volume was observed in the former (−93%), while no residual pathological ADC areas could be detected in the latter. In contrast, in 3/7 patients classified as stable (according to RANO) after treatment, an increase in FET pathological volume (5895 mm 3 , 2057 mm 3 , and 1009 mm 3 , respectively) was observed; in the same patients, the ADC pathological volume increased at similar rates (273 mm 3 , 88 mm 3 , and 425 mm 3 ).
The Kaplan-Meier analysis showed that the percentage variations of FET PAT/VOL, ADC PAT/VOL and RANO criteria were able to predict overall survival (p = 0.02, p = 0.024 and p = 0.04 respectively). TBR Max and TBR mean were not able to accurately predict overall survival.   [32][33][34][35] and many authors have suggested that the DWI methodology could play an important role in guiding response assessment, particularly when conventional contrast-enhanced and T 2 weighted/FLAIR sequences are less reliable. Although DWI sequences are routinely acquired as part of the standard MR protocol for brain tumor imaging, the most recent recommendations 36 only describe how diffusion-weighted images should be acquired and provide no guidance for clinically interpreting and quantifying the extent of the tumor for the purpose of response evaluation. Two major issues consequently   arise, the first regarding the strategy to identify the region on the DWI-ADC images to be analyzed, the second regarding the threshold for pathological ADC values. In most of the published studies, the tumor volume was outlined and the VOI constructed on contrast-enhanced T 1 weighted images, which were subsequently transferred to the corresponding DWI-ADC images. Buemi et al, 37 e.g. manually drew the VOIs encompassing the areas of tumor-related contrast enhancement, and T 2 weighted/ FLAIR abnormalities were mapped onto the corresponding ADC images, thus deriving the CE-ADC and T2/FLAIR-ADC volumes, respectively. Histogram analysis and curve fitting using a two-mixture normal distribution model were carried out to calculate the mean ADC of the lowest ADC values in these areas (CE-ADC-L and T2/FLAIR-ADC-L). 37 Interestingly, only the mean ADC in CE-ADC-L turned out to be significantly predictive of progression-free survival and overall survival in GBM patients treated with bevacizumab and fotemustine. The predictive value of the low-ADC areas is confirmed by other published papers. 38,39 Zeiner et al 40 calculated the ADC-ratio by measuring the minimum ADC values in the tumor and normalizing them by the ADC values of the contralateral, normal appearing brain tissue. In our study, instead, a standard VOI contralateral to the lesion was used to determine the appropriate threshold for the selection of pathological ADC values. This approach allowed for a more direct identification of the low ADC values in the defined VOIs, avoiding the need for complex mathematical models. However, the post-regorafenib variation in the mean ADC thus calculated did not significantly correlate with the corresponding change in FET-positive volume, nor with the RANO response categories. A possible explanation for this discrepancy may lie in the different methodological approaches used here and in the previously published studies. 12,[37][38][39][40][41][42][43][44][45][46][47][48] This highlights the importance of future efforts towards standardizing the analysis of ADC maps before considering the inclusion of this methodology in the response assessment criteria.
It is important to note that the changes in neither the pathological FET volume nor the pathological ADC volume were significantly different in the stable and progressive patients as assessed by RANO criteria. RANO criteria are based mostly on changes in T2/FLAIR and contrast-enhanced areas, which are known to be affected by edema, inflammation, gliosis, and disruption of the blood-brain barrier. DWI, instead, is sensitive to microscopic water motion, resulting in relatively restricted diffusion in areas of tightly packed tumor cells. However, diffusion may be altered by causes other than increased cellularity in neoplastic tissue, and the diagnostic performance of this methodology is influenced by the choice of the appropriate DWI parameter to analyze. 16,49,50 Moreover, the heterogeneity of the ADC signal may have translated into the wide variability we observed in the changes in the pathological ADC volume after regorafenib. This, in turn, may explain why a relatively high threshold of increase in FET pathological volume was needed to subdivide our population into groups with significantly different pathological ADC volumes.
In three cases which were classified, according to the RANO criteria, as stable (SD) after treatment with regorafenib, both the ADC and FET pathological volumes increased compared with the baseline examinations (patients #5, #8 in Figure 5, and #13). Information from subsequent follow-ups was available for two of these patients: a) patient #5 (interestingly, classified as SD by the RANO criteria) showed a slight increase in FET vol/pat (and to a lesser extent also in ADC vol/pat ) at a PET/MR examination 10 of 14 birpublications.org/bjr ( Figure 6) performed 2 months later (TP3 in Figure 5), and presented disease progression at an MR scan performed 4 months later; b) patient #8 showed a significant increase in FET vol/pat (and to a lesser extent also in ADC vol/pat ) at a subsequent PET/ MR examination (TP3 in Figure 5), and was consistently considered progressive according to the RANO criteria. We were able to carry out a follow-up PET/MR in another two cases (patients #9 and #16, both SD at the PET/MR examination after two cycles of regorafenib): a) patient # presented minimal variations in FET vol/ pat and ADC vol/pat after two cycles of regorafenib, and remained stable (presenting a decrease in FET vol/pat and ADC vol/pat ) at the follow-up PET/MR (TP3 in Figure 5); b) patient #16 showed a significant increase in FET vol/pat and ADC vol/pat between the PET/MR performed after two cycles of regorafenib ( Figure 5) and the follow-up PET/MR, and was then considered progressive according to the RANO criteria.
These four cases, although insufficient to draw definitive conclusions, seem to show consistent variations in FET and ADC pathological volumes in follow-up examinations performed after six cycles of regorafenib, and seems to confirm the greater predictive value of these parameters compared with the standard RANO criteria. In fact, the variations in FET and ADC (in the PET/MR after two cycles of regorafenib) predicted the follow-up in two out of four cases wrongly classified by RANO (subjects #5 and #8 who were categorized as stable according to the RANO criteria).
Kaplan-Meier analysis (Figure 7), performed to compare the performance in overall survival prediction, revealed that the percentage variations of FET PAT/VOL and ADC PAT/VOL performed at least as well as RANO criteria (p = 0.02, p = 0.024 and p = 0.04 respectively) or even better. TBR Max and TBR mean on the other hand, frequently used at the first tumor occurrence, are not able to accurately predict overall survival.
Therefore, the so identified [ 18 F]FET and ADC areas and values, which are correlated but were obtained from completely different measures, could serve as independent biomarkers of treatment response and could, at least, complement the RANO criteria especially in doubtful cases. Our study has some limitations, including that regorafenib was introduced only recently: (1) It was retrospective in nature and included only a relatively (considering the actual infrequent use of the treatment) small number of patients.
(2) The current RANO criteria were assumed as gold-standard.
Despite these limitations, we focused on a highly homogeneous patient population comprising GBM subjects at their first disease relapse, and all patients were treated with a recently approved chemotherapeutic agent (regorafenib). Moreover, all imaging studies were acquired at the same institution with an integrated PET/MR system and a standardized protocol. Figure 7. Kaplan-Meier analysis shows that the percentage variations of FET PAT/VOL and ADC PAT/VOL performed at least as well as RANO criteria or even better in terms of overall survival prediction (left column). TBR Max and TBR mean (right column) on the other hand are not able to accurately predict overall survival. ADC, apparent diffusion coefficient; FET, O-(2-18F-fluoroethyl)-Ltyrosine; TBR, Tumor-to-Background Ratio.
12 of 14 birpublications.org/bjr CONCLUSIONS In the present study, we have proposed a method to identify the pathological ADC volume based on the corresponding [ 18 F]FET positive region in intrinsically co-registered [ 18 F] FET PET/MR images. We found a correlation between the percentage changes in pathological FET and DWI-ADC volumes in glioblastoma patients treated with regorafenib at their first disease relapse. In 4/16 cases followed up with a third PET/MR, the results seemed encouraging compared to the RANO criteria.
Kaplan analysis showed that FET PAT/VOL and ADC PAT/VOL performed at least as well as RANO criteria in terms of overall survival prediction.
The [ 18 F]FET and ADC metrics identified could, given they were correlated but obtained from completely different measures, serve as semi-quantitative independent biomarkers of response to regorafenib treatment.

FUNDING
Open Access Funding provided by Universita degli Studi di Padova within the CRUI-CARE Agreement.