Integrated CT Radiomics Features Could Enhance the Efficacy of 18F-FET PET for Non-Invasive Isocitrate Dehydrogenase Genotype Prediction in Adult Untreated Gliomas: A Retrospective Cohort Study

Purpose We aimed to investigate the predictive models based on O-[2-(18F)fluoroethyl]-l-tyrosine positron emission tomography/computed tomography (18F-FET PET/CT) radiomics features for the isocitrate dehydrogenase (IDH) genotype identification in adult gliomas. Methods Fifty-eight consecutive pathologically confirmed adult glioma patients with pretreatment 18F-FET PET/CT were retrospectively enrolled. One hundred and five radiomics features were extracted for analysis in each modality. Three independent radiomics models (PET-Rad Model, CT-Rad Model and PET/CT-Rad Model) predicting IDH mutation status were generated using the least absolute shrinkage and selection operator (LASSO) regression analysis based on machine learning algorithms. All-subsets regression and cross validation were applied for the filter and calibration of the predictive radiomics models. Besides, semi-quantitative parameters including maximum, peak and mean tumor to background ratio (TBRmax, TBRpeak, TBRmean), standard deviation of glioma lesion standardized uptake value (SUVSD), metabolic tumor volume (MTV) and total lesion tracer uptake (TLU) were obtained and filtered for the simple model construction with clinical feature of brain midline involvement status. The area under the receiver operating characteristic curve (AUC) was applied for the evaluation of the predictive models. Results The AUC of the simple predictive model consists of semi-quantitative parameter SUVSD and dichotomized brain midline involvement status was 0.786 (95% CI 0.659-0.883). The AUC of PET-Rad Model building with three 18F-FET PET radiomics parameters was 0.812 (95% CI 0.688-0.902). The AUC of CT-Rad Model building with three co-registered CT radiomics parameters was 0.883 (95% CI 0.771-0.952). While the AUC of the combined 18F-FET PET/CT-Rad Model building with three CT and one PET radiomics features was 0.912 (95% CI 0.808-0.970). DeLong test results indicated the PET/CT-Rad Model outperformed the PET-Rad Model (p = 0.048) and simple predictive model (p = 0.034). Further combination of the PET/CT-Rad Model with the clinical feature of dichotomized tumor location status could slightly enhance the AUC to 0.917 (95% CI 0.814-0.973). Conclusion The predictive model combining 18F-FET PET and integrated CT radiomics features could significantly enhance and well balance the non-invasive IDH genotype prediction in untreated gliomas, which is important in clinical decision making for personalized treatment.

Methods: Fifty-eight consecutive pathologically confirmed adult glioma patients with pretreatment 18 F-FET PET/CT were retrospectively enrolled. One hundred and five radiomics features were extracted for analysis in each modality. Three independent radiomics models (PET-Rad Model, CT-Rad Model and PET/CT-Rad Model) predicting IDH mutation status were generated using the least absolute shrinkage and selection operator (LASSO) regression analysis based on machine learning algorithms. All-subsets regression and cross validation were applied for the filter and calibration of the predictive radiomics models. Besides, semi-quantitative parameters including maximum, peak and mean tumor to background ratio (TBRmax, TBRpeak, TBRmean), standard deviation of glioma lesion standardized uptake value (SUV SD ), metabolic tumor volume (MTV) and total lesion tracer uptake (TLU) were obtained and filtered for the simple model construction with clinical feature of brain midline involvement status. The area under the receiver operating characteristic curve (AUC) was applied for the evaluation of the predictive models.
Results: The AUC of the simple predictive model consists of semi-quantitative parameter SUV SD and dichotomized brain midline involvement status was 0.786 (95% CI 0.659-0.883). The AUC of PET-Rad Model building with three 18 F-FET PET radiomics parameters was 0.812 (95% CI 0.688-0.902). The AUC of CT-Rad Model building with three co-registered CT radiomics parameters was 0.883 (95% CI 0.771-0.952). While the INTRODUCTION Glioma is the most frequently-occurred primary malignant tumor in the brain. The 2016 WHO classification of the central nervous system tumor introduced a new integrated classification mode of glioma (1). Isocitrate dehydrogenase (IDH) mutation is considered to be an early event in the occurrence and development of glioma (2). IDH mutations were identified with a high percentage in low-grade gliomas and secondary glioblastoma multiforme (GBM) but with a much lower percentage in primary GBM (3).
IDH mutation altered the metabolism and microstructure of gliomas, thus affecting the biological characteristics and prognosis (4). Evidence supports that astrocytoma with IDH wildtype and other GBM-like molecular features has similar behavior as WHO grade IV glioma (5). At present, the identification of IDH mutation status was mainly based on surgical resection or biopsy specimens. However, additional surgical risks related to patient's comorbidities, advanced age, deep-seated tumors are the barriers to accurately detect the IDH mutation status. Therefore, reliable methods which could noninvasively detect IDH mutation status are needed.
O-[2-( 18 F)fluoroethyl]-l-tyrosine ( 18 F-FET) is an amino acid tracer of positron emission tomography (PET) imaging which could generally well outline the lesion out of the brain background, and could be applied as an effective complementary diagnostic modality besides magnetic resonance imaging (MRI) (6). The characteristic of 18 F-FET PET imaging has led to wide clinical application (7)(8)(9), and gained the recommendation in all the phases of glioma management (10,11).
Although these above-mentioned modalities could benefit the imaging diagnosis and evaluation, the intratumoral heterogeneity of glioma still challenge the precise imaging diagnosis and prediction. Our previous experience indicated conventional semi-quantitative parameters of static 18 F-FET PET imaging could contribute to glioma non-invasive prediction in some degree (12). However, multiple dimensional and more detailed description of the imaging heterogeneity of glioma are crucial in improving the predictive efficacy. Radiomics technology could further extract imaging features among voxels for quantitative analysis based on the imaging heterogeneity, which could provide valuable information about tumor spatial and microenvironment (13). MRI radiomics has shown promising results in glioma imaging research (14). Limited research has shown the potential of amino acid PET imaging based radiomics analysis in glioma noninvasive grading, prognostication, pseudoprogression differentiation and molecular markers prediction such as IDH mutation (15)(16)(17)(18)(19)(20)(21). Computed tomography (CT) radiomics features, which has shown encouraging findings in lung, breast, gastrointestinal tract, pancreas and renal neoplasms, et al. (22)(23)(24)(25)(26), have rarely been applied for glioma radiomics analysis. Whether radiomics features of the amino acid tracer PET and co-registered CT images could contribute to glioma noninvasive prediction is of interest.
The purpose of this research was to investigate the efficacy of predictive radiomics models originating from PET modality, CT modality and combined PET/CT modalities generated with machine learning algorithms based on static 18 F-FET PET/CT imaging in noninvasive IDH prediction, and to compare with the conventional semi-quantitative parameter analysis.

Patient Selection
We retrospectively analyzed the patients who received consecutive 18 F-FET PET/CT brain imaging at the PET Center of Huashan Hospital, Fudan University from November 2017 to February 2019. A total of 58 untreated adult glioma patients were identified based on inclusion and exclusion criteria Supplementary S1. Baseline clinical and demographic characteristics including age, gender, tumor location, WHO grade, and IDH mutation status, were derived from medical file system of Huashan Hospital, Fudan University. 30 MBq), a static PET scan was collected for 20 minutes with a Siemens Biograph 64 HD PET/CT (Siemens, Erlangen, Germany) in 3-dimensional (3D) mode for 50 patients in this cohort. 8 patients underwent dynamic 18 F-FET PET scanning lasted longer than 40 minutes after the tracer injection; 20-40 minutes images were reconstructed for diagnosis and analysis. Attenuation correction was performed using a low-dose CT (tube current 150mAs, voltage 120kV, Acq. 64*0.6mm, convolution kernel H30s, slice thickness 5mm, interslice gap 1.5mm) before the emission scan. After acquisition, the PET images were reconstructed by filtered back projection algorithm with Gaussian filter and full-width-at-half-maximum at the center of the field of view of 3.5mm). The reconstructed brain PET image matrix size was 168*168 with voxel size of 2.04*2.04mm. And the reconstructed brain CT image matrix size was 512*512 with voxel size of 0.59*0.59 mm.

Histological Evaluation and IDH Genotype Analysis
Histological specimens were obtained by surgical resection or stereotactic brain biopsy. H&E staining and immunohistochemical analysis were performed by experienced neuropathologists according to the 2016 WHO classification. IDH genotype status was identified with an antibody to the IDH1 (R132H) mutation by immunohistochemical staining. For 31 out of the 58 patients, IDH mutational status from the immunohistochemical staining was further confirmed by sequencing.
Conventional Threshold-Based 18 F-FET PET Imaging Analysis PET/CT imaging was first analyzed using a dedicated workstation (Siemens Syngo.via) to obtain semi-quantitative parameters in 3D volumes. Background mean standardized uptake value (SUV) was measured initially in a crescent-shape area, including both gray and white matter on the contralateral hemisphere as mentioned (27). The brain structural MRI of patients were reviewed initially to locate the tumor region. A predefined threshold of 1.6-times of the background mean SUV was applied for tumor volume of interest (VOI) delineation to derive lesion maximal SUV (SUVmax), peak SUV (SUVpeak), mean SUV (SUVmean), the standard deviation of lesion standardized uptake value (SUV SD ), metabolic tumor volume (MTV) and total lesion tracer uptake (TLU) (28). Maximal tumor to background ratio (TBRmax), peak tumor to background ratio (TBRpeak), and mean tumor to background ratio (TBRmean) were calculated by the division of tumor VOI SUVmax, SUVpeak and SUVmean with background SUVmean. TLU was defined as the MTV multiplied by the tumor lesion SUVmean. For those multifocal glioma patients in our group, the specific surgical resected or biopsied lesion for pathological examination were included in our research to avoid bias.
Two experienced nuclear medicine physicians (WY Zhou: 5 years' experience, T Hua: 10 years' experience) performed lesion delineation and obtained two sets of semi-quantitative parameters separately. Brain midline involvement status were judged respectively by the two physicians after lesion delineation based on 18 F-FET PET imaging. The brain midline structures included corpus callosum, cingulate gyrus, bilateral thalamus, third ventricle, brain stem and cerebellar vermis. The interobserver agreement indices were evaluated for those obtained semi-quantitative parameters and glioma location status.

Semi-Quantitative Parameters Filter and Simple Predictive Model Building
Univariate logistic regression was applied to filter the conventional semi-quantitative parameters for further model construction. The best-performance semi-quantitative parameter was combined with dichotomous brain midline involvement status for simple predictive model building.

Segmentation and Radiomics Features Extraction
The original 18 F-FET PET/CT data in digital imaging and communications in medicine (DICOM) format were converted into NIFTI format using SPM12 (Welcome Trust Centre for Neuroimaging, London, UK; http://www.fil.ion.ucl.ac.uk/ spm).Then the tumor lesions were manually delineated on 18 F-FET PET images using the ITK-SNAP software (http://www. itksnap.org/pmwiki/pmwiki.php) with the reference to patients' structural MRI by the above-mentioned two physicians separately. If the divergence of a patient's segmentation by the two physicians was less than 5%, the final segmentation of this specific patient was determined as the overlapping region of the two separate VOIs, and if the divergence was more than 5%, a senior nuclear medicine physician with over 25 years' experience (YH Guan) made the final decision. After that the masks of those final tumor VOIs delineation on the 18 F-FET PET images were applied to the co-registered CT images for the CT lesion delineation obtaining.
The registered 18 F-FET PET/CT images and the VOIs were used to extract features. Prior to feature extraction, image standardization was implemented as follows: sitkBSpline interpolation resampling techniques were used to standardize the image scale in the slice, resulting in a pixel size of 2.5mm×2.5mm×2.5mm for PET and 1.5mm×1.5mm×1.5mm for CT, respectively. Two sets (105*2) of radiomics features from VOI lesions of the original PET and CT images were extracted with no further transformation or filtering using PyRadiomics (https://github.com/Radiomics/pyradiomics) (29), including 13 shape and size features, 18 first-order features and 74 texture features. The shape and size features included descriptors of the 3D size and shape of the VOI, which are independent from the gray level intensity distribution in the VOI. First-order features included the maximum, mean, and average absolute deviation of the gray-level intensity values in the VOIs. The texture features consisted of second-order features and are used to express heterogeneity in the tumor, from common matrices such as gray-level co-occurrence matrix (GLCM), gray-level size zone matrix (GLSZM), gray-level run length matrix (GLRLM), gray level dependence matrix (GLDM) and neighboring gray tone difference matrix (NGTDM). The feature extraction algorithms were standardized by referring to the Image Biomarker Standardization Initiative (IBSI) (30). Considering the relative low-resolution nature of PET images, radiomics feature extraction was conducted only on the original PET images with no further transformation or filtering. The details of extracted radiomics features were presented in Supplementary S2.

Radiomics Features Filter and Predictive Model Establishment
Owing to the relatively small number of glioma patients in our cohort, less than five radiomics features were filtered for predictive model construction to minimize overfitting or selection bias in our radiomics models (13,31). Least absolute shrinkage and selection operator (LASSO) and all-subsets regression were performed to filter the most significant radiomics features combinations for IDH genotyping. The LASSO algorithm adds a L1 regularization term to a least square algorithm to avoid overfitting, which is suitable for the regression of high dimensional data (32). Logistic regression analysis was then utilized to integrate the filtered features after further all-subsets regression selection. Calculated-score of three predictive models originating from radiomics features of PET, CT and their combined modalities were obtained by the linear fusion of the selected discriminating radiomics features weighted by their respective coefficients.

Prediction Performance, Model Validation, and Calibration
The performance of the predictive models was evaluated by the receiver-operating characteristic (ROC) analysis and compared by the DeLong test. Additionally, the area under the ROC curve (AUC) with 95% confidence interval (CI), sensitivity, specificity, accuracy, positive predictive value and negative predictive value were calculated for each predictive model.
The four models for IDH genotype prediction were evaluated with k (k=3,5,10) fold cross-validation. The clinical application value of the predictive models was determined and compared through the decision curve analysis (DCA) by quantifying the net benefit to the patient under different threshold probabilities in the cohort (33).

Statistical Analyses
Descriptive statistics were presented as mean ± standard deviation or median and range. Categorical variables were expressed as percentages. An independent sample t-test was used to compare two groups, while the chi-square test was applied to calculate p values for categorical variables. The Mann-Whitney rank sum test was used when variables were not normally distributed. Interobserver agreements on 18 F-FET PET metrics and dichotomized location results were assessed with interclass correlation coefficients (ICC) and Cohen's kappa coefficient analysis respectively, defined as poor (less than 0.2), fair (0.21-0.4), moderate (0.41-0.6), good (0.61-0.8), and very good (0.8-1.0). Univariate and multivariate logistic regressions were used to identify the predictive factors for IDH mutation. The package "glmnet" was used to perform LASSO binary logistic regression analysis, and the "leaps" package to achieve all-subset regression algorithms, and the "bootstrap" package, to perform cross-validations by the R software, version 4.0.4 (http://www.r-project.org/). The ROC analyses and the DeLong test were performed by MedCalc for windows (version11.3.3.0, MedCalc software, Mariakierke, Belgium). All other statistical analysis was performed with Prism Software version 8.0 (GraphPad, San Diego. CA). In all analyses, p < 0.05 was considered to indicate statistical significance.

Clinical Characteristics of Patients
Demographic and clinical data of the 58 enrolled glioma patients were summarized in Table 1. The patient recruit process is presented in Figure 1. The time interval between PET imaging and subsequent tumor resection or biopsy was no more than 90 days for grade II or III patients and not exceeding 30 days for grade IV glioblastomas. A detailed chronologically enrolled patients' clinical data was provided in Supplementary S3.

Inter-Reader Agreement for the Semi-Quantitative Parameters
The dichotomized brain midline involvement status of the interested tumor lesion yielded unified results for two separate readers, and the inter-observer kappa was satisfactory (k=1.0, p<0.0001). The ICC also showed very good agreement for the 18 F-FET PET volume-based semi-quantitative parameters (ICC>0.95, p<0.0001). Therefore, the results of reader one (WY Zhou) was applied for further analysis. The absolute values for all 18 F-FET PET metrics based on different IDHgenotype were also provided in Table 1.

Radiomics Models Performance and Calculated-Score Analyses
Three radiomics predictive models were developed by machinelearning based algorithms. Three PET radiomics features were selected for the building of the PET-Rad Model. Three CT radiomics features were selected for the building of the CT-Rad Model. Three CT radiomics features and one PET radiomics feature were filtered for the building of the PET/CT-Rad Model (shown in Figure 2). The relative importance analysis of the PET/CT-Rad Model indicated one of the three CT radiomics features (CT_glcm_InverseVariance) had the most important weight with 30.25% and the PET radiomics feature (PET_glcm_JointEnergy) had the third weight out of the four features (26.41%) in the PET/CT-Rad Model (shown in Figure 3). The calculated-scores of the three radiomics models for each patient were obtained using related formulas and shown as follows: Results showed IDH-mutants had higher calculated-scores than IDH-wildtypes in the PET-Rad, CT-Rad and combined PET/CT-Rad models (1.1262 ± 1.8048 vs. -2.6648 ± 2.1637, p < 0.001; 1.0008 ± .2.0396 vs. -2.5214 ± 2.5610, p < 0.001; 0.1536 ± 1.2660vs. -1.6186 ± 1.6268, p < 0.001; respectively) ( Figures 4A, B).  Table 2). DeLong test results indicated that PET/CT-Rad Model outperformed the PET-Rad Model and the simple model in IDH genotype prediction (p = 0.048 and p = 0.034, respectively) ( Figure 5). No statistically significant difference could be observed among the PET-Rad Model, CT-Rad Model and simple model. The decision curve analysis results revealed that the PET/CT-Rad Model was the most satisfactory predictive model in differentiating IDH mutation status in these four models (shown in Figure 6). Representative glioma patients of different IDH genotypes were provided in Figure 7.

Predictive Efficacy of the Combination of PET/CT-Rad Model With Dichotomized Brain Midline Involvement Status
We further explored the combination of the dichotomized tumor location status and PET/CT-Rad Model for IDH mutation status prediction, and results indicated the predictive efficacy slightly enhanced in terms of AUC and accuracy (from 0.912 to 0.917 and 86.2% to 87.9%, respectively), although the enhancement was statistically insignificant.

DISCUSSION
Intratumoral heterogeneity is a significant feature of malignant tumors. Radiomics techniques have provided the quantification possibility to further investigate the imaging heterogeneity in internal space geometry, texture, gray distribution and shape (34). In our research we investigated the feasibility for PET, CT and PET/CT radiomics models for IDH mutation status prediction using LASSO regression based on static 18 F-FET resolution of PET imaging, the efficacy of PET radiomics features needs to be enhanced and stabilized with the combination of structural radiomics features. One major obstacle for CT radiomics application in glioma was the VOI delineation process, which amino acid tracer PET imaging could well complement. With the duplication of the 18 F-FET PET lesion VOI delineation to the co-registered CT imaging, we could evaluate the efficacy of CT radiomics features. The CT radiomics predictive model comprises both shape and textural features from the glcm, and gldm matrixes. An intuitive interpretation of the signature might be that relatively homogeneous tumors with less aggressive growth pattern is in concordance with relatively favorable IDH genotype. Results in our research showed that after the combination of the coregistered CT imaging radiomics features, the predictive accuracy, sensitivity, specificity and AUC of the 18 F-FET PET/ CT radiomics model could be enhanced and stabilized in noninvasive IDH genotype prediction. Different weighted images in MRI could benefit significantly in brain imaging. A recent study by Bangalore et al. evaluated the usefulness of T2-weighted MRI-based deep learning method for the determination of IDH mutation status (35). The results are inspiring since T2-weighted MRI is widely available and routinely performed in the assessment of gliomas. Owing to the incomplete structural MRI DICOM data of our patient cohort, we investigated the radiomics features of integrated CT images in our research. Results showed CT radiomics features have promising potentials in the IDH genotyping of glioma patients.
SUV SD is a semi-quantitative parameter based on static 18 F-FET PET images which could reflect the metabolic heterogeneity in some degree. We have confirmed the efficacy of this semiquantitative parameter in 11 C-MET and 18 F-FET PET imaging in glioma patients for IDH mutation status prediction (12,36). However, cautions should be noticed in interpreting this specific semi-quantitative parameter because the reproducible lesion delineation procedures are the prerequisite for the application of SUV SD. In this patient cohort, the most frequently used semiquantitative parameters including TBRmax and TBRmean were found to have borderline significance in differentiating IDH mutation status. With the enrollment number increases, further investigation could be applied to observe the efficacy of these classical semi-quantitative parameters.
Although the radiomics signatures and simple model have shown efficacy in IDH mutation status prediction, we must know that all these imaging analyses were based on robust and repetitive lesion delineation procedures. Experienced manual segmentation of glioma with the reference of structural MRI before radiomics features extraction, 1.6 times of contralateral brain background SUVmean with correct and reproducible tumor delineation for the acquirement of semi-quantitative parameters are crucial for this study. Additionally, the methodologies we used were a 3D classification approach without the concern of data leakage or the risk of introducing bias, compared to those 2D slice-wise-built models (37). With the application of radiomics techniques, abundant imaging features could be extracted for analysis. In this process caution should be made to avoid overfitting during the radiomics analysis, which could lead to bias in result interpretations. With the consideration of the numbers of our patient cohort and radiomics features, radiomics models in our research each consisting of less than five features would be acceptable to control possible overfitting. Cross-validations have been applied to evaluate the generalization of these predictive models owing to the lack of external validation.
Dynamic 18 F-FET imaging could provide additional valuable information besides regular static parameters including lesion time activity curve pattern, time to peak, and slope. These parameters could also enhance the performance of static imaging parameters for IDH mutation detection (38)(39)(40). Moreover, the combination of dynamic parameters of 18   Although the amino acid tracer PET imaging has been recommended for the glioma diagnosis and evaluation, evidence showed a small portion of glioma patients could have iso-or hypo-FET uptake than the brain background (42). Limited to the method outling the tumor VOI, our filtered radiomics model is not generally applicable for the possible glioma patients with the iso-or hypo-FET uptake. With the update of the follow-up information in our facility, methodological optimization could further facilitate the better understanding of the 18 F-FET PET/CT imaging heterogeneity in glioma patients.
There are limitations in present research. The retrospective nature of the study, the relatively low patient number, incomplete molecular biomarkers results (43), and lack of external validation limit the strength of the results. Radiomics analysis with PET/CT is substantially influenced by scanning protocol, image acquisition parameters and reconstruction algorithms (22). Harmonization and standardization for the radiomics methodology is essential for results reproductivity. Finally, the percentage of IDH mutation in our patient cohort was not well-balanced, the possible reason was that some imaging-typical glioma patients will not be recommended for amino acid tracer PET imaging owing to social-economic reasons. Thus, the general application of the radiomics model needs to be verified in an independent    validation cohort with larger sample size and more comprehensive molecular biomarker information in future studies.

CONCLUSIONS
In conclusion, we established three predictive radiomics models and one simple model using static 18 F-FET PET/CT images for the non-invasive identification of IDH genotype in adult untreated glioma patients. The combination of co-registered CT and 18 F-FET PET radiomics features could significantly enhance and well balance the IDH genotype prediction, which is crucial in treatment planning and prognostic evaluation in glioma patients.

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 authors.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by the Ethics Committee of HuaShan Hospital Fudan University. Written informed consent to participate in this study was provided by the participants' legal guardian/ next of kin.