Dynamic 18F-FDopa PET Imaging for Newly Diagnosed Gliomas: Is a Semiquantitative Model Sufficient?

Purpose Dynamic amino acid positron emission tomography (PET) has become essential in neuro-oncology, most notably for its prognostic value in the noninvasive prediction of isocitrate dehydrogenase (IDH) mutations in newly diagnosed gliomas. The 6-[18F]fluoro-l-DOPA (18F-FDOPA) kinetic model has an underlying complexity, while previous studies have predominantly used a semiquantitative dynamic analysis. Our study addresses whether a semiquantitative analysis can capture all the relevant information contained in time–activity curves for predicting the presence of IDH mutations compared to the more sophisticated graphical and compartmental models. Methods Thirty-seven tumour time–activity curves from 18F-FDOPA PET dynamic acquisitions of newly diagnosed gliomas (median age = 58.3 years, range = 20.3–79.9 years, 16 women, 16 IDH-wild type) were analyzed with a semiquantitative model based on classical parameters, with (SQ) or without (Ref SQ) a reference region, or on parameters of a fit function (SQ Fit), a graphical Logan model with input function (Logan) or reference region (Ref Logan), and a two-tissue compartmental model previously reported for 18F-FDOPA PET imaging of gliomas (2TCM). The overall predictive performance of each model was assessed with an area under the curve (AUC) comparison using multivariate analysis of all the parameters included in the model. Moreover, each extracted parameter was assessed in a univariate analysis by a receiver operating characteristic curve analysis. Results The SQ model with an AUC of 0.733 for predicting IDH mutations showed comparable performance to the other models with AUCs of 0.752, 0.814, 0.693, 0.786, and 0.863, respectively corresponding to SQ Fit, Ref SQ, Logan, Ref Logan, and 2TCM (p ≥ 0.10 for the pairwise comparisons with other models). In the univariate analysis, the SQ time-to-peak parameter had the best diagnostic performance (75.7% accuracy) compared to all other individual parameters considered. Conclusions The SQ model circumvents the complexities of the 18F-FDOPA kinetic model and yields similar performance in predicting IDH mutations when compared to the other models, most notably the compartmental model. Our study provides supportive evidence for the routine clinical application of the SQ model for the dynamic analysis of 18F-FDOPA PET images in newly diagnosed gliomas.


INTRODUCTION
As an adjunct to magnetic resonance imaging (MRI), amino acid positron emission tomography (PET) imaging provides additional diagnostic and prognostic information in newly diagnosed gliomas (1). Amino acid PET is particularly helpful for gliomas exhibiting no contrast enhancement on MRI (2)(3)(4), which can, in some cases, conceal more aggressive gliomas (5). The most commonly used radiolabels, i.e., amino acids O-(2-[ 18 F]fluoroethyl)-L-tyrosine ( 18 F-FET) (6), 11 C-methionine ( 11 C-MET) (7), and 6-[ 18 F]fluoro-L-DOPA ( 18 F-FDOPA), were able to noninvasively characterize gliomas at the initial diagnosis (8). Dynamic acquisitions are started at the time of tracer injection, to follow the full kinetic path, and are most commonly performed with 18 F-FET and 18 F-FDOPA. Dynamic 11 C-MET acquisitions are rarely performed because of the limitations of the short (20 min) half-life of this radiolabeled amino acid (9). Dynamic 18 F-FET and 18 F-FDOPA acquisitions have recently shown encouraging predictive performances for the noninvasive characterization of the mutation status of isocitrate dehydrogenase (IDH) in newly diagnosed gliomas (6,8). This is considered to be one of the most important molecular parameters in gliomas according to the 2021 classification of the World Health Organization (WHO) (10). These results were obtained using a semiquantitative model, initially developed for dynamic 18 F-FET PET imaging, that can be easily transposed to the clinic (11). The model relies on the extraction of two parameters directly from the time-activity curve (TAC) without requiring any modeling of the underlying metabolism. These parameters are the time-to-peak (TTP), which is the time from the beginning of the dynamic acquisition to the maximal TAC value, and the late phase slope of the TAC. This simple model is currently recommended for dynamic 18 F-FET PET analyses (12) and has been reported to predict IDH mutation with an accuracy of 72% (6). The same dynamic analysis has also been successfully transposed to 18 F-FDOPA dynamic images, with the time-topeak parameter demonstrating a promising 74% accuracy for predicting the presence of IDH mutations in newly diagnosed gliomas (8).
The simple dynamic 18 F-FET PET modeling approach can be applied because, unlike other radiolabeled amino acids, 18 F-FET is not metabolized by cells (13). Indeed, several studies have shown that 18 F-FET kinetics are best modeled using a one-tissue reversible compartmental model with reliable fit and stable kinetic parameters (14)(15)(16). The 18 F-FDOPA kinetics in glioma, however, involve a much more complex model that is, to date, only partially characterized (17,18) because the initial models were developed in the context of assessing Parkinson's disease (19). The complexities of the 18 F-FDOPA kinetic model arise from its different peripheral and intracellular metabolic pathways. 18 F-FDOPA is metabolized into 18 F-labeled metabolites (METS) and 18 F-labeled L-3,4-dihydroxy-6-fluoro-3-O-methylphenylalanine (OMFD) in the periphery, with OMFD able to bidirectionally cross the blood-brain barrier. In addition, and unlike 18 F-FET, 18 F-FDOPA is metabolized via the dopaminergic pathway, even though it is unclear whether it is metabolized in the same way by tumor cells. A two-tissue compartmental model, only using one input function, has nevertheless been previously proven useful for the 18 F-FDOPA imaging of gliomas (17,18). However, to the best of our knowledge, there are currently no studies that compare the semiquantitative dynamic model to other dynamic models, such as the compartmental 18 F-FDOPA PET imaging model in gliomas (17,18).
Our current study assesses whether an 18 F-FDOPA semiquantitative dynamic analysis indeed captures all the relevant information contained in time-activity curves to predict the presence of IDH mutations compared to the more sophisticated graphical and compartmental models.

Patients
We retrospectively selected newly diagnosed glioma patients for whom 18 F-FDOPA PET had been performed as part of the initial tumor characterization, in the Department of Nuclear Medicine at the CHRU of Nancy, between February 2018 and June 2020. The final selection included: i) patients with a neuropathological diagnosis based on the WHO 2016 classification (20) and with a maximum time interval of 150 days between the 18 F-FDOPA PET and the histological confirmation for diffuse grade II or III gliomas and 60 days for glioblastomas (6,8); ii) patients with available raw data for a posteriori reconstruction; and iii) patients with a visually abnormal 18 F-FDOPA uptake, i.e., by excluding isometabolic and photopenic gliomas (21,22). The institutional ethics committee (Comitéd'Ethique du CHRU de Nancy-FRANCE) approved the evaluation of retrospective patient data on August 26, 2020. The trial was registered at ClinicalTrials.gov (NCT04469244). This research complied with the principles of the Declaration of Helsinki. Informed consent was obtained from all individuals included in the study. 18

F-FDopa PET Acquisition and Image Reconstruction
Patients were instructed to fast for at least 4 h prior to the examination and were pre-medicated with Carbidopa 1 h prior to the examination to increase tracer uptake in the brain (17). Patients were scanned with a digital Vereos PET/computed tomography (CT) camera (Vereos, Philips, Cleveland, OH, USA). Immediately after recording the CT images (100 kV, 80 mAs), a 30-min 3D list-mode PET acquisition was initiated concomitantly to the bolus injection of 2 MBq of 18 F-FDOPA per kilogram of body weight. Static PET images were reconstructed from the list-mode data acquired 10-30 min post-injection (12,23). Dynamic images of all patients were reconstructed using two different temporal sampling protocols depending on the dynamic analyses carried out. For models requiring an input function, we used the recommended temporal sampling protocol from the EANM/SNMMI guidelines, i.e., 12 frames of 5 s, 6 frames of 10 s, 6 frames of 30 s, 5 frames of 60 s, and 4 frames of 300 s (12). This protocol was used because very short frames at the start of the acquisition are needed to capture the large variations of the radiotracer concentration in the blood that occur at the beginning of the acquisition. The temporal sampling protocol consisted of 30 × 60 s frames for models not requiring any input function (24). For models that do not require an input function, the short initial frames only contribute very noisy data points that do not contain any useful information because vascular phase data are not used. Moreover, using a uniform frame duration has the advantage of eliminating one source of noise due to the variations between the time points of the TACs.
Static images were reconstructed using the time-of-flight information and a high-resolution protocol with the Ordered Subset Expectation Maximization (OSEM) 3D algorithm [two iterations, 10 subsets, a deconvolution of the point spread function (PSF), and 256 × 256 × 164 voxels of 1 × 1 × 1 mm 3 ], while a protocol with a lower spatial resolution was used to limit the level of noise in dynamic images, i.e., three iterations, 15 subsets, without PSF, and 128 × 128 × 82 voxels of 2 × 2 × 2 mm 3 (25).
All images were corrected for attenuation using CT, dead time, and random and scattered coincidences during the reconstruction process.

Segmentation
Healthy brain uptake was initially measured from static images using a merged volume of interest (VOI) consisting of a crescentshaped region of interest manually positioned on three consecutive slices of the semi-oval center of the unaffected hemisphere to include both white and gray matter. Tumor VOI was segmented semi-automatically from static images using a threshold of 1.6 healthy brain SUV mean , as previously recommended (24,26,27). The arterial input function VOI was subsequently placed into the internal carotid using initial dynamic frames to identify the early vascular phases (28).
All volumes of interest were segmented using the LifeX software (lifexsoft.org) (29) and were visually inspected by an experienced physician (AV, nuclear physician with more than 10 years of experience) to ensure the quality of the methods applied. Healthy brain was considered as the reference region due to its nonspecific uptake, as required (26).

Extraction of Time-Activity Curves
For dynamic images reconstructed using the protocol with 30 × 60 s frames, each dynamic frame was first registered to the associated CT image in order to correct for any potential patient movement during the acquisition (30). These transformations, representing the evolution of the patient's movements over time, were interpolated to the time frames of the other protocol, for models involving an input function. Indeed, the first frames from images reconstructed with models involving an input function are very short and suffer from noise, which makes the registration very challenging.
Blood and brain TACs were extracted by retrieving the mean standard uptake value (SUV mean ) for each frame in the respective VOIs. Tumor TACs were computed by retrieving the SUV mean for each frame in the volume corresponding to the SUV peak of the tumor VOI on the static image in order to represent the most aggressive part of the tumor (24).

Input Function Pre-Processing
Since no arterial blood sampling was performed in this study, an image-derived input function was used for analyses that required one. TACs representing the evolution of the arterial blood activity were obtained from internal carotid VOIs and were fitted using linear interpolation to the peak followed by a triexponential function after the peak (31). The fitted blood TAC was then corrected for the spill-out effect, the coefficient of which had been estimated as 0.51 (32). In the case of 18 F-FDOPA, the plasma 18 F-FDOPA TAC was obtained after correcting for OMFD and other METS generated in the peripheral tissues (19). The plasma 18 F-FDOPA TAC can be obtained from the blood TAC if the hematocrit level and the proportion of each 18 F-labeled entity in the respective plasma TACs are known (18). These values were retrieved from the literature, specifically from Huang et al. (19), who used a hematocrit of 40%, and from Melega et al. (33), who reported the metabolite proportions for patients pre-medicated with 100 mg of Carbidopa, which is identical to the pre-medication schedule of our patients. To extrapolate the proportions of metabolites at any time, the measured fractions of plasma radioactivity were fitted using the following equations for the plasma fractions of DOPA, OMFD, and METS, respectively: where t is the time in minutes.

Dynamic Models
The pre-processing steps and input data for each dynamic model are presented in Figure 1.

Semiquantitative Models
To overcome noise effects, tumor TACs were first fitted using nonlinear least square optimizations and a specific tumor vascularization function (patent WO/2008/053268, entitled "Method and System for Quantification of Tumoral Vascularization") (8,24,34): where t* is the time interval from injection to the arrival of the radiotracer and a 0 the baseline intensity that is fixed at 0. a 1 and a 2 reflect the maximal value of contrast agent uptake and the time to peak intensity, respectively. p and q are coefficients related to the increase in intensity and the decrease in intensity, respectively. A and B are arbitrary parameters.  Semiquantitative (SQ) model parameters, time-to-peak (TTP) and slope, were respectively computed as the time from the beginning of the dynamic acquisition to the maximum uptake value and the slope of the linear regression of the data between the 10th and the 30th minute (8). Semiquantitative fitting parameters (SQ Fit)-a 1 , a 2 , p, q, A, and B-extracted from the equation of the fit were also assessed. The reference semiquantitative (Ref SQ) model was conducted as an assessment of other studies where the tumor-to-normal brain ratio dynamic values were used to overcome the Carbidopa effect (8,24,34), even though such normalization was not needed in this study. To achieve this, healthy brain TACs were fitted similarly to tumor TACs, and TAC ratios (TAC ratio ), representing the evolution of the ratios between tumor and brain fitted TACs, were computed. The TTP ratio and slope ratio were computed from the TAC ratio similarly to the tumor TACs.

Graphical Models
Among all graphical models available, the Logan graphical model (35) with the computation of the equilibrium volume of distribution is particularly suited to 18 F-FDOPA in gliomas since there is no evidence that 18 F-FDOPA is trapped in tumors (17). The Logan graphical model (Logan) was performed with the slope computed between 15 and 30 min post-injection, as previously suggested (17). The equilibrium volumes of distribution (V ed ) and Int Logan were computed as the slope and the intercept of the graphical model, respectively. We also used the Logan graphical model with a healthy brain reference region (Ref Logan) similarly to the Patlak graphical model for 18 F-FDOPA PET imaging of parkinsonian syndromes (36). Two parameters, with regression between 15 and 30 min, were extracted from the Ref Logan model, namely, the distribution volume ratio (DVR) and the relative residence time (RRT), computed respectively as the slope and the negative intercept.

Compartmental Model
The model used was a simplified two-tissue compartmental model (2TCM) adapted from the original publication of Huang et al. (19) and previously used for the compartmental analysis of 18 F-FDOPA glioma imaging (18). Four rate constants (K 1 , k 2 , k 3 , and k 4 ) and the blood volume fraction (V b ) were estimated by fitting the 2TCM to tumor TACs. The net influx rate constant, Ki, was computed from the previously estimated four rate constants as Ki = (K 1 × k 3 ) / (k 2 + k 3 ).

Statistical Analysis
Categorical variables are expressed as numbers and percentages and continuous variables as medians (first quartile to third quartile) because the variables did not follow a normal distribution. Intergroup comparisons were performed with the chi-squared test for categorical variables and the Mann-Whitney test for continuous variables.
For the overall comparison of the different kinetic models, parameters belonging to the same model were used to construct a multivariate model. This multivariate model was a general linear model with variables selected automatically in a stepwise manner with both forward and backward selections minimizing the Akaike information criterion. Comparisons between the performances of the final models were carried out with the onesided comparison of superiority pairwise Delong tests (37).
The ability of each individually extracted parameter to predict an IDH mutation was assessed using receiver operating characteristic (ROC) curves from which the area under the curve (AUC), sensitivity, specificity, and accuracy were computed. The optimal threshold was determined by selecting the point on the curve closest to (0,1). Spearman's coefficients were calculated to assess the correlations between each extracted parameter of the different models.
P-values were adjusted using the Benjamini-Hochberg correction, and p-values lower than 0.05 were considered significant. Analyses were performed with the R software version 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria).

Dynamic Models
The SQ model with an AUC of 0.733 showed similar performance to the other models with AUCs of 0.752, 0.814, 0.693, 0.786, and 0.863, respectively corresponding to SQ Fit, Ref SQ, Logan, Ref Logan, and 2TCM (p ≥ 0.10 for the pairwise comparisons with the other models). We tested all the possible pairwise dynamic model combinations and found no significant differences (p > 0.05) ( Table 1 and Figure 2). Detailed diagnostic performances of all individual parameters and the correlations between individual parameters are presented in Table 2 and Figure 3, respectively. The TTP parameter from the SQ model showed the highest accuracy (75.7%) among all the parameters examined. The other SQ parameter, i.e., slope, also had a strong diagnostic performance, with an accuracy of 73.0%, which is within the range of performances for the parameters from the Ref SQ and Ref Logan models (respective accuracies of 73.0% and 70.3% for both TTP ratio and Slope ratio and both DVR and RRT). Despite its high overall performance, none of the parameters obtained with the 2TCM were significant for predicting IDH mutations (p > 0.14). SQ parameters were highly correlated with Ref SQ Slope ratio , the a 2 parameter from SQ Fit, and the intercepts from Logan and Ref Logan, namely, Int Logan and RRT, respectively. The 2TCM parameters, however, did not correlate with any of the parameters from the other models. Most parameters extracted from the 2TCM were highly intracorrelated. Representative IDH-wild-type and IDH-mutant gliomas with all the models considered are shown in Figure 4.

DISCUSSION
The current study compared different 18 F-FDOPA kinetic models to assess whether the SQ model, which is currently recommended for amino acid PET in clinical routine practice, is specifically suited to a 18 F-FDOPA PET dynamic analysis (12). This comparison was performed on newly diagnosed gliomas for which the amino acid PET dynamic analysis had previously been shown to successfully predict the status of IDH mutation (6,8,24). The SQ model was found to have a similar performance to the other models, including the 2TCM, as previously reported for the dynamic 18 F-FDOPA PET imaging of gliomas (17,18) ( Table 1 and Figure 2). Our study therefore provides supportive evidence for the application of the SQ model, to predict IDH mutations by 18 F-FDOPA PET imaging in newly diagnosed gliomas, in the clinic.
The SQ model has the advantage of being easily integrated into clinical routine practice with no need for heavy processing,  a reference region, or an input function. Although compartmental analysis, which is considered the current gold standard, requires an input function, input functions are notoriously difficult to obtain in clinical routine practice because they either require blood sampling or, if extracted directly from images, heavy postprocessing and many approximations (see Figure 1), which limits their use in the clinic. There has been renewed interest in dynamic PET acquisitions because of the significant technological advances in the field, such as digital signal (38) and/or the longer field of view of PET devices. These improvements make dynamic acquisitions accessible to clinical routine practice (39). A simple dynamic model is thus preferable to harmonize protocols across different centers and to promote a wider acceptance in the scientific medical community, not only among nuclear physicians but also clinicians requiring access to easily understandable and interpretable data.
Our study confirms the high diagnostic performance of dynamic 18 F-FDOPA PET imaging using the SQ model to predict IDH mutations, with the TTP analysis yielding an accuracy of 75.7%. TACs with shorter TTPs may, in addition, identify more aggressive IDH-wild-type gliomas, which may correspond not only to tumors that express high concentrations of LAT transporters (40) but also those that are characterized by more extensive tracer perfusion, as shown for 18 F-FET (11) and discussed in our previous publications (8,24). Such performances are within the range of previously published studies that reported accuracies of 74% for TTP alone (8) and 75% when associated with other specific radiomics features of 18 F-FDOPA PET imaging (24). It should nevertheless be noted   that these latter two studies used a reference region for the SQ model to account for the population heterogeneity with regard to the Carbidopa pre-medication (8,24,34 (8,24,34). Even though no significant differences were observed between the different models studied, the 2TCM provided better performance (AUC = 0.863, range = 0.734-0.993, p = 0.11, for the pairwise test with the SQ model) with parameters that reflect different kinetic information compared to those from the other models ( Figure 3). However, as explained above, adapting the 2TCM to clinical routine practice presents serious challenges and would make it more difficult to harmonize data sourced from different centers. Among the radiolabeled amino acids recommended for the assessment of gliomas, 18 (6), while an accuracy of 74% was described for the 18 F-FDOPA TTP ratio (8). These results were also confirmed in the Lohmann et al. radiomics study, which reported a model accuracy of 80% (42). The study of Lohmann et al. nevertheless showed that dynamic parameters were particularly pertinent in determining diagnostic performance. Since the underlying kinetic model of 18 F-FDOPA is much more complex than that of 18 F-FET, there was no evidence that a simple model such as the SQ model would suffice. Our current study showed that the SQ model is appropriate for interpreting 18 F-FDOPA kinetics and that it might be considered in future guidelines, as is currently the case for 18 F-FET imaging (12).
The main limitations of our study are related to the assumptions that underpin how the input function was determined. Since no blood sampling was performed for any of our patients, we extrapolated the input function from an imagederived blood TAC that was converted to a 18 F-FDOPA plasma TAC using a previously published method (18). This method is based on hematocrit data and the proportions of individual metabolites previously reported in the literature (19,33). Since this assumption is not adapted to each patient, it may have a negative impact on the results from the model that are dependent on this input function. Moreover, our study used a digital PET device with high count sensitivity and improved dynamic image quality. The values of the different parameters may therefore not be directly comparable to those extracted from noisier dynamic images, captured with older PET devices. This may significantly affect the fitting process that is used for each dynamic model and, thus, affect the stability and the ability to directly extrapolate our results to other PET devices. The number of patients in this study was also limited, even though the diagnostic performances reported were similar to those of other amino acid dynamic PET studies. Although our population comprised a highly selected group of patients, it included representative proportions of the different types of gliomas (43).

CONCLUSION
Despite the complexities of the 18 F-FDOPA kinetic model, our study confirms that a simple semiquantitative analysis, which is currently recommended for amino acid PET imaging in clinical routine practice, captures all the relevant information contained in TACs to predict the presence of IDH mutations when compared to the more sophisticated graphical and compartmental models. Although dynamic parameters play important roles in the interpretation of radiolabeled amino acid PET imaging in gliomas, our current study shows that the application of this easily transposable method can be extended to 18 F-FDOPA PET imaging for the noninvasive characterization of newly diagnosed gliomas.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Comitéd'Ethique du CHRU de Nancy -FRANCE. The patients/participants provided their written informed consent to participate in this study.

AUTHOR CONTRIBUTIONS
TZ, MD, and AV conceived and designed the study and analyzed the data. FR, MB, LT, and LI participated either in the study conception and design or the analysis and interpretation of the data. All authors helped with the drafting of the manuscript or revising it critically for important intellectual content. FR, MB, LT, LI, and AV approved the submitted version. All authors contributed to the article and approved the submitted version.

ACKNOWLEDGMENTS
The authors thank Petra Neufing for critical review of the manuscript.