We conducted a retrospective study of patients with histologically proven diffuse glioma who underwent ¹¹C-MET PET/CT between February 2012 and November 2017 at a single center. Inclusion criteria: (1) all patients were confirmed to have glioma histological diagnosis and IDH1 R132H immunohistochemistrical staining results. (2) PET images for every patient were of good quality with no obvious artifacts. Exclusion criteria: (i) patients who received treatment by radiotherapy, chemotherapy, or chemoradiotherapy before PET imaging. (ii) glioma patients with no precise histological grading or IDH1R132H staining results. (iii) poor image quality with artifacts affecting the semi-quantitative analysis. (iv) hypo- or iso-metabolism of ¹¹C-MET compared to the background which is not applicable for threshold-based tumor volume delineation procedures. Moreover, the interval between PET imaging and subsequent tumor resection or biopsy was no more than 100 days for grade II or III gliomas and no more than 30 days for grade IV glioblastomas. A total of 110 cases were eligible for inclusion.

¹¹C-MET was synthesized by the GE Healthcare-Tracerlab-FXc ¹¹C radiolabelling module semi-automatically. ¹¹C-CO₂ was produced by SIEMENS RDS III cyclotron, and homo-cysteine was used as the precursor. The radiochemical purity of the obtained sterile product was higher than 95%. All patients had fasted for at least 4 h before imaging. At 10–15 min after an intravenous bonus injection of ¹¹C-MET (370–550 MBq), a static PET scan was subsequently collected for 20 min with a Siemens Biograph 64 HD PET/CT (Siemens, Erlangen, Germany) in 3-dimensional (3D) mode. PET images were reconstructed using the filter back projection (FBP) with Gaussian filter (FWHM 3.5 mm) and a 256^*256 matrix, providing 64 contiguous transaxial slices of 5 mm-thick spacing. Attenuation correction was performed using a low-dose CT (150 mAs, 120 kV, Acq. 64^*0.6 mm) before the emission scan.

All PET/CT images were analyzed using a dedicated workstation (Siemens.syngo.via). Semi-quantitative analysis of tumor metabolic activity was obtained using SUV normalized to body weight. All parameters were assessed in 3-dimensional volumes. Mean standardized uptake values (SUVmean) of the normal contralateral frontal cortex were calculated as references. A predefined threshold method at 1.3-times of the corresponding reference SUVmean (23, 24) was applied. The brain MRI of patients were reviewed initially to locate the possible tumor region. A VOI isocontour of the tumor region were applied, semi-quantitative PET imaging analysis were carried out after the lesion delineation procedures. The above-mentioned procedures were carried out by two experienced nuclear medicine physicians separately to double confirm the correct inclusion and reproducible parameters measurements of the glioma lesion. For those multifocal glioma patients in our group, the specific surgical resected or biopsied lesion for pathological examination were included in our research in order to avoid bias. Each VOI generated a maximum of SUV (SUVmax), a mean SUV (SUVmean), a peak SUV (SUVpeak), a standard deviation of SUVmean (SUV_SD), a metabolic tumor volume (MTV) and a total lesion methionine uptake (TLMU). The total lesion methionine uptake (TLMU) was defined as the MTV multiplied by the SUVmean within the tumor boundary. SUVpeak was the highest mean SUV from a fixed 1-cm³ spherical volume centered over the highest metabolic part of the tumor. The lesion SUV/normal contralateral cortical SUVmean was defined as the tumor-to-normal brain tissue ratio (TNR) of ¹¹C-MET uptake.

Physicians would examine the interested glioma lesions to decide whether brain midline structure were involved or not, mainly taking MET PET images for reference. The brain midline structures included corpus calloum, cingulate gyrus, thalamus, third ventricle and brain stem. As illustrated above, two physicians performed VOI delineation for each included patient to confirm the brain midline structure involvement status and to obtain two sets of MET metric features. In order to build a relatively stable integrated ¹¹C-MET PET/CT metrics-based model, we evaluated the inter-observer agreement indices for those obtained results.

Histological specimens were obtained by surgery or stereotactic brain biopsy. H&E staining and immunohistochemical analysis were performed by an experienced neuropathologist according to the current WHO guidelines. IDH status of the surgical samples was identified with an antibody to the IDH1 (R132H) mutation by immunohistochemical staining.

Participant's age, gender and brain midline structure involvement were used as potential predictors, together with those MET-PET metrics, to perform the univariate logistic regression analysis for developing a prediction model of IDH mutation. Those MET-PET metrics share a deep homology, so we first evaluated their correlations to avoid overfitting in the nomogram model building. After that, the MET PET metrics-based nomogram was then designed based on a multivariable logistic analysis results in the whole group with the aim of providing the clinician with a quantitative tool used in the prediction of IDH mutation status. The nomogram model validation involved the quantitative assessment of the nomogram's accuracy in IDH mutation prediction by use of Harrell's concordance index (C-statistic) and calibration curve. The corrected C-index, which is used to quantify the level of concordance between predicted probabilities and actual chance, was measured to predict the accuracy (discrimination) of the nomogram (20, 25). A relatively corrected C-index could be calculated after bootstrap analyses using 1,000 resamples. The calibration curve was used to estimate how closely the modeled nomogram estimated the risk relative to the actual risk of IDH status (mutant or wildtype), accompanied by the Hosmer-Lemeshow test (26).

All continuous variables are expressed as mean ± standard deviation or median and range. Categorical variables are expressed as percentages. For continuous variables, an independent sample t-test was used to compare the two groups, while the chi-square test was applied to calculate P-values for categorical variables. Inter-observer agreements on ¹¹C-MET PET metrics and dichotomized location results were assessed with interclass correlation coefficients (ICC) and Cohen's kappa coefficient analysis, respectively, defined as poor (<0.2), fair (0.21–0.4), moderate (0.41–0.6), good (0.61–0.8), and very good (0.8–1.0). All PET activity measuring indices were compared with each other using scatter plots and Pearson correlation coefficients. Receiver operating characteristic (ROC) analysis was performed to calculate the area under the ROC curve (AUC) for each PET semi-quantitative parameters. The Delong test was used in the comparison of ROC curves. The AUC of ROC curves analysis and the Delong test were performed by using MedCalc for windows (version11.3.3.0, MedCalc software, Mariakierke, Belgium). Univariate and multivariate logistic regression models were used to identify the predictive factors for an IDH mutation. A nomogram was formulated based on the results of multivariate logistic regression analysis and by using the rms package of R, version 3.6.1 (http://www.r-project.org/). The predictive performance of the nomogram was measured by concordance index (C-Statistics) and calibration with 1000 bootstrap samples to decrease the overfit bias. All other statistical analysis was performed using the Prism Software version 8.0 (GraphPad, San Diego. CA). In all analyses, P < 0.05 was considered to indicate statistical significance.

The demographic data of the patients included in this study are listed in Table 1. Of the 110 patients who were evaluated retrospectively, 67 (59.32%) were male, and 43 (40.68%) were female, with a mean age of 45.5 years (range 10–71). The majority of patients (80/110) underwent tumor resection. The post-surgical histological examination demonstrated 59 grade II diffuse glioma, 32 grade III anaplastic tumors, and 19 grade IV glioblastomas, among which 61 patients confirmed IDH-wildtype while 49 patients were IDH-mutant. Patients with IDH-wildtype were more likely to present with lesions involving the midline structures, while there was no significant difference in gender distribution between these two groups (detailed in Table 2).

The dichotomized location results of the interested tumor lesion yielded very similar values for both readers, and accordingly, the inter-observer kappa was satisfactory (κ = 1.0, p < 0.0001). The ICC also showed perfect agreement for the five ¹¹C-MET PET volume-based metrics (ICC > 0.95, p < 0.0001). Therefore, only the results of reader one were considered for further analysis. The absolute values for all ¹¹C-MET PET metrics based on IDH-genotype were given in Table 2.

All pairs of volume-based ¹¹C-MET PET metrics showed a linear association, which was quantified by Pearson correlation coefficients. There were strong correlations between paired TNRs, i.e., TNRmax, TNRmean, and TNRpeak, and SUV_SD with r values ranging from 0.843 to 0.986 (p < 0.0001). The volume-related features, including MTV and TLMU, also correlated strongly with each other (r = 0.927, p < 0.0001). Intratumoral heterogeneity feature SUV_SD and TNRs demonstrated fair or moderate associations with MTV (r = 0.228–0.370, p < 0.05) and TLMU (r = 0.342–0.430, p < 0.0001) (detailed in Table 3 and Supplementary Figures 1A,B).

As shown in Table 2, IDH-wildtype patients had significantly higher TNRmax, TNRmean, and TNRpeak values. Lower SUV_SD values were shown in IDH-mutant patients. Lower MTV and TLMU values were also observed in the IDH-mutant group, albeit not significantly so.

In the ROC analysis, the highest AUC of 0.731 (95%CI: 0.638–0.811) was reached by the SUV_SD value, with the best cut-off value at 0.29, a specificity of 60.66% and a sensitivity of 77.55%, followed by the TNRmax value with an AUC of 0.678, the best cut-off value at 2.99, a specificity of 59.02% and a sensitivity of 75.51%. Their optimal cutoff, AUC, sensitivity, specificity values, etc., for the abovementioned ¹¹C-MET MET PET metrics were listed in Table 4. Their AUC curves were displayed in Supplementary Figure 2.

Further pairwise comparisons of ROC curves confirmed that the AUC of SUV_SD differed significantly from any other MET PET metrics (p < 0.05, details in Supplementary Table 1).

In univariate analysis, the MET PET metrics including SUV_SD, TNRmax, TNRpeak and TNRmean, except for MTV and TLMU, were significantly associated with IDH mutation (p < 0.05 for all the variables). Considering their collinearity and the AUC curve comparison results for the MET PET metrics, SUV_SD was selected as the only MET PET feature for further multivariate logistic regression analysis. In multivariate logistic regression analysis, the three factors, i.e., participant's age, the involvement of midline structure, and SUV_SD, were found to be significant independent predictors. We demonstrated that SUV_SD (>0.29 vs. ≤ 0.29 OR: 0.053, p = 0.010), brain midline structure involvement (no vs. yes OR: 26.52, p = 0.000) and age (≤ 45 vs. >45 years OR: 3.23, p = 0.023), were associated with a higher incidence of IDH mutation (shown Table 5).

Age does not correlate with SUV_SD (r = 0.0370, p = 0.7010). Lesions involving brain midline structure (73 cases) showed higher SUV_SD (0.3147 ± 0.2501 vs. 0.5038 ± 0.3017, p = 0.007) compared to lesions without brain midline involvement (37 cases), but no age predominance (45.67 ± 12.10 vs. 43.92 ± 16.19, p = 0.5245) was observed for brain midline involvement (shown in Supplementary Figure 3).

After that, a nomogram was constructed on the basis of the multivariate logistic regression (for details, see Figure 1A). The nomogram (model 1) showed good discrimination efficacy, with a C-statistics of 0.866 (95%CI: 0.796–0.937). The calibration curve of the nomogram also indicated good agreement between predicted probability and actual occurrence in the whole cohort (Figure 1B). The Hosmer-Lemeshow test indicated no significant difference (p > 0.05), suggesting that there was no departure from a perfect fit. Meanwhile, we built model 2 only using MET PET information, i.e., brain midline structure involvement and SUV_SD derived from lesion VOI, with a C-statistics of 0.843 (95%CI: 0.766–0.920), suggesting that age information dichotomized by 45 years old do enhance the predictive ability for IDH genotype. Figure 2 illustrates a comparison of two representative grade II glioma cases with similar images by visual analysis. Our ¹¹C-MET PET/CT-based nomogram could effectively distinguish between IDH-mutant and IDH-wildtype gliomas.