Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Individualized discrimination of tumor recurrence from radiation necrosis in glioma patients using an integrated radiomics-based model

Individualized discrimination of tumor recurrence from radiation necrosis in glioma patients... Purpose To develop and validate an integrated model for discriminating tumor recurrence from radiation necrosis in glioma patients. Methods Data from 160 pathologically confirmed glioma patients were analyzed. The diagnostic model was developed in a 18 18 primary cohort (n = 112). Textural features were extracted from postoperative F-fluorodeoxyglucose ( F-FDG) positron 11 11 emission tomography (PET), C-methionine ( C-MET) PET, and magnetic resonance images. The least absolute shrinkage and selection operator regression model was used for feature selection and radiomics signature building. Multivariable logistic regression analysis was used to develop a model for predicting tumor recurrence. The radiomics signature, quantitative PET parameters, and clinical risk factors were incorporated in the model. The clinical value of the model was then assessed in an independent validation cohort using the remaining 48 glioma patients. Results The integrated model consisting of 15 selected features was significantly associated with postoperative tumor recurrence (p < 0.001 for both primary and validation cohorts). Predictors contained in the individualized diagnosis model included the radiomics signature, the mean of tumor-background ratio (TBR) of F-FDG, maximum of TBR of C-MET PET, and patient age. The integrated model demonstrated good discrimination, with an area under the curve (AUC) of 0.988, with a 95% confidence interval (CI) of 0.975–1.000. Application in the validation cohort showed good differentiation (AUC of 0.914 and 95% CI of 0.881–0.945). Decision curve analysis showed that the integrated diagnosis model was clinically useful. Conclusions Our developed model could be used to assist the postoperative individualized diagnosis of tumor recurrence in patients with gliomas. . . . . Keywords Glioma Radiomics Recurrence MRI PET Introduction Glioma is the most common and aggressive malignant brain This article is part of the Topical Collection on Oncology – Brain tumor in adults [1]. The accurate identification of tumor re- Electronic supplementary material The online version of this article currence in patients with gliomas is crucial for selecting treat- (https://doi.org/10.1007/s00259-019-04604-0) contains supplementary material, which is available to authorized users. ment strategies to provide better therapeutic management. Early and accurate postoperative knowledge of tumor recur- * Lin Ai rence can provide valuable information for determining adju- [email protected] vant therapies. Previous studies revealed that F-fluorodeoxyglucose Department of Nuclear Medicine, Beijing Tiantan Hospital, Capital 18 11 11 18 ( F-FDG) [2, 3], C-methionine ( C-MET) [4], F- Medical University, 119, West Road of South 4th Ring, Fengtai 18 11 fluoroethyl-L-tyrosine ( F-FET) [5, 6], and C-choline [7] District, Beijing, China 2 PET, along with MRI, can differentiate between tumor recur- Department of PET/MR Advanced Application, GE Healthcare, rence and radiation necrosis with various levels of diagnostic Beijing, China Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1401 efficiencies [8, 9]. However, conventional hybrid PET/ within 48 h of surgery for contrast-enhanced tumors, MRI studies did not fully perform deep mining of the or all the abnormal hyperintense changes on preopera- intrinsic features of the images, which could be further tive MR images for tumors not demonstrating contrast investigated using advanced methodology in a larger enhancement [19]. The postoperative adjuvant treatment cohort [8–11]. was radiation therapy alone or concomitant temozolo- Radiomics has attracted increased attention in recent mide administration with fractionated radiotherapy, years as it has the potential to improve the accuracy of followed by up to six cycles of adjuvant temozolomide. recurrence predictions in oncology [12–15]. The appli- Follow-up visit, MRI, and telephone interviews were cation of radiomics enables parallel investigation of conducted periodically after surgery with a minimum multiple imaging features and enables high-throughput follow-up duration of 3 months after the completion of mining of quantitative image features from standard-of- chemoradiotherapy. Tumor progression and radiation ne- care medical imaging to improve diagnostic, classifica- crosis were defined according to the criteria in a previ- tion, prognostic, and predictive accuracy, providing a ous study [20]. The overall follow-up duration of the powerful tool in modern medicine [12, 16–18]. study was 40 months, between May 2015 and Therefore, the aim of this study was to develop and September 2018. Accordingly, 118 patients (73 males validate an integrated model that incorporated features and 45 females, mean age 44.48 ± 10.32 years with a 18 11 from PET (with both F-FDG and C-MET) and MRI range of 16 to 66 years) had tumor recurrence, and 42 images, along with clinical risk factors for individual patients (23 males and 19 females, mean age 44.74 ± discriminating tumor recurrence from radiation necrosis 12.13 years with a range of 24 to 74 years) were identified as in glioma patients. having radiation necrosis. Data assignment and MR and PET imaging Materials and methods Of the 160 patients, 70% (112 patients) were assigned to the Patients primary cohort by stratified sampling, including 83 cases of tumor recurrence and 29 cases of radiation necrosis; the re- For this retrospective analysis, ethical approval was obtained, maining 30% (48 patients) were selected for the validation and the informed consent requirement was waived by our cohort with 35 cases of tumor recurrence and 13 cases of institutional reviewing board. Selection of the cohort followed radiation necrosis. an evaluation of the institutional database in Beijing Tiantan MR images were obtained from GE 3.0T scanners Hospital for medical records from April 2015 to March 2018 (Genesis Signa and Signa HDe) and Siemens 3.0T scan- to identify patients with cerebral gliomas who underwent sur- ners (Trio Tim and Verio). Post-contrast images were gical resection. The inclusion and exclusion criteria are as acquired immediately after injection of the contrast follows: inclusion criteria: (1) patients who underwent surgery agent. The interval between contrast injection and the for cerebral gliomas, (2) pathologically confirmed cerebral start of contrast-enhanced T1-weighted image acquisi- gliomas, (3) postoperative MRI (including contrast-enhanced tion wasalways75–85 s. Postoperative MR scans for T1-weighted imaging) and PET (including both F-FDG and determining the extent of resection were performed C-MET PET) were performed (the time between MRI and within 72 h of this procedure, and the radiological pa- PET scans was less than 2 weeks), (4) postoperative radiother- rameters were maintained in accordance with the preoperative apy received with or without chemotherapy, and (5) interview scans. 18 11 or telephone follow-up information available; exclusion F-FDG and C-MET PET images were acquired using a criteria: (1) preoperative central nervous system disease of PET/CTscanner (Elite Discovery, GE Healthcare, USA) using other kinds, (2) unknown histological grade, and (3) loss of a 5-mm axial resolution and full-width-at-half-maximum at contact post-operation/patient did not return for postoperative the center of the field of view of 4 mm. Imaging data were procedures. Those patients who satisfied each inclusion or reconstructed into 30 axial planes with a slice thickness of exclusion criterion were identified for the whole cohort and 5 mm and a 192 × 192 image matrix. All patients underwent 18 11 were further assigned to either the primary cohort or validation F-FDG or C-MET PET scans according to the same pro- cohort randomly. tocol. F-FDG was intravenously injected at a dose of 3.7 MBq/kg and whole-brain image acquisition was started 11 11 Treatment and follow-up 60 min later. For C-MET PET, 555–740 MBq of C-MET was intravenously injected and whole-brain imaging was Gross total resection (GTR) was defined as there was no vis- started 10 min later. Subjects were scanned in the supine po- ible contrast enhancement on postoperative MR images sition and instructed to remain completely quiet throughout 1402 Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 the scanning procedure. The scanning times for both F-FDG the data processing, images feature, and biomarker selection and C-MET PET were 8–10 min. Postoperative PET scans procedure [22]. were performed according to the onset of worsening Two physicians performed ROI delineation for each patient symptoms of the patients after operation, and the time and obtained two sets of radiomics features. In order to obtain 18 11 interval between F-FDG and C-MET PET was at a relatively stable integrated radiomics-based model, we cal- least 2 days in order to eliminate the potential biologi- culated the relatively stable radiomics by calculating the intra- cal radiotracer crossover effect. class coefficient correlation (ICC) index. A total of 1188 (396 × 3) imaging ensembles were obtained for the three sequences Image pre-processing of FDG, MET, and MR images, and the characteristics of ICC > 0.8 were retained, which yielded a relative high inter-observer variability in the segmented tumor PET and MR images with different resolutions were resampled and normalized to the same dimensions and gray- volume. The texture analysis–based 3D ROIs are reported in the scale level. The PET and MR images were not resampled simultaneously, but separately; and the resolution of PET im- Supplemental Data (Appendix 1). A flow chart of the analysis process used in the present study is shown in Fig. 1. All tex- ages and MR images was not used. In order to minimize the loss of information, we separately perform image group fea- ture features were standardized and normalized with a regres- sion model to remove error discrepancies introduced using ture extraction on them. The standardization process is carried out for the statistical analysis of the omics characteristics. For different scanning instruments and methods. all 160 glioma patients, texture analysis was applied to their 18 11 MR and PET ( F-FDG and C-MET) images using an in- house texture analysis software, called AnalysisKit (GE Feature selection and radiomics signatures Healthcare, China). Contrast-enhanced T1WI, FLAIR, and 18 11 PET ( F-FDG and C-MET) data were retrieved from the The least absolute shrinkage and selection operator (LASSO) method, which is suitable for the regression of high- institution archive in Beijing Tiantan Hospital for the texture analysis herein. By using T1 contrast-enhanced (lesion dimensional data [23], was used to select the most useful predictive features from the primary data set. A radiomics showed contrast enhancement) or FLAIR (lesion without con- trast enhancement) MR images as the reference modality of score (rad-score) was calculated for each patient via a linear combination of selected features that were weighted by their the delineation, the regions of interest (ROI) of the lesion for each slice of images were delineated manually by two expe- respective coefficients. For the model with three imaging mo- dalities (model ), the performance of a specific rienced neuroradiologists. For each patient, the lesion mask [FDG+MET+MRI] (ROIs of the lesion) was combined to generate the final ROI radiomics signature for predicting tumor recurrence was first for further texture analysis. The patient information was hid- evaluated in the primary cohort and then confirmed in the validation cohort using an independent t test. Then, we com- den during this process using ITK-SNAP software [21]. The image biomarker standardization initiative (IBSI) was pared the diagnostic efficiency of the radiomics signature be- tween models with three modalities (model ) regarded as reference and taken into consideration in most of [FDG+MET+MRI] Fig. 1 Schematic diagram showing the image analysis and model clinical factors was performed. Finally, the radiomics signature and 18 11 building processes. The abnormal signal region of F-FDG, C-MET, patient features were applied for diagnostic evaluation to achieve and MRI images was firstly segmented manually, followed by use of a personalized discrimination of tumor recurrence from radiation necrosis feature extraction algorithm. Then, selection of image features and Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1403 and two modalities (model , model , and Statistical analysis [FDG+MET] [FDG+MRI] model ). [MET+MRI] For all radiomics features, after obtaining 912 (FDG 303; Statistical analysis was performed using R Studio soft- MET 297; MRI 312) features with high consistency, since the ware (version 1.2.1335) [26]. LASSO binary logistic features do not satisfy the normality, we use Spearman’srank regression was performed using the “glmnet” package. correlation coefficient redundancy analysis. The Spearman Multivariate binary logistic regression and diagnosis correlation coefficient takes a value of 0.9; that is, for all modeling were performed using the “stats” package. 912 features, a two-two correlation calculation is performed. Decision curve analysis was performed using the When the coefficient r ≥ 0.9, the system will randomly delete “DecisionCurve” function. one feature and retain another feature. In the end, there are 354 The differences in patient features between patients with radiomics features; that is, the dimensions of the entire process tumor recurrence and radiation necrosis in both the primary feature range from 912 to 354. and validation cohorts were assessed by the independent sample t test or Mann-Whitney test according to the data distribution type. The chi-squared test was used Integrated diagnosis model to compare the significance of the differences between categorical variables. The same statistical analysis was The integrated model included patient features (age, gender, performed to assess the difference between the two co- and body height and weight), contrast enhancement (+/−), the horts, where the tumor recurrence and radiation necrosis maximum of tumor-background ratio (TBR )and the mean max groups were evaluated separately. The diagnostic perfor- of tumor-background ratio (TBR )of both F-FDG and mean C-MET PET images, and tumor grade. Patient features and mance of models was evaluated using the receiver op- erating characteristic (ROC) curve. The statistical signif- the radiomics signature were applied to develop an integrated icance levels were all two-sided; the statistical significance diagnostic model for tumor recurrence using LASSO binary was set at p <0.05. logistic regression analysis in the primary cohort. Similarly, an integrated score (int-score) was calculated for each patient via a linear combination of selected features that were weighted Results by their respective coefficients. Decision curve analysis was conducted to determine whether the model is clinically useful Clinical characteristics by quantifying the net benefits at different threshold probabil- ities in the validation cohort [24]. From April 2015 to March 2018, there are 1562 patients with cerebral gliomas who underwent surgical resection Cross-validation in our institute. In total, 160 patients were identified for the whole cohort in the present study according to the To improve the performance of the integrated model, a tenfold inclusion and exclusion criteria, and were further cross-validation of the model was carried out in the study. A assigned to either the primary cohort or validation co- lot of features were improved in the regularized L1 logistic hort randomly (Fig. 2). The characteristics of the pa- regression with penalty term. As expressed in the following tients in the primary and validation cohorts are shown equation, in Table 1. The rate of tumor recurrence in the primary hi 1 1 ðÞ i ðÞ i ðÞ i and validation cohorts was 74.1% and 72.9%, respec- LwðÞ¼ ∑ ln 1 þ exp β  x −y β  x þ λ kk β m 2 i¼1 tively; this difference was not significant (p =0.875). In addition, there were no significant differences in the ‖β‖ was the penalty term, also expressed as ‖β‖ =|β |+ patient features between the primary and validation co- 1 1 1 |β |+ … +|β |. L(w)was theloss function. horts, either within the tumor recurrence cohort or in 2 p For better performance of the integrated model, the best λ the radiation necrosis cohort (Supplemental Tables 1-3). was obtained during the cross-validation procedure. Five in- The difference between the rad-scores of the tumor recurrence dependent sub-cohorts were divided in the training cohort, and radiation necrosis patients in the primary cohort was sig- and four of which were applied for the model fitting; the other nificant (p < 0.001), which was also confirmed in the valida- one sub-group was applied for the validation cohort. With 5 tion cohort (p <0.001). times repetition, each sub-group was treated as validation co- Representative MRI and PET images indicating the hort. And finally, the λ was gained in the cross-validation set. features of tumor recurrence and radiation necrosis are And the results were displayed with such regularized L1 lo- showninFig. 3. Of the texture features, 396 features gistic regression [25]. The cross-validation procedure was car- were reduced to 20 potential features considering 112 ried out using R Studio software (version 1.2.1335). patients in the primary cohorts (Supplemental Figure 1A). 1404 Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 Diagnostic performance of radiomics signature With a differential diagnosis threshold value of 0.710 for tumor recurrence and radiation necrosis, the model [FDG+MET+MRI] yielded an AUC of 0.932 (95% CI, 0.887–0.986) in the primary cohort and 0.910 (95% CI, 0.855–0.973) in the validation co- hort (Fig. 4a, b). In clinical diagnostic practice, this model dem- onstrated good diagnostic performance in distinguishing tumor recurrence in both primary and validation cohorts (Fig. 5a, b). In addition, we further investigated and compared other three types of models using two of the three imaging modalities, i.e., model , model , and model .The re- [FDG+MET] [FDG+MRI] [MET+MRI] sults of evaluation of the diagnostic performance by ROC anal- ysis are presented in Table 2. The diagnostic accuracy of model (AUC = 0.932; 95% CI = 0.887–0.986) [FDG+MET+MRI] was significantly higher than that of model[ (AUC = MET+MRI] 0.811; 95% CI = 0.711–0.912). However, although the AUC of model was higher than those of the other two [FDG+MET+MRI] types of models (model : AUC = 0.898; 95% CI = [FDG+MET] 0.841–0.955 and model : AUC = 0.891; 95% CI = [FDG+MRI] Fig. 2 Flow chart of the selection of patients with cerebral gliomas who 0.823–0.958), the differences were not statistically significant. underwent surgical resection from April 2015 to March 2018. Based on In addition, the diagnostic performance of the models based on the inclusion and exclusion criteria, a total of 160 glioma patients were 18 11 F-FDG, C-MET, and MRI, respectively, is provided in enrolled in this study as the whole cohort and were further distributed randomly to either the primary cohort or validation cohort to explore and Table 3. verify the discrimination performance of the model between tumor recurrence and radiation necrosis Integrated diagnosis model Calculation of the rad-score was performed using the formula Combined with clinical characteristics, we further developed shown as follows: an integrated diagnosis model by logistic regression. Finally, Radiomics score (rad-score) calculation 18 11 the age, TBR of F-FDG PET, TBR of C-MET PET, mean max and other 12 textual features were shown to be significant contributors for discriminating tumor recurrence from radia- Rad score ¼ −1:161464−0:111113 tion necrosis (Supplemental Figure 1B). Thesefeatureswere Quantile0:025−0:187241  RMS−0:154078 included in the integrated score (int-score) calculation, along ClusterProminence AllDirection offset4 SD with the int-score distribution (Fig. 6). þ0:007201  ClusterShade angle0 offset7 þ 0:266849 Integrated score (int-score) calculation ClusterShade angle135 offset7 þ 0:202809 Int−score ¼ 1:55460−0:06206  age þ 0:11767  TBR mean Correlation AllDirection offset4 SD þ 0:150674 þ1:17543  TBR þ 0:13864 max Correlation angle135 offset4−0:119975 ClusterProminence AllDirection offset4 SD−0:24507 Correlation angle45 offset7−0:014077 ClusterShade angle135 offfset7−0:19557 HaralickCorrelation AllDirection offset4 SD InverseDifferenceMoment AllDirection offset4 SD−0:18425 þ0:137885  Inertia AllDirection offset7 SD InverseDifferenceMoment AllDirection offset7 SD−0:11953 −0:048716  LongRunHighGreyLevelEmphasis angle135 offset7 InverseDifferenceMoment angle135 offset4−0:16515 þ0:222189  ShortRunLowGreyLevelEmphasis angle90 offset7 ShortRunEmphasis AllDirection offset4 SD−0:01222 −0:025067  RelativeDeviation−0:254328  stdDeviation ClusterProminence angle45 offset7−0:29295 þ0:102539  GLCMEnergy angle45 offset7 þ 0:014467 HaralickCorrelation AllDirection offset7 SD þ 0:20089 HaralickCorrelation AllDirection offset1 SD þ 0:111329 ShortRunHighGreyLevelEmphasis AllDirection offset1 SD HaralickCorrelation AllDirection offset7 SD−0:112513 þ0:03032  Quantile0:025 þ 0:12080 Sphericity−0:211177  Correlation angle135 offset7 Correlation angle45 offset7 þ 0:02933 þ0:074150  HaralickCorrelation AllDirection offset7 SD ShortRunHighGreyLevelEmphasis AllDirection offset4 SD Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1405 Table 1 Summary of the patient data in the primary and validation cohorts (n = 160) used in the study Primary cohort (n = 112) Validation cohort (n =48) Tumor recurrence Radiation necrosis p* Tumor recurrence Radiation necrosis p* Age, mean ± SD (years) 43.87 ± 9.90 46.45 ± 11.61 0.251 45.94 ± 11.26 40.92 ± 12.84 0.193 Gender, no. (%) 0.878 0.571 Male 50 (60.2) 17 (58.6) 22 (62.9) 7 (53.8) Female 33 (39.8) 12 (41.4) 13 (37.1) 6 (46.2) MRI contrast enhancement 0.095 0.323 Yes 76(91.6) 23(79.3) 32(91.4) 10(76.9) No 7 (8.4) 6 (20.7) 3 (8.6) 3 (23.1) F-FDG uptake TBR 4.15 ± 2.41 2.28 ± 2.29 < 0.001 4.53 ± 2.96 2.32 ± 1.16 0.012 max TBR 2.83 ± 1.38 1.54 ± 1.21 < 0.001 3.04 ± 1.75 1.63 ± 0.73 0.008 mean C-methionine uptake TBR 4.17 ± 2.62 1.74 ± 1.05 < 0.001 4.15 ± 1.53 2.05 ± 2.14 < 0.001 max TBR 2.81 ± 2.12 1.23 ± 0.62 < 0.001 2.65 ± 1.07 1.33 ± 1.06 < 0.001 mean WHO grade, no. (%) 0.292 0.828 II 38 (45.8) 18 (62.1) 11 (31.4) 5 (38.4) III 21 (25.3) 6 (20.7) 14 (40.0) 4 (30.8) IV 24 (28.9) 5 (17.2) 10 (28.6) 4 (30.8) Radiomics score, mean ± SD 1.49 ± 0.52 0.19 ± 0.78 < 0.001 1.46 ± 0.55 0.43 ± 0.68 < 0.001 Integrated score, mean ± SD 2.27 ± 1.53 − 0.52 ± 0.95 < 0.001 2.20 ± 1.18 −0.09 ± 1.76 < 0.001 *p values were derived from the univariable association analysis between clinical variables SD, standard deviation; MRI, magnetic resonance imaging; FDG, fluorodeoxyglucose; TBR, tumor-to-background ratio The difference in the int-score values between the tumor recurrence generally had higher int-score values in both the recurrence and radiation necrosis patients in the primary co- primary and validation cohorts (Table 1). hort was significant (p < 0.001), which was then confirmed in Notably, the integrated model yielded the largest AUC of the validation cohort (p < 0.001). Patients with tumor 0.988 (95% CI, 0.975–1.000) in the primary cohort and 0.914 18 11 11 Fig. 3 Representative MRI and PET images showing features of tumor fluorodeoxyglucose ( F-FDG) and C-methionine ( C-MET) PET, and recurrence and radiation necrosis, including images from T1-weighted pathological analyses (T1WI), T2-weighted (T2WI), and contrast-enhanced T1W1 MRI, F- 1406 Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 Fig. 4 Sensitivity versus 1- specificity for the primary (a)and validation (b) cohorts using the diagnosis model based only on radiomics signatures and primary (c) and validation (d) cohorts for the integrated diagnosis model with both clinical features and radiomics signatures. The area under the curve (AUC) values are given, along with the threshold (sensitivity, specificity) for each case (95% CI, 0.881–0.945) in the validation cohort (Fig. 4c, d). necrosis. Incorporating the clinical factors and radiomics sig- With a threshold of 0.712, the integrated model demonstrated natures into an integrated model could provide better assis- better diagnostic performance between prediction and obser- tance for the postoperative diagnosis of tumor recurrence. vation than that of the model (Fig. 5c, d). The accurate differentiation between tumor recurrence and [FDG+MET+MRI] Compared with the predictive models derived only from tex- radiation necrosis in postoperative follow-up is crucial for tural features, the integrated model was significantly better at decision-making regarding further clinical treatment, and has distinguishing postoperative tumor recurrence from radiation been investigated in many studies by comparing quantitative necrosis in patients with gliomas. imaging parameters and advanced imaging processing The decision curve for the integrated diagnosis model is methods [27–30]. To improve the diagnostic efficiency, the compared with those of the other models (based only on synergetic effect of multiparametric PET and MRI parameters radiomics signatures) in Fig. 7. The decision curve analysis was highlighted in previous studies. This indicates that the 18 18 showed that if the threshold probability of the patients was > integrated F-FET or F-FDG PET/MRI analysis could as- 0.15, performing tumor recurrence diagnosis using the inte- sist in the management of glioma patients by timely and con- grated diagnostic model added more benefit than either the clusive recognition of true tumor recurrence [9, 10, 31]. Being treat-all-patients scheme or the treat-none scheme; with the embedded in clinical practice, radiomics could provide a com- optimal threshold of 0.741, the patients would receive the prehensive quantification of imaging information. Papp et al. most benefit from clinical treatment. [32] proposed that survival prediction could be improved using computer-supported predictive models considering in vivo, ex vivo, and patient features. Our integrated model demonstrated adequate discrimina- Discussion tion between tumor recurrence and radiation necrosis in both primary and validation cohorts. As the difference between In the present study, we developed and validated a radiomics signature–based diagnostic model for individualized discrim- AUC values of the primary and validation cohorts was not statistically significant, we propose that the integrated model ination of postoperative glioma recurrence from radiation Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1407 Fig. 5 Diagnostic performance for the model based on only radiomics signature derived from 18 11 F-FDG, C-MET, and MRI. Radiomics score for the primary (a) and validation (b) cohorts. Integrated score for the primary (c) and validation (d)cohorts. The diagrams show the differentiation ability of each model in terms of the agreement between the predicted risk and observed outcomes of tumor recurrence. The dotted line represents the threshold for tumor recurrence diagnosis: 0.895 and 0.905 for the radiomics score and integrated score, respectively was robust for diagnosis and could be applied in the validation clinical management of glioma patients [33, 34]. Clinical phy- cohort. This suggests that multidimensional individual infor- sicians and radiologists could use our integrated diagnostic mation might be a more promising approach for improving model (with radiomics signatures and clinical variables Table 2 Diagnostic performance 18 11 18 11 Modalities F-FDG + C-MET F-FDG + MRI C-MET + MRI of textural features in models with two imaging modalities Cohort Primary Validation Primary Validation Primary Validation AUC 0.898 0.891 0.891 0.863 0.811 0.806 Accuracy 0.813 0.792 0.857 0.812 0.759 0.688 Sensitivity 0.781 0.750 0.854 0.833 0.780 0.722 Specificity 0.900 0.917 0.867 0.750 0.700 0.583 Threshold 0.749 0.711 0.740 Feature number 13 15 17 18 18 11 11 F-FDG, F-fluorodeoxyglucose; C-MET, C-methionine; MRI, magnetic resonance imaging; AUC,area under the curve 1408 Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 Table 3 Diagnostic performance 18 11 Modalities F-FDG PET C-MET PET MRI of textural features in single- modality model Cohort Primary Validation Primary Validation Primary Validation AUC 0.868 0.810 0.767 0.750 0.699 0.622 Accuracy 0.784 0.714 0.721 0.735 0.694 0.691 Sensitivity 0.744 0.694 0.732 0.750 0.683 0.628 Specificity 0.897 0.769 0.690 0.692 0.744 0.651 Threshold 0.782 0.755 0.739 Feature number 8 5 5 18 18 11 11 F-FDG, F-fluorodeoxyglucose; C-MET, C-methionine; MRI, magnetic resonance imaging; AUC,area under the curve available postoperatively) to perform an individualized diag- assess whether the radiomics-based integrated model nosis of the risk of glioma recurrence, which follows the cur- could assist clinical treatment decisions provides further rent trend of personalized medicine [16, 35]. information about clinical consequences based on The proposed use of the integrated diagnostic model threshold probability, and quantifies the net benefit is assisting clinical decision-making for postoperative [35, 36]. glioma patients during the follow-up process. However, Performance differences in between single modalities the recurrence diagnosis could not provide a specific revealed that the diagnostic model based on only F- level of discrimination, which is necessary for clinical FDG PET image features had higher AUC that sug- practice [36, 37]. The decision curve analysis used to gested a better differential diagnosis performance, Fig. 6 Integrated scores (int- score) distribution for all patients in the primary (a) and validation (b) cohorts, where the tumor re- currence (red bar) and radiation necrosis (green bar) confirmed by pathological results are indicated in different colors Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1409 The histological grade of the gliomas has been re- ported to be a predictor of patient prognosis [40–42]. Unexpectedly, the addition of the histologic grade to our integrated discrimination model did not improve the diagnostic performance, which may be attributed to the introduction of sampling bias due to the heterogenicity of glioma tissue, which may decrease the accuracy of the model. Therefore, the use of the radiomics signature, age, and uptake parameters of PET are recommended for tumor recurrence diagnosis with satisfactory discrimination. Although IDH1 mutation has remained an indepen- dent favorable prognostic molecular marker for gliomas, Fig. 7 Decision curves (net benefit vs. threshold probability) for the integrated diagnosis model and the four models based on only and is more objective and reliable than clinical criteria 18 11 radiomics signatures ( F-FDG, C-MET, and MRI). The gray dashed [43, 44], all malignant gliomas with various molecular curve represents the assumption that all patients have tumor recurrence, characteristics have the possibility of recurrence after while the black dashed curve line represents the assumption that no operation. In the present study, the integrated model patients have tumor recurrence. The intersection of these curves at 0.741 indicates the point where the patients could receive the most net could yield higher accuracy in tumor recurrence evalu- benefit from the integrated model ation without the assistance of glioma-related molecular markers. Furthermore, it is speculated that the inclusion of molecular markers into the model may further en- hance its diagnosis power. followedbymodels basedon C-MET and MRI in There are some limitations in the present study. First, turn. Furthermore, when the combined differentiation the sample size was relatively small for radiomics anal- power of two-modality models was considered, the ysis, and further studies are required to verify the cur- model still yielded a superior differential abil- rent findings. Second, the radiation necrosis group was [FDG+MET] ity for tumor recurrence, compared with the model relatively small for analysis, and the diagnostic thresh- [FDG+ and model . As the most widely used ra- olds of the integrated model may be cohort-specific; the MRI] [MET+MRI] diotracer in clinical practice, F-FDG biological metab- results shall be carefully interpreted. Third, genetic char- olism may incorporate more invisible image information acteristics, such as IDH1 mutations, were not available of lesions compared with C-MET and MRI in the for the whole cohort. In addition, the whole cohort was present study, which could potentially strengthen the not divided by tumor histologic type for further stratifi- crucial role of clinical utility of F-FDG PET. This cation. However, our integrated diagnostic model is ex- information would be useful for clinicians to help opti- pected to assist and facilitate individualized postopera- mize future diagnostic protocols for gliomas. tive discrimination of tumor recurrence from radiation The repeatability radiomics model is of an important necrosis in glioma patients. issue that could be affected by several factors, and im- age segmentation approaches are a common influencing factor. In our study, the ROIs were delineated manually Conclusion that may not be favored in radiomics models. Although automated segmentation solutions may provide better In conclusion, this paper presents an integrated model that support for the repeatability of radiomics results, ac- incorporates both patient features and radiomics signature. counting for clinical information not present in the im- The model presented can be conveniently used to facilitate ages is beyond the capabilities of the automated meth- postoperative individualized discrimination of tumor recur- od. In addition, the method to be chosen also depends rence in glioma patients. on tumor type, involvement of neighboring structures, and image features [38]. Therefore, there is a need for Acknowledgments The authors thank Yongzhong Zhang for the efforts active radiologist involvement in the segmentation pro- of radiopharmaceuticals synthesis; Wei Zhang, Qingsong Long, and Tong cess for both automated and semi-automated methods; Wu for the image data acquisition; and Ying Zhang, Xuelian Wang, and moreover, automatically generated contours can be used Zheng He for their assistance in clinical information collection. only as a starting point for lesion delineation by the physician who may decide to modify them according Funding information This work was supported by funds from the to his/her knowledge [39]. National Basic Research Program (2015CB755500), Beijing Excellent 1410 Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 Talents Project (2017000021469G278), and Beijing Natural Science 10. Pyka T, Hiob D, Preibisch C, Gempt J, Wiestler B, Schlegel J, et al. Foundation (7184207). Diagnosis of glioma recurrence using multiparametric dynamic 18F-fluoroethyl-tyrosine PET-MRI. Eur J Radiol. 2018;103:32–7. 11. Verger A, Filss CP, Lohmann P, Stoffels G, Sabel M, Wittsack HJ, Compliance with ethical standards et al. Comparison of O-(2-(18)F-fluoroethyl)-L-tyrosine positron emission tomography and perfusion-weighted magnetic resonance Ethical approval For this retrospective analysis, ethical approval was imaging in the diagnosis of patients with progressive and recurrent obtained, and the informed consent requirement was waived by the insti- glioma: a hybrid positron emission tomography/magnetic reso- tutional reviewing board of Beijing Tiantan Hospital, Capital Medical nance study. World Neurosurg. 2018;113:e727–e37. University. 12. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology. 2016;278:563–77. Conflict of interest Tingfan Wu and Zhongwei Chen are employees of 13. Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout GE Healthcare in China. RG, Granton P, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48:441–6. Open Access This article is distributed under the terms of the Creative 14. Kickingereder P, Burth S, Wick A, Gotz M, Eidel O, Schlemmer Commons Attribution 4.0 International License (http:// HP, et al. Radiomic profiling of glioblastoma: identifying an imag- creativecommons.org/licenses/by/4.0/), which permits unrestricted use, ing predictor of patient survival with improved performance over distribution, and reproduction in any medium, provided you give established clinical and radiologic risk models. Radiology. appropriate credit to the original author(s) and the source, provide a link 2016;280:880–9. to the Creative Commons license, and indicate if changes were made. 15. Lohmann P, Stoffels G, Ceccon G, Rapp M, Sabel M, Filss CP, et al. Radiation injury vs. recurrent brain metastasis: combining textural feature radiomics analysis and standard parameters may increase (18)F-FET PET accuracy without dynamic scans. Eur Radiol. 2017;27:2916–27. References 16. Lambin P, Leijenaar RTH, Deist TM, Peerlings J, de Jong EEC, van Timmeren J, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14:749–62. 1. Ostrom QT, Bauchet L, Davis FG, Deltour I, Fisher JL, Langer CE, 17. Kickingereder P, Gotz M, Muschelli J, Wick A, Neuberger U, et al. The epidemiology of glioma in adults: a “state of the science” Shinohara RT, et al. Large-scale radiomic profiling of recurrent review. Neuro Oncol. 2014;16:896–913. glioblastoma identifies an imaging predictor for stratifying anti- 2. Arora G, Sharma P, Sharma A, Mishra AK, Hazari PP, angiogenic treatment response. Clin Cancer Res. 2016;22:5765–71. Biswas A, et al. 99mTc-Methionine hybrid SPECT/CT for 18. Kickingereder P, Neuberger U, Bonekamp D, Piechotta PL, Gotz detection of recurrent glioma: comparison with 18F-FDG M, Wick A, et al. Radiomic subtyping improves disease strat- PET/CT and contrast-enhanced MRI. Clin Nucl Med. ification beyond key molecular, clinical, and standard imag- 2018;43:e132–e8. ing characteristics in patients with glioblastoma. Neuro 3. Li H, Deng L, Bai HX, Sun J, Cao Y, Tao Y, et al. Diagnostic Oncol. 2018;20:848–57. accuracy of amino acid and FDG-PET in differentiating brain me- 19. Wang K, Wang Y, Fan X, Wang J, Li G, Ma J, et al. Radiological tastasis recurrence from radionecrosis after radiotherapy: a system- features combined with IDH1 status for predicting the survival out- atic review and meta-analysis. AJNR Am J Neuroradiol. 2018;39: come of glioblastoma patients. Neuro Oncol. 2016;18:589–97. 280–8. 20. Wen PY, Macdonald DR, Reardon DA, Cloughesy TF, Sorensen 4. Deuschl C, Kirchner J, Poeppel TD, Schaarschmidt B, Kebir AG, Galanis E, et al. Updated response assessment criteria for high- S, El Hindy N, et al. (11)C-MET PET/MRI for detection of grade gliomas: response assessment in neuro-oncology working recurrent glioma. Eur J Nucl Med Mol Imaging. 2018;45: group. J Clin Oncol. 2010;28:1963–72. 593–601. 21. Yushkevich PA, Piven J, Hazlett HC, Smith RG, Ho S, Gee JC, 5. Dhermain FG, Hau P, Lanfermann H, Jacobs AH, van den Bent MJ. et al. User-guided 3D active contour segmentation of anatomical Advanced MRI and PET imaging for assessment of treatment re- structures: significantly improved efficiency and reliability. sponse in patients with gliomas. Lancet Neurol. 2010;9:906–20. Neuroimage. 2006;31:1116–28. 6. Hutterer M, Nowosielski M, Putzer D, Waitz D, Tinkhauser G, 22. Zwanenburg A, Leger S, Vallieres M, Lock S. Image biomarker Kostron H, et al. O-(2-18F-fluoroethyl)-L-tyrosine PET predicts standardisation initiative. arXiv preprint arXiv:161207003. failure of antiangiogenic treatment in patients with recurrent high- 23. Sauerbrei W, Royston P, Binder H. Selection of important variables grade glioma. J Nucl Med. 2011;52:856–64. and determination of functional form for continuous predictors in 7. Gao L, Xu W, Li T, Zheng J, Chen G. Accuracy of 11C-choline multivariable model building. Stat Med. 2007;26:5512–28. positron emission tomography in differentiating glioma recurrence 24. Vickers AJ, Cronin AM, Elkin EB, Gonen M. Extensions to deci- from radiation necrosis: a systematic review and meta-analysis. sion curve analysis, a novel method for evaluating diagnostic tests, Medicine (Baltimore). 2018;97:e11556. prediction models and molecular markers. BMC Med Inform Decis 8. Jena A, Taneja S, Gambhir A, Mishra AK, D’Souza MM, Verma Mak. 2008;8:53. SM, et al. Glioma recurrence versus radiation necrosis: single- 25. Hastie T, Tibshirani R, Friedman J. The elements of statistical learn- session multiparametric approach using simultaneous O-(2- ing: data mining, inference, and prediction. 2nd ed. New York: 18F-fluoroethyl)-L-tyrosine PET/MRI. Clin Nucl Med. Springer; 2011. 2016;41:e228–36. 26. Huang YQ, Liang CH, He L, Tian J, Liang CS, Chen X, et al. 9. Sogani SK, Jena A, Taneja S, Gambhir A, Mishra AK, D’Souza Development and validation of a radiomics nomogram for preop- MM, et al. Potential for differentiation of glioma recurrence from erative prediction of lymph node metastasis in colorectal cancer. J radionecrosis using integrated (18)F-fluoroethyl-L-tyrosine (FET) Clin Oncol. 2016;34:2157–64. positron emission tomography/magnetic resonance imaging: a pro- 27. Hojjati M, Badve C, Garg V, Tatsuoka C, Rogers L, Sloan A, et al. spective evaluation. Neurol India. 2017;65:293–301. Role of FDG-PET/MRI, FDG-PET/CT, and dynamic susceptibility Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1411 contrast perfusion MRI in differentiating radiation necrosis from 36. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent tumor recurrence in glioblastomas. J Neuroimaging. 2018;28: reporting of a multivariable prediction model for individual prog- 118–25. nosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 28. Nael K, Bauer AH, Hormigo A, Lemole M, Germano IM, Puig J, 2015;350:g7594. et al. Multiparametric MRI for differentiation of radiation necrosis 37. Van Calster B, Vickers AJ. Calibration of risk prediction models: from recurrent tumor in patients with treated glioblastoma. AJR Am impact on decision-analytic performance. Med Decis Making. J Roentgenol. 2018;210:18–23. 2015;35:162–9. 29. Suh CH, Kim HS, Jung SC, Choi CG, Kim SJ. Perfusion MRI as a 38. Reuze S, Schernberg A, Orlhac F, Sun R, Chargari C, Dercle L, diagnostic biomarker for differentiating glioma from brain metasta- et al. Radiomics in nuclear medicine applied to radiation therapy: sis: a systematic review and meta-analysis. Eur Radiol. 2018;28: methods, pitfalls, and challenges. Int J Radiat Oncol Biol Phys. 3819–31. 2018;102:1117–42. 30. Zhang Z, Yang J, Ho A, Jiang W, Logan J, Wang X, et al. A 39. HattM,Lee JA,Schmidtlein CR,Naqa IE,CaldwellC,De predictive model for distinguishing radiation necrosis from tumour Bernardi E, et al. Classification and evaluation strategies of auto- progression after gamma knife radiosurgery based on radiomic fea- segmentation approaches for PET: report of AAPM task group No. tures from MR images. Eur Radiol. 2018;28:2255–63. 211. Med Phys. 2017;44:e1–e42. 31. Jena A, Taneja S, Jha A, Damesha NK, Negi P, Jadhav GK, et al. 40. Pallud J, Mandonnet E, Duffau H, Kujas M, Guillevin R, Galanaud Multiparametric evaluation in differentiating glioma recurrence D, et al. Prognostic value of initial magnetic resonance imaging from treatment-induced necrosis using simultaneous (18)F-FDG- growth rates for World Health Organization grade II gliomas. Ann PET/MRI: a single-institution retrospective study. AJNR Am J Neurol. 2006;60:380–3. Neuroradiol. 2017;38:899–907. 41. Qaddoumi I, Sultan I, Gajjar A. Outcome and prognostic features in 32. Papp L, Potsch N, Grahovac M, Schmidbauer V, Woehrer A, pediatric gliomas: a review of 6212 cases from the Preusser M, et al. Glioma survival prediction with combined anal- Surveillance, Epidemiology, and End Results database. ysis of in vivo (11)C-MET PET features, ex vivo features, and Cancer. 2009;115:5761–70. patient features by supervised machine learning. J Nucl Med. 42. Tortosa A, Vinolas N, Villa S, Verger E, Gil JM, Brell M, et al. 2018;59:892–9. Prognostic implication of clinical, radiologic, and pathologic fea- 33. Kashani-Sabet M, Venna S, Nosrati M, Rangel J, Sucker A, Egberts tures in patients with anaplastic gliomas. Cancer. 2003;97:1063–71. F, et al. A multimarker prognostic assay for primary cutaneous 43. Ohgaki H, Kleihues P. The definition of primary and secondary melanoma. Clin Cancer Res. 2009;15:6987–92. glioblastoma. Clin Cancer Res. 2013;19:764–72. 34. Newby LK, Storrow AB, Gibler WB, Garvey JL, Tucker JF, 44. Sanson M, Marie Y, Paris S, Idbaih A, Laffaire J, Ducray F, et al. Kaplan AL, et al. Bedside multimarker testing for risk stratification Isocitrate dehydrogenase 1 codon 132 mutation is an important in chest pain units: the chest pain evaluation by creatine kinase-MB, prognostic biomarker in gliomas. J Clin Oncol. 2009;27:4150–4. myoglobin, and troponin I (CHECKMATE) study. Circulation. 2001;103:1832–7. 35. Balachandran VP, Gonen M, Smith JJ, DeMatteo RP. Nomograms Publisher’snote Springer Nature remains neutral with regard to jurisdic- in oncology: more than meets the eye. Lancet Oncol. 2015;16: tional claims in published maps and institutional affiliations. e173–80. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png European Journal of Nuclear Medicine and Molecular Imaging Springer Journals

Individualized discrimination of tumor recurrence from radiation necrosis in glioma patients using an integrated radiomics-based model

Download PDF
12 pages

Loading next page...
 
/lp/springer-journals/individualized-discrimination-of-tumor-recurrence-from-radiation-FNwMkwCTaL

References (51)

Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2019
Subject
Medicine & Public Health; Nuclear Medicine; Imaging / Radiology; Orthopedics; Cardiology; Oncology
ISSN
1619-7070
eISSN
1619-7089
DOI
10.1007/s00259-019-04604-0
Publisher site
See Article on Publisher Site

Abstract

Purpose To develop and validate an integrated model for discriminating tumor recurrence from radiation necrosis in glioma patients. Methods Data from 160 pathologically confirmed glioma patients were analyzed. The diagnostic model was developed in a 18 18 primary cohort (n = 112). Textural features were extracted from postoperative F-fluorodeoxyglucose ( F-FDG) positron 11 11 emission tomography (PET), C-methionine ( C-MET) PET, and magnetic resonance images. The least absolute shrinkage and selection operator regression model was used for feature selection and radiomics signature building. Multivariable logistic regression analysis was used to develop a model for predicting tumor recurrence. The radiomics signature, quantitative PET parameters, and clinical risk factors were incorporated in the model. The clinical value of the model was then assessed in an independent validation cohort using the remaining 48 glioma patients. Results The integrated model consisting of 15 selected features was significantly associated with postoperative tumor recurrence (p < 0.001 for both primary and validation cohorts). Predictors contained in the individualized diagnosis model included the radiomics signature, the mean of tumor-background ratio (TBR) of F-FDG, maximum of TBR of C-MET PET, and patient age. The integrated model demonstrated good discrimination, with an area under the curve (AUC) of 0.988, with a 95% confidence interval (CI) of 0.975–1.000. Application in the validation cohort showed good differentiation (AUC of 0.914 and 95% CI of 0.881–0.945). Decision curve analysis showed that the integrated diagnosis model was clinically useful. Conclusions Our developed model could be used to assist the postoperative individualized diagnosis of tumor recurrence in patients with gliomas. . . . . Keywords Glioma Radiomics Recurrence MRI PET Introduction Glioma is the most common and aggressive malignant brain This article is part of the Topical Collection on Oncology – Brain tumor in adults [1]. The accurate identification of tumor re- Electronic supplementary material The online version of this article currence in patients with gliomas is crucial for selecting treat- (https://doi.org/10.1007/s00259-019-04604-0) contains supplementary material, which is available to authorized users. ment strategies to provide better therapeutic management. Early and accurate postoperative knowledge of tumor recur- * Lin Ai rence can provide valuable information for determining adju- [email protected] vant therapies. Previous studies revealed that F-fluorodeoxyglucose Department of Nuclear Medicine, Beijing Tiantan Hospital, Capital 18 11 11 18 ( F-FDG) [2, 3], C-methionine ( C-MET) [4], F- Medical University, 119, West Road of South 4th Ring, Fengtai 18 11 fluoroethyl-L-tyrosine ( F-FET) [5, 6], and C-choline [7] District, Beijing, China 2 PET, along with MRI, can differentiate between tumor recur- Department of PET/MR Advanced Application, GE Healthcare, rence and radiation necrosis with various levels of diagnostic Beijing, China Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1401 efficiencies [8, 9]. However, conventional hybrid PET/ within 48 h of surgery for contrast-enhanced tumors, MRI studies did not fully perform deep mining of the or all the abnormal hyperintense changes on preopera- intrinsic features of the images, which could be further tive MR images for tumors not demonstrating contrast investigated using advanced methodology in a larger enhancement [19]. The postoperative adjuvant treatment cohort [8–11]. was radiation therapy alone or concomitant temozolo- Radiomics has attracted increased attention in recent mide administration with fractionated radiotherapy, years as it has the potential to improve the accuracy of followed by up to six cycles of adjuvant temozolomide. recurrence predictions in oncology [12–15]. The appli- Follow-up visit, MRI, and telephone interviews were cation of radiomics enables parallel investigation of conducted periodically after surgery with a minimum multiple imaging features and enables high-throughput follow-up duration of 3 months after the completion of mining of quantitative image features from standard-of- chemoradiotherapy. Tumor progression and radiation ne- care medical imaging to improve diagnostic, classifica- crosis were defined according to the criteria in a previ- tion, prognostic, and predictive accuracy, providing a ous study [20]. The overall follow-up duration of the powerful tool in modern medicine [12, 16–18]. study was 40 months, between May 2015 and Therefore, the aim of this study was to develop and September 2018. Accordingly, 118 patients (73 males validate an integrated model that incorporated features and 45 females, mean age 44.48 ± 10.32 years with a 18 11 from PET (with both F-FDG and C-MET) and MRI range of 16 to 66 years) had tumor recurrence, and 42 images, along with clinical risk factors for individual patients (23 males and 19 females, mean age 44.74 ± discriminating tumor recurrence from radiation necrosis 12.13 years with a range of 24 to 74 years) were identified as in glioma patients. having radiation necrosis. Data assignment and MR and PET imaging Materials and methods Of the 160 patients, 70% (112 patients) were assigned to the Patients primary cohort by stratified sampling, including 83 cases of tumor recurrence and 29 cases of radiation necrosis; the re- For this retrospective analysis, ethical approval was obtained, maining 30% (48 patients) were selected for the validation and the informed consent requirement was waived by our cohort with 35 cases of tumor recurrence and 13 cases of institutional reviewing board. Selection of the cohort followed radiation necrosis. an evaluation of the institutional database in Beijing Tiantan MR images were obtained from GE 3.0T scanners Hospital for medical records from April 2015 to March 2018 (Genesis Signa and Signa HDe) and Siemens 3.0T scan- to identify patients with cerebral gliomas who underwent sur- ners (Trio Tim and Verio). Post-contrast images were gical resection. The inclusion and exclusion criteria are as acquired immediately after injection of the contrast follows: inclusion criteria: (1) patients who underwent surgery agent. The interval between contrast injection and the for cerebral gliomas, (2) pathologically confirmed cerebral start of contrast-enhanced T1-weighted image acquisi- gliomas, (3) postoperative MRI (including contrast-enhanced tion wasalways75–85 s. Postoperative MR scans for T1-weighted imaging) and PET (including both F-FDG and determining the extent of resection were performed C-MET PET) were performed (the time between MRI and within 72 h of this procedure, and the radiological pa- PET scans was less than 2 weeks), (4) postoperative radiother- rameters were maintained in accordance with the preoperative apy received with or without chemotherapy, and (5) interview scans. 18 11 or telephone follow-up information available; exclusion F-FDG and C-MET PET images were acquired using a criteria: (1) preoperative central nervous system disease of PET/CTscanner (Elite Discovery, GE Healthcare, USA) using other kinds, (2) unknown histological grade, and (3) loss of a 5-mm axial resolution and full-width-at-half-maximum at contact post-operation/patient did not return for postoperative the center of the field of view of 4 mm. Imaging data were procedures. Those patients who satisfied each inclusion or reconstructed into 30 axial planes with a slice thickness of exclusion criterion were identified for the whole cohort and 5 mm and a 192 × 192 image matrix. All patients underwent 18 11 were further assigned to either the primary cohort or validation F-FDG or C-MET PET scans according to the same pro- cohort randomly. tocol. F-FDG was intravenously injected at a dose of 3.7 MBq/kg and whole-brain image acquisition was started 11 11 Treatment and follow-up 60 min later. For C-MET PET, 555–740 MBq of C-MET was intravenously injected and whole-brain imaging was Gross total resection (GTR) was defined as there was no vis- started 10 min later. Subjects were scanned in the supine po- ible contrast enhancement on postoperative MR images sition and instructed to remain completely quiet throughout 1402 Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 the scanning procedure. The scanning times for both F-FDG the data processing, images feature, and biomarker selection and C-MET PET were 8–10 min. Postoperative PET scans procedure [22]. were performed according to the onset of worsening Two physicians performed ROI delineation for each patient symptoms of the patients after operation, and the time and obtained two sets of radiomics features. In order to obtain 18 11 interval between F-FDG and C-MET PET was at a relatively stable integrated radiomics-based model, we cal- least 2 days in order to eliminate the potential biologi- culated the relatively stable radiomics by calculating the intra- cal radiotracer crossover effect. class coefficient correlation (ICC) index. A total of 1188 (396 × 3) imaging ensembles were obtained for the three sequences Image pre-processing of FDG, MET, and MR images, and the characteristics of ICC > 0.8 were retained, which yielded a relative high inter-observer variability in the segmented tumor PET and MR images with different resolutions were resampled and normalized to the same dimensions and gray- volume. The texture analysis–based 3D ROIs are reported in the scale level. The PET and MR images were not resampled simultaneously, but separately; and the resolution of PET im- Supplemental Data (Appendix 1). A flow chart of the analysis process used in the present study is shown in Fig. 1. All tex- ages and MR images was not used. In order to minimize the loss of information, we separately perform image group fea- ture features were standardized and normalized with a regres- sion model to remove error discrepancies introduced using ture extraction on them. The standardization process is carried out for the statistical analysis of the omics characteristics. For different scanning instruments and methods. all 160 glioma patients, texture analysis was applied to their 18 11 MR and PET ( F-FDG and C-MET) images using an in- house texture analysis software, called AnalysisKit (GE Feature selection and radiomics signatures Healthcare, China). Contrast-enhanced T1WI, FLAIR, and 18 11 PET ( F-FDG and C-MET) data were retrieved from the The least absolute shrinkage and selection operator (LASSO) method, which is suitable for the regression of high- institution archive in Beijing Tiantan Hospital for the texture analysis herein. By using T1 contrast-enhanced (lesion dimensional data [23], was used to select the most useful predictive features from the primary data set. A radiomics showed contrast enhancement) or FLAIR (lesion without con- trast enhancement) MR images as the reference modality of score (rad-score) was calculated for each patient via a linear combination of selected features that were weighted by their the delineation, the regions of interest (ROI) of the lesion for each slice of images were delineated manually by two expe- respective coefficients. For the model with three imaging mo- dalities (model ), the performance of a specific rienced neuroradiologists. For each patient, the lesion mask [FDG+MET+MRI] (ROIs of the lesion) was combined to generate the final ROI radiomics signature for predicting tumor recurrence was first for further texture analysis. The patient information was hid- evaluated in the primary cohort and then confirmed in the validation cohort using an independent t test. Then, we com- den during this process using ITK-SNAP software [21]. The image biomarker standardization initiative (IBSI) was pared the diagnostic efficiency of the radiomics signature be- tween models with three modalities (model ) regarded as reference and taken into consideration in most of [FDG+MET+MRI] Fig. 1 Schematic diagram showing the image analysis and model clinical factors was performed. Finally, the radiomics signature and 18 11 building processes. The abnormal signal region of F-FDG, C-MET, patient features were applied for diagnostic evaluation to achieve and MRI images was firstly segmented manually, followed by use of a personalized discrimination of tumor recurrence from radiation necrosis feature extraction algorithm. Then, selection of image features and Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1403 and two modalities (model , model , and Statistical analysis [FDG+MET] [FDG+MRI] model ). [MET+MRI] For all radiomics features, after obtaining 912 (FDG 303; Statistical analysis was performed using R Studio soft- MET 297; MRI 312) features with high consistency, since the ware (version 1.2.1335) [26]. LASSO binary logistic features do not satisfy the normality, we use Spearman’srank regression was performed using the “glmnet” package. correlation coefficient redundancy analysis. The Spearman Multivariate binary logistic regression and diagnosis correlation coefficient takes a value of 0.9; that is, for all modeling were performed using the “stats” package. 912 features, a two-two correlation calculation is performed. Decision curve analysis was performed using the When the coefficient r ≥ 0.9, the system will randomly delete “DecisionCurve” function. one feature and retain another feature. In the end, there are 354 The differences in patient features between patients with radiomics features; that is, the dimensions of the entire process tumor recurrence and radiation necrosis in both the primary feature range from 912 to 354. and validation cohorts were assessed by the independent sample t test or Mann-Whitney test according to the data distribution type. The chi-squared test was used Integrated diagnosis model to compare the significance of the differences between categorical variables. The same statistical analysis was The integrated model included patient features (age, gender, performed to assess the difference between the two co- and body height and weight), contrast enhancement (+/−), the horts, where the tumor recurrence and radiation necrosis maximum of tumor-background ratio (TBR )and the mean max groups were evaluated separately. The diagnostic perfor- of tumor-background ratio (TBR )of both F-FDG and mean C-MET PET images, and tumor grade. Patient features and mance of models was evaluated using the receiver op- erating characteristic (ROC) curve. The statistical signif- the radiomics signature were applied to develop an integrated icance levels were all two-sided; the statistical significance diagnostic model for tumor recurrence using LASSO binary was set at p <0.05. logistic regression analysis in the primary cohort. Similarly, an integrated score (int-score) was calculated for each patient via a linear combination of selected features that were weighted Results by their respective coefficients. Decision curve analysis was conducted to determine whether the model is clinically useful Clinical characteristics by quantifying the net benefits at different threshold probabil- ities in the validation cohort [24]. From April 2015 to March 2018, there are 1562 patients with cerebral gliomas who underwent surgical resection Cross-validation in our institute. In total, 160 patients were identified for the whole cohort in the present study according to the To improve the performance of the integrated model, a tenfold inclusion and exclusion criteria, and were further cross-validation of the model was carried out in the study. A assigned to either the primary cohort or validation co- lot of features were improved in the regularized L1 logistic hort randomly (Fig. 2). The characteristics of the pa- regression with penalty term. As expressed in the following tients in the primary and validation cohorts are shown equation, in Table 1. The rate of tumor recurrence in the primary hi 1 1 ðÞ i ðÞ i ðÞ i and validation cohorts was 74.1% and 72.9%, respec- LwðÞ¼ ∑ ln 1 þ exp β  x −y β  x þ λ kk β m 2 i¼1 tively; this difference was not significant (p =0.875). In addition, there were no significant differences in the ‖β‖ was the penalty term, also expressed as ‖β‖ =|β |+ patient features between the primary and validation co- 1 1 1 |β |+ … +|β |. L(w)was theloss function. horts, either within the tumor recurrence cohort or in 2 p For better performance of the integrated model, the best λ the radiation necrosis cohort (Supplemental Tables 1-3). was obtained during the cross-validation procedure. Five in- The difference between the rad-scores of the tumor recurrence dependent sub-cohorts were divided in the training cohort, and radiation necrosis patients in the primary cohort was sig- and four of which were applied for the model fitting; the other nificant (p < 0.001), which was also confirmed in the valida- one sub-group was applied for the validation cohort. With 5 tion cohort (p <0.001). times repetition, each sub-group was treated as validation co- Representative MRI and PET images indicating the hort. And finally, the λ was gained in the cross-validation set. features of tumor recurrence and radiation necrosis are And the results were displayed with such regularized L1 lo- showninFig. 3. Of the texture features, 396 features gistic regression [25]. The cross-validation procedure was car- were reduced to 20 potential features considering 112 ried out using R Studio software (version 1.2.1335). patients in the primary cohorts (Supplemental Figure 1A). 1404 Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 Diagnostic performance of radiomics signature With a differential diagnosis threshold value of 0.710 for tumor recurrence and radiation necrosis, the model [FDG+MET+MRI] yielded an AUC of 0.932 (95% CI, 0.887–0.986) in the primary cohort and 0.910 (95% CI, 0.855–0.973) in the validation co- hort (Fig. 4a, b). In clinical diagnostic practice, this model dem- onstrated good diagnostic performance in distinguishing tumor recurrence in both primary and validation cohorts (Fig. 5a, b). In addition, we further investigated and compared other three types of models using two of the three imaging modalities, i.e., model , model , and model .The re- [FDG+MET] [FDG+MRI] [MET+MRI] sults of evaluation of the diagnostic performance by ROC anal- ysis are presented in Table 2. The diagnostic accuracy of model (AUC = 0.932; 95% CI = 0.887–0.986) [FDG+MET+MRI] was significantly higher than that of model[ (AUC = MET+MRI] 0.811; 95% CI = 0.711–0.912). However, although the AUC of model was higher than those of the other two [FDG+MET+MRI] types of models (model : AUC = 0.898; 95% CI = [FDG+MET] 0.841–0.955 and model : AUC = 0.891; 95% CI = [FDG+MRI] Fig. 2 Flow chart of the selection of patients with cerebral gliomas who 0.823–0.958), the differences were not statistically significant. underwent surgical resection from April 2015 to March 2018. Based on In addition, the diagnostic performance of the models based on the inclusion and exclusion criteria, a total of 160 glioma patients were 18 11 F-FDG, C-MET, and MRI, respectively, is provided in enrolled in this study as the whole cohort and were further distributed randomly to either the primary cohort or validation cohort to explore and Table 3. verify the discrimination performance of the model between tumor recurrence and radiation necrosis Integrated diagnosis model Calculation of the rad-score was performed using the formula Combined with clinical characteristics, we further developed shown as follows: an integrated diagnosis model by logistic regression. Finally, Radiomics score (rad-score) calculation 18 11 the age, TBR of F-FDG PET, TBR of C-MET PET, mean max and other 12 textual features were shown to be significant contributors for discriminating tumor recurrence from radia- Rad score ¼ −1:161464−0:111113 tion necrosis (Supplemental Figure 1B). Thesefeatureswere Quantile0:025−0:187241  RMS−0:154078 included in the integrated score (int-score) calculation, along ClusterProminence AllDirection offset4 SD with the int-score distribution (Fig. 6). þ0:007201  ClusterShade angle0 offset7 þ 0:266849 Integrated score (int-score) calculation ClusterShade angle135 offset7 þ 0:202809 Int−score ¼ 1:55460−0:06206  age þ 0:11767  TBR mean Correlation AllDirection offset4 SD þ 0:150674 þ1:17543  TBR þ 0:13864 max Correlation angle135 offset4−0:119975 ClusterProminence AllDirection offset4 SD−0:24507 Correlation angle45 offset7−0:014077 ClusterShade angle135 offfset7−0:19557 HaralickCorrelation AllDirection offset4 SD InverseDifferenceMoment AllDirection offset4 SD−0:18425 þ0:137885  Inertia AllDirection offset7 SD InverseDifferenceMoment AllDirection offset7 SD−0:11953 −0:048716  LongRunHighGreyLevelEmphasis angle135 offset7 InverseDifferenceMoment angle135 offset4−0:16515 þ0:222189  ShortRunLowGreyLevelEmphasis angle90 offset7 ShortRunEmphasis AllDirection offset4 SD−0:01222 −0:025067  RelativeDeviation−0:254328  stdDeviation ClusterProminence angle45 offset7−0:29295 þ0:102539  GLCMEnergy angle45 offset7 þ 0:014467 HaralickCorrelation AllDirection offset7 SD þ 0:20089 HaralickCorrelation AllDirection offset1 SD þ 0:111329 ShortRunHighGreyLevelEmphasis AllDirection offset1 SD HaralickCorrelation AllDirection offset7 SD−0:112513 þ0:03032  Quantile0:025 þ 0:12080 Sphericity−0:211177  Correlation angle135 offset7 Correlation angle45 offset7 þ 0:02933 þ0:074150  HaralickCorrelation AllDirection offset7 SD ShortRunHighGreyLevelEmphasis AllDirection offset4 SD Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1405 Table 1 Summary of the patient data in the primary and validation cohorts (n = 160) used in the study Primary cohort (n = 112) Validation cohort (n =48) Tumor recurrence Radiation necrosis p* Tumor recurrence Radiation necrosis p* Age, mean ± SD (years) 43.87 ± 9.90 46.45 ± 11.61 0.251 45.94 ± 11.26 40.92 ± 12.84 0.193 Gender, no. (%) 0.878 0.571 Male 50 (60.2) 17 (58.6) 22 (62.9) 7 (53.8) Female 33 (39.8) 12 (41.4) 13 (37.1) 6 (46.2) MRI contrast enhancement 0.095 0.323 Yes 76(91.6) 23(79.3) 32(91.4) 10(76.9) No 7 (8.4) 6 (20.7) 3 (8.6) 3 (23.1) F-FDG uptake TBR 4.15 ± 2.41 2.28 ± 2.29 < 0.001 4.53 ± 2.96 2.32 ± 1.16 0.012 max TBR 2.83 ± 1.38 1.54 ± 1.21 < 0.001 3.04 ± 1.75 1.63 ± 0.73 0.008 mean C-methionine uptake TBR 4.17 ± 2.62 1.74 ± 1.05 < 0.001 4.15 ± 1.53 2.05 ± 2.14 < 0.001 max TBR 2.81 ± 2.12 1.23 ± 0.62 < 0.001 2.65 ± 1.07 1.33 ± 1.06 < 0.001 mean WHO grade, no. (%) 0.292 0.828 II 38 (45.8) 18 (62.1) 11 (31.4) 5 (38.4) III 21 (25.3) 6 (20.7) 14 (40.0) 4 (30.8) IV 24 (28.9) 5 (17.2) 10 (28.6) 4 (30.8) Radiomics score, mean ± SD 1.49 ± 0.52 0.19 ± 0.78 < 0.001 1.46 ± 0.55 0.43 ± 0.68 < 0.001 Integrated score, mean ± SD 2.27 ± 1.53 − 0.52 ± 0.95 < 0.001 2.20 ± 1.18 −0.09 ± 1.76 < 0.001 *p values were derived from the univariable association analysis between clinical variables SD, standard deviation; MRI, magnetic resonance imaging; FDG, fluorodeoxyglucose; TBR, tumor-to-background ratio The difference in the int-score values between the tumor recurrence generally had higher int-score values in both the recurrence and radiation necrosis patients in the primary co- primary and validation cohorts (Table 1). hort was significant (p < 0.001), which was then confirmed in Notably, the integrated model yielded the largest AUC of the validation cohort (p < 0.001). Patients with tumor 0.988 (95% CI, 0.975–1.000) in the primary cohort and 0.914 18 11 11 Fig. 3 Representative MRI and PET images showing features of tumor fluorodeoxyglucose ( F-FDG) and C-methionine ( C-MET) PET, and recurrence and radiation necrosis, including images from T1-weighted pathological analyses (T1WI), T2-weighted (T2WI), and contrast-enhanced T1W1 MRI, F- 1406 Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 Fig. 4 Sensitivity versus 1- specificity for the primary (a)and validation (b) cohorts using the diagnosis model based only on radiomics signatures and primary (c) and validation (d) cohorts for the integrated diagnosis model with both clinical features and radiomics signatures. The area under the curve (AUC) values are given, along with the threshold (sensitivity, specificity) for each case (95% CI, 0.881–0.945) in the validation cohort (Fig. 4c, d). necrosis. Incorporating the clinical factors and radiomics sig- With a threshold of 0.712, the integrated model demonstrated natures into an integrated model could provide better assis- better diagnostic performance between prediction and obser- tance for the postoperative diagnosis of tumor recurrence. vation than that of the model (Fig. 5c, d). The accurate differentiation between tumor recurrence and [FDG+MET+MRI] Compared with the predictive models derived only from tex- radiation necrosis in postoperative follow-up is crucial for tural features, the integrated model was significantly better at decision-making regarding further clinical treatment, and has distinguishing postoperative tumor recurrence from radiation been investigated in many studies by comparing quantitative necrosis in patients with gliomas. imaging parameters and advanced imaging processing The decision curve for the integrated diagnosis model is methods [27–30]. To improve the diagnostic efficiency, the compared with those of the other models (based only on synergetic effect of multiparametric PET and MRI parameters radiomics signatures) in Fig. 7. The decision curve analysis was highlighted in previous studies. This indicates that the 18 18 showed that if the threshold probability of the patients was > integrated F-FET or F-FDG PET/MRI analysis could as- 0.15, performing tumor recurrence diagnosis using the inte- sist in the management of glioma patients by timely and con- grated diagnostic model added more benefit than either the clusive recognition of true tumor recurrence [9, 10, 31]. Being treat-all-patients scheme or the treat-none scheme; with the embedded in clinical practice, radiomics could provide a com- optimal threshold of 0.741, the patients would receive the prehensive quantification of imaging information. Papp et al. most benefit from clinical treatment. [32] proposed that survival prediction could be improved using computer-supported predictive models considering in vivo, ex vivo, and patient features. Our integrated model demonstrated adequate discrimina- Discussion tion between tumor recurrence and radiation necrosis in both primary and validation cohorts. As the difference between In the present study, we developed and validated a radiomics signature–based diagnostic model for individualized discrim- AUC values of the primary and validation cohorts was not statistically significant, we propose that the integrated model ination of postoperative glioma recurrence from radiation Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1407 Fig. 5 Diagnostic performance for the model based on only radiomics signature derived from 18 11 F-FDG, C-MET, and MRI. Radiomics score for the primary (a) and validation (b) cohorts. Integrated score for the primary (c) and validation (d)cohorts. The diagrams show the differentiation ability of each model in terms of the agreement between the predicted risk and observed outcomes of tumor recurrence. The dotted line represents the threshold for tumor recurrence diagnosis: 0.895 and 0.905 for the radiomics score and integrated score, respectively was robust for diagnosis and could be applied in the validation clinical management of glioma patients [33, 34]. Clinical phy- cohort. This suggests that multidimensional individual infor- sicians and radiologists could use our integrated diagnostic mation might be a more promising approach for improving model (with radiomics signatures and clinical variables Table 2 Diagnostic performance 18 11 18 11 Modalities F-FDG + C-MET F-FDG + MRI C-MET + MRI of textural features in models with two imaging modalities Cohort Primary Validation Primary Validation Primary Validation AUC 0.898 0.891 0.891 0.863 0.811 0.806 Accuracy 0.813 0.792 0.857 0.812 0.759 0.688 Sensitivity 0.781 0.750 0.854 0.833 0.780 0.722 Specificity 0.900 0.917 0.867 0.750 0.700 0.583 Threshold 0.749 0.711 0.740 Feature number 13 15 17 18 18 11 11 F-FDG, F-fluorodeoxyglucose; C-MET, C-methionine; MRI, magnetic resonance imaging; AUC,area under the curve 1408 Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 Table 3 Diagnostic performance 18 11 Modalities F-FDG PET C-MET PET MRI of textural features in single- modality model Cohort Primary Validation Primary Validation Primary Validation AUC 0.868 0.810 0.767 0.750 0.699 0.622 Accuracy 0.784 0.714 0.721 0.735 0.694 0.691 Sensitivity 0.744 0.694 0.732 0.750 0.683 0.628 Specificity 0.897 0.769 0.690 0.692 0.744 0.651 Threshold 0.782 0.755 0.739 Feature number 8 5 5 18 18 11 11 F-FDG, F-fluorodeoxyglucose; C-MET, C-methionine; MRI, magnetic resonance imaging; AUC,area under the curve available postoperatively) to perform an individualized diag- assess whether the radiomics-based integrated model nosis of the risk of glioma recurrence, which follows the cur- could assist clinical treatment decisions provides further rent trend of personalized medicine [16, 35]. information about clinical consequences based on The proposed use of the integrated diagnostic model threshold probability, and quantifies the net benefit is assisting clinical decision-making for postoperative [35, 36]. glioma patients during the follow-up process. However, Performance differences in between single modalities the recurrence diagnosis could not provide a specific revealed that the diagnostic model based on only F- level of discrimination, which is necessary for clinical FDG PET image features had higher AUC that sug- practice [36, 37]. The decision curve analysis used to gested a better differential diagnosis performance, Fig. 6 Integrated scores (int- score) distribution for all patients in the primary (a) and validation (b) cohorts, where the tumor re- currence (red bar) and radiation necrosis (green bar) confirmed by pathological results are indicated in different colors Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1409 The histological grade of the gliomas has been re- ported to be a predictor of patient prognosis [40–42]. Unexpectedly, the addition of the histologic grade to our integrated discrimination model did not improve the diagnostic performance, which may be attributed to the introduction of sampling bias due to the heterogenicity of glioma tissue, which may decrease the accuracy of the model. Therefore, the use of the radiomics signature, age, and uptake parameters of PET are recommended for tumor recurrence diagnosis with satisfactory discrimination. Although IDH1 mutation has remained an indepen- dent favorable prognostic molecular marker for gliomas, Fig. 7 Decision curves (net benefit vs. threshold probability) for the integrated diagnosis model and the four models based on only and is more objective and reliable than clinical criteria 18 11 radiomics signatures ( F-FDG, C-MET, and MRI). The gray dashed [43, 44], all malignant gliomas with various molecular curve represents the assumption that all patients have tumor recurrence, characteristics have the possibility of recurrence after while the black dashed curve line represents the assumption that no operation. In the present study, the integrated model patients have tumor recurrence. The intersection of these curves at 0.741 indicates the point where the patients could receive the most net could yield higher accuracy in tumor recurrence evalu- benefit from the integrated model ation without the assistance of glioma-related molecular markers. Furthermore, it is speculated that the inclusion of molecular markers into the model may further en- hance its diagnosis power. followedbymodels basedon C-MET and MRI in There are some limitations in the present study. First, turn. Furthermore, when the combined differentiation the sample size was relatively small for radiomics anal- power of two-modality models was considered, the ysis, and further studies are required to verify the cur- model still yielded a superior differential abil- rent findings. Second, the radiation necrosis group was [FDG+MET] ity for tumor recurrence, compared with the model relatively small for analysis, and the diagnostic thresh- [FDG+ and model . As the most widely used ra- olds of the integrated model may be cohort-specific; the MRI] [MET+MRI] diotracer in clinical practice, F-FDG biological metab- results shall be carefully interpreted. Third, genetic char- olism may incorporate more invisible image information acteristics, such as IDH1 mutations, were not available of lesions compared with C-MET and MRI in the for the whole cohort. In addition, the whole cohort was present study, which could potentially strengthen the not divided by tumor histologic type for further stratifi- crucial role of clinical utility of F-FDG PET. This cation. However, our integrated diagnostic model is ex- information would be useful for clinicians to help opti- pected to assist and facilitate individualized postopera- mize future diagnostic protocols for gliomas. tive discrimination of tumor recurrence from radiation The repeatability radiomics model is of an important necrosis in glioma patients. issue that could be affected by several factors, and im- age segmentation approaches are a common influencing factor. In our study, the ROIs were delineated manually Conclusion that may not be favored in radiomics models. Although automated segmentation solutions may provide better In conclusion, this paper presents an integrated model that support for the repeatability of radiomics results, ac- incorporates both patient features and radiomics signature. counting for clinical information not present in the im- The model presented can be conveniently used to facilitate ages is beyond the capabilities of the automated meth- postoperative individualized discrimination of tumor recur- od. In addition, the method to be chosen also depends rence in glioma patients. on tumor type, involvement of neighboring structures, and image features [38]. Therefore, there is a need for Acknowledgments The authors thank Yongzhong Zhang for the efforts active radiologist involvement in the segmentation pro- of radiopharmaceuticals synthesis; Wei Zhang, Qingsong Long, and Tong cess for both automated and semi-automated methods; Wu for the image data acquisition; and Ying Zhang, Xuelian Wang, and moreover, automatically generated contours can be used Zheng He for their assistance in clinical information collection. only as a starting point for lesion delineation by the physician who may decide to modify them according Funding information This work was supported by funds from the to his/her knowledge [39]. National Basic Research Program (2015CB755500), Beijing Excellent 1410 Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 Talents Project (2017000021469G278), and Beijing Natural Science 10. Pyka T, Hiob D, Preibisch C, Gempt J, Wiestler B, Schlegel J, et al. Foundation (7184207). Diagnosis of glioma recurrence using multiparametric dynamic 18F-fluoroethyl-tyrosine PET-MRI. Eur J Radiol. 2018;103:32–7. 11. Verger A, Filss CP, Lohmann P, Stoffels G, Sabel M, Wittsack HJ, Compliance with ethical standards et al. Comparison of O-(2-(18)F-fluoroethyl)-L-tyrosine positron emission tomography and perfusion-weighted magnetic resonance Ethical approval For this retrospective analysis, ethical approval was imaging in the diagnosis of patients with progressive and recurrent obtained, and the informed consent requirement was waived by the insti- glioma: a hybrid positron emission tomography/magnetic reso- tutional reviewing board of Beijing Tiantan Hospital, Capital Medical nance study. World Neurosurg. 2018;113:e727–e37. University. 12. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology. 2016;278:563–77. Conflict of interest Tingfan Wu and Zhongwei Chen are employees of 13. Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout GE Healthcare in China. RG, Granton P, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48:441–6. Open Access This article is distributed under the terms of the Creative 14. Kickingereder P, Burth S, Wick A, Gotz M, Eidel O, Schlemmer Commons Attribution 4.0 International License (http:// HP, et al. Radiomic profiling of glioblastoma: identifying an imag- creativecommons.org/licenses/by/4.0/), which permits unrestricted use, ing predictor of patient survival with improved performance over distribution, and reproduction in any medium, provided you give established clinical and radiologic risk models. Radiology. appropriate credit to the original author(s) and the source, provide a link 2016;280:880–9. to the Creative Commons license, and indicate if changes were made. 15. Lohmann P, Stoffels G, Ceccon G, Rapp M, Sabel M, Filss CP, et al. Radiation injury vs. recurrent brain metastasis: combining textural feature radiomics analysis and standard parameters may increase (18)F-FET PET accuracy without dynamic scans. Eur Radiol. 2017;27:2916–27. References 16. Lambin P, Leijenaar RTH, Deist TM, Peerlings J, de Jong EEC, van Timmeren J, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14:749–62. 1. Ostrom QT, Bauchet L, Davis FG, Deltour I, Fisher JL, Langer CE, 17. Kickingereder P, Gotz M, Muschelli J, Wick A, Neuberger U, et al. The epidemiology of glioma in adults: a “state of the science” Shinohara RT, et al. Large-scale radiomic profiling of recurrent review. Neuro Oncol. 2014;16:896–913. glioblastoma identifies an imaging predictor for stratifying anti- 2. Arora G, Sharma P, Sharma A, Mishra AK, Hazari PP, angiogenic treatment response. Clin Cancer Res. 2016;22:5765–71. Biswas A, et al. 99mTc-Methionine hybrid SPECT/CT for 18. Kickingereder P, Neuberger U, Bonekamp D, Piechotta PL, Gotz detection of recurrent glioma: comparison with 18F-FDG M, Wick A, et al. Radiomic subtyping improves disease strat- PET/CT and contrast-enhanced MRI. Clin Nucl Med. ification beyond key molecular, clinical, and standard imag- 2018;43:e132–e8. ing characteristics in patients with glioblastoma. Neuro 3. Li H, Deng L, Bai HX, Sun J, Cao Y, Tao Y, et al. Diagnostic Oncol. 2018;20:848–57. accuracy of amino acid and FDG-PET in differentiating brain me- 19. Wang K, Wang Y, Fan X, Wang J, Li G, Ma J, et al. Radiological tastasis recurrence from radionecrosis after radiotherapy: a system- features combined with IDH1 status for predicting the survival out- atic review and meta-analysis. AJNR Am J Neuroradiol. 2018;39: come of glioblastoma patients. Neuro Oncol. 2016;18:589–97. 280–8. 20. Wen PY, Macdonald DR, Reardon DA, Cloughesy TF, Sorensen 4. Deuschl C, Kirchner J, Poeppel TD, Schaarschmidt B, Kebir AG, Galanis E, et al. Updated response assessment criteria for high- S, El Hindy N, et al. (11)C-MET PET/MRI for detection of grade gliomas: response assessment in neuro-oncology working recurrent glioma. Eur J Nucl Med Mol Imaging. 2018;45: group. J Clin Oncol. 2010;28:1963–72. 593–601. 21. Yushkevich PA, Piven J, Hazlett HC, Smith RG, Ho S, Gee JC, 5. Dhermain FG, Hau P, Lanfermann H, Jacobs AH, van den Bent MJ. et al. User-guided 3D active contour segmentation of anatomical Advanced MRI and PET imaging for assessment of treatment re- structures: significantly improved efficiency and reliability. sponse in patients with gliomas. Lancet Neurol. 2010;9:906–20. Neuroimage. 2006;31:1116–28. 6. Hutterer M, Nowosielski M, Putzer D, Waitz D, Tinkhauser G, 22. Zwanenburg A, Leger S, Vallieres M, Lock S. Image biomarker Kostron H, et al. O-(2-18F-fluoroethyl)-L-tyrosine PET predicts standardisation initiative. arXiv preprint arXiv:161207003. failure of antiangiogenic treatment in patients with recurrent high- 23. Sauerbrei W, Royston P, Binder H. Selection of important variables grade glioma. J Nucl Med. 2011;52:856–64. and determination of functional form for continuous predictors in 7. Gao L, Xu W, Li T, Zheng J, Chen G. Accuracy of 11C-choline multivariable model building. Stat Med. 2007;26:5512–28. positron emission tomography in differentiating glioma recurrence 24. Vickers AJ, Cronin AM, Elkin EB, Gonen M. Extensions to deci- from radiation necrosis: a systematic review and meta-analysis. sion curve analysis, a novel method for evaluating diagnostic tests, Medicine (Baltimore). 2018;97:e11556. prediction models and molecular markers. BMC Med Inform Decis 8. Jena A, Taneja S, Gambhir A, Mishra AK, D’Souza MM, Verma Mak. 2008;8:53. SM, et al. Glioma recurrence versus radiation necrosis: single- 25. Hastie T, Tibshirani R, Friedman J. The elements of statistical learn- session multiparametric approach using simultaneous O-(2- ing: data mining, inference, and prediction. 2nd ed. New York: 18F-fluoroethyl)-L-tyrosine PET/MRI. Clin Nucl Med. Springer; 2011. 2016;41:e228–36. 26. Huang YQ, Liang CH, He L, Tian J, Liang CS, Chen X, et al. 9. Sogani SK, Jena A, Taneja S, Gambhir A, Mishra AK, D’Souza Development and validation of a radiomics nomogram for preop- MM, et al. Potential for differentiation of glioma recurrence from erative prediction of lymph node metastasis in colorectal cancer. J radionecrosis using integrated (18)F-fluoroethyl-L-tyrosine (FET) Clin Oncol. 2016;34:2157–64. positron emission tomography/magnetic resonance imaging: a pro- 27. Hojjati M, Badve C, Garg V, Tatsuoka C, Rogers L, Sloan A, et al. spective evaluation. Neurol India. 2017;65:293–301. Role of FDG-PET/MRI, FDG-PET/CT, and dynamic susceptibility Eur J Nucl Med Mol Imaging (2020) 47:1400–1411 1411 contrast perfusion MRI in differentiating radiation necrosis from 36. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent tumor recurrence in glioblastomas. J Neuroimaging. 2018;28: reporting of a multivariable prediction model for individual prog- 118–25. nosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 28. Nael K, Bauer AH, Hormigo A, Lemole M, Germano IM, Puig J, 2015;350:g7594. et al. Multiparametric MRI for differentiation of radiation necrosis 37. Van Calster B, Vickers AJ. Calibration of risk prediction models: from recurrent tumor in patients with treated glioblastoma. AJR Am impact on decision-analytic performance. Med Decis Making. J Roentgenol. 2018;210:18–23. 2015;35:162–9. 29. Suh CH, Kim HS, Jung SC, Choi CG, Kim SJ. Perfusion MRI as a 38. Reuze S, Schernberg A, Orlhac F, Sun R, Chargari C, Dercle L, diagnostic biomarker for differentiating glioma from brain metasta- et al. Radiomics in nuclear medicine applied to radiation therapy: sis: a systematic review and meta-analysis. Eur Radiol. 2018;28: methods, pitfalls, and challenges. Int J Radiat Oncol Biol Phys. 3819–31. 2018;102:1117–42. 30. Zhang Z, Yang J, Ho A, Jiang W, Logan J, Wang X, et al. A 39. HattM,Lee JA,Schmidtlein CR,Naqa IE,CaldwellC,De predictive model for distinguishing radiation necrosis from tumour Bernardi E, et al. Classification and evaluation strategies of auto- progression after gamma knife radiosurgery based on radiomic fea- segmentation approaches for PET: report of AAPM task group No. tures from MR images. Eur Radiol. 2018;28:2255–63. 211. Med Phys. 2017;44:e1–e42. 31. Jena A, Taneja S, Jha A, Damesha NK, Negi P, Jadhav GK, et al. 40. Pallud J, Mandonnet E, Duffau H, Kujas M, Guillevin R, Galanaud Multiparametric evaluation in differentiating glioma recurrence D, et al. Prognostic value of initial magnetic resonance imaging from treatment-induced necrosis using simultaneous (18)F-FDG- growth rates for World Health Organization grade II gliomas. Ann PET/MRI: a single-institution retrospective study. AJNR Am J Neurol. 2006;60:380–3. Neuroradiol. 2017;38:899–907. 41. Qaddoumi I, Sultan I, Gajjar A. Outcome and prognostic features in 32. Papp L, Potsch N, Grahovac M, Schmidbauer V, Woehrer A, pediatric gliomas: a review of 6212 cases from the Preusser M, et al. Glioma survival prediction with combined anal- Surveillance, Epidemiology, and End Results database. ysis of in vivo (11)C-MET PET features, ex vivo features, and Cancer. 2009;115:5761–70. patient features by supervised machine learning. J Nucl Med. 42. Tortosa A, Vinolas N, Villa S, Verger E, Gil JM, Brell M, et al. 2018;59:892–9. Prognostic implication of clinical, radiologic, and pathologic fea- 33. Kashani-Sabet M, Venna S, Nosrati M, Rangel J, Sucker A, Egberts tures in patients with anaplastic gliomas. Cancer. 2003;97:1063–71. F, et al. A multimarker prognostic assay for primary cutaneous 43. Ohgaki H, Kleihues P. The definition of primary and secondary melanoma. Clin Cancer Res. 2009;15:6987–92. glioblastoma. Clin Cancer Res. 2013;19:764–72. 34. Newby LK, Storrow AB, Gibler WB, Garvey JL, Tucker JF, 44. Sanson M, Marie Y, Paris S, Idbaih A, Laffaire J, Ducray F, et al. Kaplan AL, et al. Bedside multimarker testing for risk stratification Isocitrate dehydrogenase 1 codon 132 mutation is an important in chest pain units: the chest pain evaluation by creatine kinase-MB, prognostic biomarker in gliomas. J Clin Oncol. 2009;27:4150–4. myoglobin, and troponin I (CHECKMATE) study. Circulation. 2001;103:1832–7. 35. Balachandran VP, Gonen M, Smith JJ, DeMatteo RP. Nomograms Publisher’snote Springer Nature remains neutral with regard to jurisdic- in oncology: more than meets the eye. Lancet Oncol. 2015;16: tional claims in published maps and institutional affiliations. e173–80.

Journal

European Journal of Nuclear Medicine and Molecular Imaging – Springer Journals

Published: Jun 26, 2020

There are no references for this article.