Effects of carbidopa premedication on 18F-FDOPA PET imaging of brain tumors: a static, dynamic and radiomics analysis

Purpose This study aims to determine the impact of carbidopa premedication on static, dynamic and radiomics parameters of 18 F-FDOPA PET imaging in brain tumors. The study included 54 patients that underwent 18 F-FDOPA PET imaging for newly diagnosed gliomas. Among these, 18 patients received 100 mg of carbidopa. SUV parameters and 105 radiomics features were extracted from the static images. Dynamic data were available for 41 patients. Time to Peak (TTP) values were extracted from dynamic acquisitions. These parameters were obtained from volumes of interest in healthy brain as well as tumors. Simulation of the effects of carbidopa premedication on TTP values were also generated. and of of which (absolute In carbidopa p and TTP (ΔTTP + p = 0.025). all parameters were by carbidopa premedication when using tumor-to-healthy-brain image and TAC ratios. Simulated data that carbidopa leads to tumor TTP values which is corrected by using these ratios.


Introduction
L-3,4-dihydroxy-6-18 F-uoro-phenyl-alanine ( 18 F-FDOPA) is a PET amino-acid radiotracer that has been used to assess gliomas for over 20 years [1]. The PET-RANO group (Response Assessment in Neuro-Oncology) recommends its use at the primary diagnosis, for monitoring disease and therapy, and for diagnosing tumor recurrence [2][3][4]. 18 F-FDOPA has a relatively high speci city for gliomas, conferred by its ability to cross an intact blood-brain barrier and the overexpression of Large Aminoacid Transporters (LATs) in tumors [5,6], making it a useful adjunct to contrast enhanced brain MRI which remains the gold standard for the diagnostic assessment of gliomas.
Carbidopa (L-a-hydrazino-a-methyl-b-(3,4-dihydroxyphenyl)propionic acid) is a peripheral inhibitor of aromatic amino acid decarboxylase. Its use as a premedication therefore results in increased plasma concentrations of 18 F-FDOPA and of its metabolite 18 F-OMFD (3-O-methyl-6-[ 18 F] uoro-L-DOPA) [7]. As uptake in both healthy brain and glioma [9]. The proportion of uptake increase in these two structures and the effects of carbidopa premedication on the radiomics parameters as well as the dynamic analysis nevertheless remain to be determined. To date only one study consisting of two patients premedicated with 200 mg of carbidopa showed an average 50% increase in uptake in the cerebellum, striatum and the tumor, based on acquisitions obtained 15 to 25 min post injection [9].
In contrast to movement disorders PET imaging [10], international guidelines do not recommend administering carbidopa before 18 F-FDOPA PET for brain tumor imaging, solely based on the fact that most of the published studies do not use it. There is currently no evidence-based data to determine whether carbidopa should be used in the clinical setting of brain tumor imaging, particularly in the current era of routine semi-quantitative analyses [3].
The objective of this study is therefore to determine the impact of carbidopa premedication on static, radiomics and dynamic parameters in brain tumor 18 F-FDOPA PET imaging.

Materials And Methods
Population and 18 F-FDOPA PET imaging We retrospectively selected newly diagnosed glioma patients, classi ed according to the WHO 2016 classi cation [11], who underwent an 18 F-FDOPA PET at the CHRU of Nancy between January 2013 and October 2017. Premedication data was available for all patients included in the study and depended on the routine examination protocol performed; patients analyzed from March 2016 to October 2017 were premedicated with carbidopa and patients analyzed from January 2013 to February 2016 were not premedicated prior to PET imaging. The data evaluation process was approved by the local ethics committee (Comité d'Éthique du CHRU de Nancy) on August 26, 2020. The trial was registered at Clinical Trials.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-computed tomography (CT) scans were performed on a Biograph hybrid system involving a six-detector CT for attenuation correction (Biograph 6 True Point, SIEMENS, Erlangen, Germany). All patients were instructed to fast for at least 4h, and patients analyzed from March 2016 to October 2017 received 100 mg of carbidopa 1 hr. prior to their examination. A CT scan was rst recorded for each patient, immediately followed by a 30-min 3D list mode PET recording initiated during the bolus injection of 3 MBq of 18 F-FDOPA per kilogram of body weight. Static PET images were reconstructed from the list mode data acquired 10 to 30 min post-injection, while dynamic PET images consisted of 30 frames of one minute each [12]. Static and dynamic images were reconstructed using the OSEM 2D algorithm (2 iterations, 21 subsets, 4-mm Gaussian post-reconstruction lter, 256 x 256 x 148 voxels of 2.7 x 2.7 x 3.0 mm3). All images were corrected for attenuation using CT, dead time, random and scattered coincidences during the reconstruction process.

Image analyses
Segmentation, The LIFEx software (lifexsoft.org) was used to de ne volumes of interest (VOIs) for tumors and contralateral healthy brain [13].
A patient speci c crescent shape VOI, which encompassed both white and grey matter, was manually drawn on the unaffected hemisphere to measure healthy brain uptake as recommended [14].
In tumors, VOIs were segmented semi-automatically using a threshold of 1.6 healthy brain SUV mean [15]. For tumors with multiple loci, we only considered the site on which the neuropathological diagnosis was performed. All nal VOIs were visually inspected by an experienced physician (A.V.) to ensure that the quality of the methods applied was consistent throughout.
Extraction of parameters, For static images, the SUV mean , SUV max , and SUV peak parameters were extracted from the previously described VOIs for healthy brain and tumors.
For the radiomics analysis, 105 features were extracted from the same brain and tumor VOIs. These included morphological, local intensity, intensity-based statistical, intensity histogram and textural parameters. In accordance with the guidelines and benchmark values of the image standard biomarker initiative [16], 103 parameters were extracted using PyRadiomics and 2 local-intensity parameters, that were not available on PyRadiomics, were extracted with in-house software [12]. These radiomics parameters were extracted as detailed elsewhere [12]. Brie y, isotropic voxel resampling was performed using tricubic spline interpolations, before carrying out an absolute discretization of PET intensities with a xed bin size of 0.1. Parameters were computed from a single matrix after merging all 3D directional matrices.
To potentially correct for any carbidopa premedication effects in our population, all static and radiomics parameters, except morphological features, in tumors were re-extracted after normalizing each static image for the SUV mean of healthy brain VOI, to compute the Tumor-to-normal-Brain Ratio (TBR) parameters.
To take into account any potential patient movement during the dynamic acquisition, each dynamic frame was rst registered to the associated CT image [17]. The SUV mean values for each frame were respectively computed in the brain VOI and in the VOI corresponding to the tumor SUV peak on the static image to extract the brain and tumor time activity curves (TACs). TACs were tted to overcome noise effects [12]. As previously de ned, two dynamic parameters were extracted: time-to-peak (TTP) and slope [18].
As for parameters extracted from static images, a normalized version of the parameters was extracted from a TAC representing the evolution of the ratio between tumor and brain tted TACs to potentially correct for any carbidopa premedication effect [12].
Further analyses to con rm our hypotheses were performed on simulated data (methodology in Supplemental Materials and Supplemental Figure 1). Statistical analysis, Categorical variables are expressed as percentages and continuous variables as medians and interquartile ranges. Intergroup comparisons were performed with the Chi-squared test for categorical variables and the Mann-Whitney test for continuous variables. Mann-Whitney tests were performed to compare carbidopa-naïve and premedicated patients in healthy brain VOIs. Correlations between radiomics features and SUV mean were assessed using the Spearman correlation coe cient. Correction of multiples tests was performed with the Benjamini-Hochberg correction and p < 0.05 was considered signi cant. For tumor VOIs, linear regression analyses were performed to predict the parameters using carbidopa status and histo-molecular diagnosis as covariates, as histo-molecular diagnosis is known to in uence static and dynamic parameters (gliomas were classi ed as IDH-wildtype and IDH-mutant astrocytomas, IDH-mutant and 1p/19q co-deleted oligodendrogliomas, and IDH-wildtype and IDH mutant glioblastomas) [18]. The signi cance of each covariate was tested using a type III analysis of variance. Analyses were performed with the R software version 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria), and Python (Python Software Foundation).
In tumors, carbidopa premedication was an independent predictor of SUV mean (ΔSUV mean = +52%, p<0.001) and TTP (ΔTTP = +24%, p=0.025). Histomolecular diagnosis was predictive of TTP (p = 0.010) and slope (p < 0.001) and a trend was observed for the SUV mean (p = 0.07). Interestingly, all static, dynamic and radiomics parameters were no longer signi cantly modi ed by carbidopa premedication when using tumor-to-healthy-brain image ratios or timeactivity curve ratios (Table 3 and Supplementary Table).
To con rm our hypotheses about the impact of carbidopa premedication on TTP and better understand its effect on slope, simulated TACs were performed with carbidopa premedication on the assumption that carbidopa premedication induces an increase of radiotracer availability through the input function, i.e. the plasma concentration of 18 F-FDOPA, without modifying the rate constants.
Examples of simulated carbidopa premedication leading to increased TTP are displayed in Figure 2.

Discussion
This study shows that carbidopa premedication before 18 F-FDOPA PET imaging of brain tumors is associated with an increase in SUV, SUV-related radiomics and TTP dynamic parameters, of a global same order of magnitude as healthy brain. For neuro-oncological PET indications, the effect of the costly carbidopa premedication is thus limited by the use of TBR images and TAC ratios which are e cient tools for multicentric studies harmonization.
Carbidopa premedication in the present study increases the SUV mean by about 50% in healthy brain tissue as well as and in brain tumors (Table 2 and 3). To the best of our knowledge only one previously published study has to date evaluated the effects of carbidopa in brain tumors, by comparing two patients with stable brain tumors, premedicated with 200mg of carbidopa at baseline and reevaluating these same patients without premedication one year later. Consistently with our study, this previous study also reported a 50% increased uptake in all brain structures, including healthy brain and tumors from acquisitions performed 15 to 25mn post-injection [9].
Our current study found that 81% of all radiomics parameters in healthy brain were signi cantly modi ed by carbidopa premedication, among which 92% correlated with SUV mean (85 parameters, absolute correlation coe cient ≥ 0.4 in 78 parameters) (Figure 1). This underlines the fact that the effect of carbidopa premedication is related to a relatively homogeneous increase in the uptake of the 18 F-FDOPA radiotracer within the VOIs. The impact of SUV values on textural features has already been highlighted by Orlhac et al. [19]. As we are using absolute discretization here, the increase in SUV related to carbidopa premedication induced two concomitant observations: i. a shift of textural matrices to higher bin values (bin shift) mainly responsible for the modi cation of the parameters correlated with the SUV values and, ii. a relative spread of the distribution of SUV values over a larger number of bins (bin spread). Eight percent of radiomics features (7 parameters, Supplemental Figure 2) were signi cantly modi ed by carbidopa premedication but poorly correlated with SUV mean values (absolute correlation coe cients ranging from 0.27 to 0.34). Beyond the impact of a threshold effect, we presume that these parameters are less correlated with SUV mean because they mainly depend on the bin spread effect, for which the impact on radiomics parameters is less directly correlated with SUV mean than the bin shift effect. The bin spread effect on the involved matrices is illustrated in Supplemental Figure 2. Among the 20 remaining parameters, most are not signi cantly modi ed by carbidopa premedication since they are not correlated with SUV mean values (Figure 1).
In our study, carbidopa also induced changes in the dynamic analysis parameters with an increase of TTP in healthy brain and in tumors. In tumors, this effect was independent of the histo-molecular diagnosis, which is known to affect the TTP of the dynamic analysis [18,[20][21][22]. Even if the slope parameter was in uenced by the histo-molecular diagnosis, no signi cant predictive value of carbidopa premedication was observed in tumors while carbidopa induced a signi cant increase of the slope in healthy brain. To con rm our ndings in patients, we performed simulations on the assumption that carbidopa premedication leads to an increase in the plasma availability of 18 F-FDOPA without modifying the rate constants (Supplemental Materials). The two cases based on these simulations presented in Figure 2 con rm that the TTP increase was related to carbidopa premedication, similarly to what was observed in patients. Moreover, an increase of the slope was also observed in the simulated data. It is of concern that the TTP increase due to carbidopa premedication could potentially lead to an underestimation of tumor aggressiveness. In fact, this probability is moderate as regards the small degree of variation observed in the absolute values of TTP and slope in the tumors after simulation of carbidopa premedication (around +4 minutes and +0.5 SUV/h respectively).
Interestingly, when extracted from TBR images and TAC ratios, all static, dynamic and radiomics parameters were no longer signi cantly modi ed by the carbidopa premedication (Table 3 and Supplementary Table). This is an important observation, as TBRs are typically implemented in the routine analysis of neuro-oncology scans. These ratios were used in most articles in the literature for static [23,24], dynamic [18] and radiomics [12] parameters but without having been yet validated as to the effect of the carbidopa premedication.
Since glioma is a rare pathology, the need of multicentric studies to gather enough patients is even more important. These multicentric studies could be conducted only after performing harmonization between the participating centers. Regarding the carbidopa premedication, since its use is very dependent of each center, the harmonization process could come from the use of TBR images and TAC ratios, that have been shown to be insensitive to carbidopa premedication in this study, without modifying the protocols of participating centers. While, the 18 F-FDOPA PET guidelines for imaging of Parkinsonian syndromes [10] recommend carbidopa premedication to increase the systemic and thus central nervous system availability of 18 F-FDOPA. This recommendation is completely different in brain tumors, given that, unlike striatum, brain tumor cells do not metabolize 18 F-FDOPA [9]. This is presumably why recommendations for 18 F-FDOPA PET imaging do not include mandatory carbidopa premedication prior to brain tumor scans. This recommendation is however not founded on any evidence-based data from the literature and our study therefore provides additional information which supports the current recommendation.
Our retrospective study was performed on a heterogeneous population of patients, and therefore limits the ability to directly compare the effects of carbidopa premedication in tumors. Our hypotheses on the effects of carbidopa premedication were nevertheless con rmed by using simulated data. In our department, the patient carbidopa status was predicated by the recruitment period, this may have introduced an inclusion bias. However, no signi cant differences were observed between the two groups of patients (Table 1).

Conclusion
Our current study documents the effects of carbidopa premedication on the 18 F-FDOPA PET imaging of brain tumors. Carbidopa premedication leads to an increase in the availability of 18 F-FDOPA in plasma without modifying the rate constants. As carbidopa premedication increases the SUV, SUV-dependent radiomics and TTP dynamic parameters by a global same order of magnitude in healthy brain as in tumor, these effects are compensated for after taking into account the tumor-to-healthy-brain ratios in static images or in time-activity curves, which is an important point for multicentric studies harmonization.   Effect of carbidopa premedication on healthy brain radiomics parameters.