Radiomics improves the prognosis assessment of glioma recurrences: Focus on reliability analysis of MRI features

Radiomics improves the prognosis assessment of glioma recurrences: Focus on reliability analysis of MRI features | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Radiomics improves the prognosis assessment of glioma recurrences: Focus on reliability analysis of MRI features Linlin Li, Ying Yan, Jiaxin Zhang, Zhiru Lv, Bing Liu, Guiyuan Tong, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4647708/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Purpose To investigate whether imaging biomarkers could improve the efficacy of recurrent glioma survival prediction compared with that of the established clinical factors model. Method The clinical information of 80 patients was recorded in detail along with the radiomic features of the tumor region on recurrent MR images. An overall survival (OS) prediction model was proposed that combines clinical information and radiomic features. To improve the model’s generalizability and reliability, three-level feature selection methods (Kruskal‒Wallis test, Pearson correlation coefficient, and LASSO) were utilized. Finally, feature maps were constructed to explain the radiomic features. Results Six radiomic features and three clinical factors were identified to have prognostic value for recurrent glioma. The model combining radiomics features and clinical factors achieved better predictive performance (C-index = 0.787) than the clinical-based model (C-index = 0.734). KM survival curves showed clear differences between the high- and low-risk OS groups, with C-indexes of 0.751 ( p < .0001) and 0.687 ( p = 0.018), respectively. Conclusion Radiomics features improve overall survival prediction for recurrent glioma patients. Glioma recurrence prognosis MRI radiomic features Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1. Introduction Glioma, which originates from glial cells [ 1 ], is a malignant central nervous system (CNA) tumor [ 2 ]. The recurrence rate of malignant gliomas after initial treatment is at least 70% [ 3 ]. The median survival time for patients with malignant gliomas is approximately one year [ 4 ]. There is an urgent need to explore what factors influence the survival of patients with recurrent glioma. Traditionally, histopathology obtained from surgery and biopsy is considered the gold standard for cancer diagnosis and prognosis assessment [ 5 ]. However, the error rate of cancer pathology is as high as 23%[ 6 ]. An optimal method is expected to characterize the phenotype of the tumor. We aimed to investigate whether imaging features could improve the accuracy of recurrent glioma survival prediction by comparing only clinical factors. Radiomics is a growing trend in cancer diagnosis and prognosis. As a noninvasive method, it enables the transformation of imaging data into high-throughput quantitative imaging features [ 7 ][ 8 ], which are significantly related to disease prognosis [ 9 ][ 10 ]. In view of tumor heterogeneity, Gerlinger et al. [ 11 ] and Sottoriva et al. [ 12 ] divided whole tumor regions into different subregions to explore tumor heterogeneity. Liu S et al. [ 13 ] extracted the necrotic area of a tumor and quantified the geometric shape and complexity of the necrotic lesion with fractal dimensions to quantitatively evaluate the tumor. Jain et al. [ 6 ] reported that cerebral blood volume in peritumoral edema lesions was a risk factor significantly associated with survival. Kickingereder et al. [ 14 ] extracted 11 radiomic features and divided glioblastoma patients into high- and low-risk groups while simultaneously predicting pseudoprogression and survival. For the artificially defined tumor subregions, [ 15 ] K-clustering was used to classify 294 tumor subregions from 104 glioblastomas, and all subregions were classified into high-risk and low-risk groups. Patients with high-risk subregions in the whole tumor region were considered to have a shorter survival period. In survival analysis, various kinds of models have been adopted to predict patient prognosis. Philipp et al. [ 16 ] extracted 1043 features from different MRI sequences, tested the repeatability of the features by hypothesis and then conducted Cox regression using the penalty likelihood method. Ishwaran et al. [ 17 ] proposed random survival forests (RSFs) in 2008. In addition to carrying out survival analysis and variable screening on high-dimensional data, RSF can also prevent overfitting through random sampling. Suchorska et al. [ 18 ] used random survival forests (RSFs) in machine learning to reevaluate the predictive value of image markers related to the survival period in MR images. Lao et al. [ 19 ] used transfer learning to extract features related to glioma survival using a pretrained convolutional neural network (CNN). Radiomics features are sensitive to various factors, such as image acquisition, image preprocessing, and software used to extract radiomics features [ 20 ], which leads to poor repeatability and reproducibility. Consequently, we developed a prognosis prediction model that links radiomics and clinical signatures to prognostic outcomes in patients with recurrent glioma. To improve the model's generalizability and reliability, three-level feature selection methods (Kruskal‒Wallis test, Pearson correlation coefficient, and LASSO) were utilized. K‒M survival analysis was performed to verify the effect of the radio-score on survival. Finally, predictive radiomic feature maps were generated, which can aid physicians in obtaining quantitative imaging information. 2. Materials and Methods 2.1 Patients In this retrospective study, the patient cohort from a hospital consisted of 88 patients who experienced recurrence between September 2010 and December 2023. This study passed the medical ethics review of the Branch of Medical Ethics Committee, and as it was a retrospective study, the requirement for informed consent was waived. All workflows were in line with standard regulations and guidelines. The recurrence criteria in this study consisted of radiological and clinical standards. Radiological criteria rely on imaging examinations to determine tumor recurrence, typically manifesting as new abnormal signals or enhanced areas at the primary tumor site. Clinical criteria assess whether patients develop new neurological symptoms or experience disease progression. Two common radiotherapy modalities are often used in the clinical treatment of glioblastoma in hospitals: Gamma Knife Radiosurgery (GKRS) and conventional linear accelerator radiotherapy. However, no definitive conclusions have been drawn about the difference in the impact of the two treatments on the survival of glioblastoma patients. Given the extensive use of GKRS in our collaborating hospital and the results of our previous study [ 21 ] showing that it can significantly prolong OS and PFS, especially in patients with small recurrent lesions, short intervals, high KPS scores and multiple GKRS treatments, coupled with its lower radiation-induced damage to healthy tissue, we adopted a controlled variable approach and selected data from recurrent glioblastoma patients treated with GKRS for our study. The inclusion criteria were as follows: (a) histopathologically confirmed, recurrent, grade I-IV glioma; (b) clinical variables and corresponding follow-up records; and (c) treatment with gamma knife radiosurgery (GKRS). Patients were excluded if (a) MRI sequences were absent when gliomas recurred (n = 8). Finally, a total of 80 patients were randomly divided into two groups at a ratio of 3:1. The complete inclusion and exclusion criteria used in this study are shown in Fig. 1 . Position for Fig. 1 . Flowchart of the inclusion and exclusion criteria. The clinical characteristics of the patients, including demographics (sex, age), pre-GK treatment specifics (initial surgical time, multiple craniotomies, surgery-to-GKRS interval, adjuvant treatments after initial surgery, KPS scale), GKRS treatment parameters (number of targets, volume of PTV, maximum dose, marginal dose, central does, peritumoral does, multiple GKRS, concurrent/adjuvant chemotherapy), recurrence interval, OS, number of GKRS procedures, clinical notes, radiological images, and reports, and telephone follow-up, were recorded. All T1-weighted FLAIR images were obtained in the routine clinical workup with two MR scanners from 1.5T Signal HDxt and 3.0T Discovery MR 750 W (GE Healthcare, Fairfield, Connecticut, USA). The parameters were repetition time (TR)/echo time (TE), 2317 ms/11 ms; inversion time (TI), 860 ms; 3-mm-thick sections; field of view (FOV), 225 × 225 mm2 for the GE Signal; TR/TE, 4.044 ms/1.834 ms; and 1.5-mm slice thickness for the GE Discovery MR 750 W. 2.2 Image feature extraction One radiologist initially delineated the treatment target area, which was subsequently reviewed and confirmed by two senior doctors. The voxels were resampled to 1 mm × 1 mm × 1 mm using trilinear interpolation [ 22 ]. The detailed parameter settings are listed in Supplementary Table S1 . The flowchart of the survival analysis in this study is shown in Fig. 2 . A total of 1300 image features were extracted from the region of interest (ROI) on T1-weighted FLAIR images with an open-source Python tool named PyRadiomics 2.2.2. The feature pool contains I) first-order features, Ⅱ) shape and size features, and Ⅲ) textural features. Supplementary Table S2 shows the details of the radiomic features. Position for Fig. 2 . Flowchart of the radiomics model for survival analysis. The reproducibility and repeatability of radiomic features are critical problems introduced by image acquisition, preprocessing, and feature extraction. A three-step procedure reduced feature dimensionality. First, features dependent on the imaging scanner used were removed using Kruskal‒Wallis tests ( p > 0.05). Second, correlated features (correlation > 0.9) were eliminated by removing one. Third, LASSO regression with cross-validation was repeated 300 times with different seeds to obtain stable feature selection. The final features met two criteria: 1) were selected in most repetitions, and 2) had a C-index > 0.7. In each repetition, 10-fold cross-validation was used to determine λ to minimize error, and LASSO regression was used to select features by setting irrelevant coefficients to zero. The selected features formed the final predictive model, which maintained its predictive accuracy. The third step intermediate results are provided in Supplementary Fig. S1 . A radiomics score was then computed for each patient by a linear combination of selected features weighted by their respective coefficients. The formula for the combined radiomics features (referred to as the radio-score) is as follows: $$\sum _{i}^{m}{x}_{i}{w}_{i}= Radio\_score \left(1\right)$$ where \({w}_{i}\) is the weight of the retained features \({x}_{i}\) . 2.3 Statistical analysis Differences in clinical characteristics between the training and test cohorts were assessed using independent sample t tests (for discrete variables) and the Mann‒Whitney-Wilcoxon test (for continuous variables). The patients in both the training and test sets were divided into high-risk and low-risk groups. The classification cutoff was determined based on the radio-score obtained using X-tile software (Yale University School of Medicine, New Haven, CT, USA). The association between the radio-score and overall survival (OS) was assessed using univariate Cox regression analysis. This was followed by Kaplan‒Meier survival analysis to examine the correlation between the radio-score and OS in the training set, which was then validated in the test set. The log-rank test was used to determine whether the Kaplan‒Meier survival curves were statistically significant. 2.4 Assessment and explanation of the model Univariate Cox regression analysis was first performed to identify clinical risk factors significantly associated with survival in patients with gliomas. A nomogram was then constructed by incorporating the radio-score and significant clinical risk factors identified by univariate Cox regression analysis. Calibration curves were generated to assess the predictive accuracy of the nomogram by comparing the concordance between the predicted and observed outcomes. To facilitate interpretation and clinical application, we not only visualized the feature maps comprising the radio-score but also compared the predictive probabilities of OS obtained from the clinical model (including only clinical risk factors, C-model) versus the combined model (including radio-score and clinical risk factors, RC-model). 3. Results Table 1 summarizes the clinical risk factors. Statistical analyses revealed that the median survival of patients with glioma recurrence was approximately one year, which is similar to the survival of patients with glioblastoma [ 4 ]. No significant difference was observed between the training and test datasets ( p = 0.099–0.922). Table 1 The clinical factors of the patients who experienced recurrence are listed in the training and test datasets. Characteristic Training set Test set p Range Median Mean \(\pm\) SD Range Median Mean \(\pm\) SD Age (year) 17–78 51.6 50.28 \(\pm\) 13.41 24–73 48 59.85 \(\pm\) 14.35 0.906* OS (day) 15-4680 255 583.75 \(\pm\) 835.48 60-2385 307.5 681.45 \(\pm\) 731.38 0.621* Volume 1.6-124.1 16 28.98 \(\pm\) 29.95 2.80-106.30 18.65 28.10 \(\pm\) 26.62 0.903* Peritumoral dose 7.5–24.5 13 13.40 \(\pm\) 3.44 9–20 13.5 13.52 \(\pm\) 3.28 0.892* Central dose 13.5–42 24.75 25.89 \(\pm\) 6.28 18–40 24.75 26.04 \(\pm\) 5.68 0.922* Recurrence interval 30-6330 450 888.5 \(\pm\) 1155.7 30-2940 397.5 769.5 \(\pm\) 859.5 0.627* KPS 40–90 70 69.15 \(\pm\) 15.65 40–90 70 67.5 \(\pm\) 15.52 0.681* Gender (%) 0.201 † Female 26 (43.3%) 12 (60.0%) Male 34 (56.7%) 8 (40.0%) Status (%) 0.099 † Alive 17 (28.3%) 2 (10.0%) Dead 43(71.7%) 18 (90.0%) Times of GKRS 0.683 † One 42 (70.0%) 13 (65.0%) More 18 (30.0%) 7 (35.0%) * and †indicate that the p-value was calculated using the log-rank test and Mann-Withney-Wilcoxon test. Position for Table 1 . The clinical factors of the patients who experienced recurrence are listed in the training and test datasets. 3.1 Validation of radiomic features Irrelevant and highly correlated radiomic features were removed after the selection step. After the Kruskal‒Wallis test, 1091 imaging features were robust to the impact of the two scanner parameters. All 527 individual imaging features remained after PCC analysis. These remaining features were input into the LASSO model. Figure 3 shows the frequency of feature selection across 300 LASSO repetitions. Finally, six features ( \({f}_{1}-\) \({f}_{6}\) ) with the highest prognostic values were selected based on frequency, feature numbers, and the C-index, as shown in the formula below. The weight coefficients of the selected features and the formula for calculating the radio-score are: Radio-score = original-first order-10Percentile × (0.018) (2) + original-glcm-ClusterShade × (-0.121) + log.sigma.4.0.mm.3D-gldm-SmallDependenceHighGrayLevelEmphasis × (0.106) + wavelet. LHL-glcm-DifferenceVariance × (0.067) + wavelet. LLL-glcm-ClusterShade × (-0.100) + wavelet. LLL-gldm-LargeDependenceHighGrayLevelEmphasis × (0.021) Position for Fig. 3 . Frequency of features retained after 300 LASSO cycles. To investigate whether the radio-score has a significant prognostic value for survival in patients with gliomas, KM survival analysis was performed to compare the high-risk (radio-score > cutoff) and low-risk (radio-score < cutoff) groups. An optimal cutoff of -0.1 was used in the training set ( p < 0.0001) and test set ( p = 0.0068), as shown in Fig. 4 . The C-indexes of the radio-scores in the training and test sets were 0.751 (95% CI, 0.711 to 0.830; p < .0001) and 0.687 (95% CI, 0.506 to 0.866; p = 0.018), respectively, indicating good discriminative ability. Position for Fig. 4 . K‒M survival analysis according to the radio-score. 3.2 Construction and assessment of the nomogram Multivariate analysis demonstrated that the radiomic signature (radio-score) derived from MRI was a significant independent predictor of overall survival in recurrent glioma patients (hazard ratio 45). The RC model achieved improved predictive performance (C-index of 0.787) compared to the C-model (C-index of 0.734). A higher radio-score, greater tumor volume, shorter recurrence interval, and greater age were associated with significantly worse survival. In conclusion, MRI radiomics enables superior recurrence risk stratification and survival prediction in recurrent glioma patients when combined with clinical variables. These imaging biomarkers may inform individualized prognostication and personalized management for this patient population. The results of univariate Cox regression are shown in Supplementary Table S3 and Table S4. The prognostic value of the identified risk factors in the multivariable Cox regression analysis can be found in Table 2 . Table 2 Multivariable prognostic value of the model in the training and test datasets. Factor P- value Hazard Ratio P -value Hazard Ratio train test radio-score < 0.0001 45 (8.7, 240) 0.025 45 (1.6, 1300) age 0.01 0.96 (0.94, 0.99) 0.022 0.94(0.89, 0.99) Volume 0.072 1 (1, 1) 0.013 1 (1, 1.1) Recurrence interval 0.54 1 (1, 1) 0.088 1 (1, 1) Position for Table 2 . Multivariable prognostic value of the model in the training and test datasets. Nomograms were developed for visualizing the RC model incorporating radiomics and clinical factors compared to the C model incorporating clinical factors alone. The calibration curves demonstrated improved agreement between the predicted and observed OS with the RC model versus the C model, as shown in Fig. 5 . The proposed RC model offered superior prognostic performance and more accurate survival prediction than the use of clinical factors alone. Position for Fig. 5 . Nomogram and calibration curves for the C-model and RC-model. 3.3 Radiomics feature visualization Radiomic phenotypes of 1- and 3-year survivors (Patients A and B) were analyzed. The radiomic feature values and corresponding maps are shown in Fig. 6 . First-order statistics and texture features quantifying intensity distribution skew, asymmetry, heterogeneity, and fluctuation were greater for Patient A, reflecting greater genomic instability. Patient radiomic phenotypes aligned with cohort patterns, validating the feature selection. In particular, Table 3 presents the six feature values of patients at one and three years and the mean across the dataset. Obviously, except for \({f}_{6}\) , the mean values of the other five features are consistent with those of the special cases, indicating that this feature has statistical significance on the whole dataset. Table 3 The six feature values from two patients with 1- and 3-year OS and the mean across the entire dataset. f 1 f 2 f 3 f 4 f 5 f 6 1-year OS -0.029 -0.601 1.059 1.288 0.315 1.632 mean 0.025 -0.051 -0.060 0.083 -1.131 0.028 3-year OS -1.632 0.830 -1.575 0.143 1.288 -0.372 mean -0.047 -0.006 -0.103 -0.007 -0.089 0.044 Position for Table 3 . The six feature values from two patients with 1- and 3-year OS and the mean across the entire dataset. Position for Fig. 6 . Six-feature heatmaps from 1-year and 3-year patient ROIs. 4. Discussion The proposed RC model integrates radiomics and clinical data to predict survival in recurrent glioma patients. By incorporating age, tumor volume, and recurrence interval as clinical prognostic factors, the model with radiomic features achieved greater accuracy than did the model with clinical parameters alone. A robust radiomic signature was derived through a three-step feature selection process, enhancing reproducibility. K‒M analysis validated the radio-score as a significant predictor of OS. Calibration curves confirmed the superior survival prediction ability of the RC model compared to that of clinical factors. The presented intuitive radiomic feature maps provide practical insight into habitat quantification. This study demonstrated that combining radiomics and clinical factors creates an accurate, generalizable tool for individualized outcome prediction in recurrent glioma patients. Further validation in independent cohorts could support clinical implementation. While the three-step selection method yielded a stable radiomic signature, only a small proportion of the models achieved a C-index above 0.7 across hundreds of repeated LASSO experiments. All signature features except \({f}_{1}\) were frequently reproduced, demonstrating the method's effectiveness at identifying reliable prognostic markers in recurrent glioma. In addition, we presented maps of six radiomic features and explained the relationships among the feature values, feature maps, and survival status of patients. No compelling feature map of 2-year surviving patients was presented due to similar survival rates to those of the 1- and 3-year surviving patients in this limited sample. Dividing the ROI into peritumoral, enhanced, and necrotic subregions may unveil more nuanced texture phenotypes. In summary, the presented selection methodology derived a signature with potential generalizability, pending validation in external cohorts. The generation of intuitive feature maps holds promise as a technique for elucidating the prognostic value and clinical utility of radiomic biomarkers. This study has certain limitations. The radiomic features were derived solely from T1-weighted FLAIR images and were partly constrained by clinical examination protocols. The incorporation of multimodal MRI could enable more comprehensive image analysis and outcome prediction [ 23 ][ 24 ]. Additionally, the limited sample size precluded robust univariable Cox regression analysis, as the test set failed to identify risk factors similar to those in the training set. Larger, multicenter datasets are needed to fully validate the proposed model. Further research could integrate genetic information and refine tumor subregion analysis. In this study, we hypothesized that radiomic features may capture genetic heterogeneity. An analysis of additional genetic subtypes could elucidate this relationship. Previous studies [ 25 ] have demonstrated that most glioma recurrences occur in the peritumoral region. Segmenting the tumor into necrotic, enhanced, and edematous subregions may enable the extraction of more precise radiomic biomarkers. In conclusion, our study revealed that integrating radiomics into a predictive model with clinical factors could provide an accurate survival assessment for glioma recurrence. We developed a relatively stable and highly predictive model by facilitating a multistep selection feature. The six imaging features were individually interpreted using a common language, so we hoped that this approach would facilitate the recognition of these numeric radiomic signatures by physicians. In the future, more studies are still needed to explore imaging markers with diagnostic or survival relevance for recurrent gliomas. 5. Conclusion In conclusion, this study demonstrated that integrating radiomic features with clinical factors can improve the prognostic accuracy for recurrent glioma patients. The multistep selection process yielded a relatively robust and predictive model. Visualizing signature imaging phenotypes through intuitive habitat maps may facilitate the clinical interpretation and application of these numeric radiomic biomarkers. Further research is warranted to identify additional diagnostic or prognostic imaging markers in this patient population. Declarations Data availability The original image data and analysis results presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding author. Conflict of interest statement The authors declare that they have no competing interests. Ethics approval and consent to participate This study is a retrospective analysis of existing data collected as part of routine clinical practice. As it does not involve any direct interaction with participants and uses anonymized data. Informed consent was waived by the Medical Ethics Review of the Branch of the Medical Ethics Committee of the General Hospital of Northern Theater Command [Approval Number: Y (2020) 089], as this retrospective study utilized anonymized data from routine clinical practice, which does not require direct patient interaction. The study was conducted in accordance with the Declaration of Helsinki and all relevant national and institutional guidelines. Funding Declaration This research was supported by the National Natural Science Foundation of China (Grant Number: 62101357), Liaoning Province Science and Technology Joint Fund (Grant Number: 2023-BSBA-256), and 2023 Liaoning Province Artificial Intelligence Innovation Development Plan Project (Grant Number: 2023JH26/10200013). The funders had no role in study design, data collection, data analysis, data interpretation, or writing of the report. Open Access funding was provided by the National Natural Science Foundation of China (Grant Number: 62101357), which covered the Article Processing Charges (APCs) for this publication.” References Gore S, Chougule T, Jagtap J, Saini J, Ingalhalikar M. A Review of Radiomics and Deep Predictive Modeling in Glioma Characterization, Acad Radiol, vol. 28, no. 11, pp. 1599–1621, Nov. 2021, 10.1016/j.acra.2020.06.016 . Omuro A. Glioblastoma and Other Malignant Gliomas. JAMA. Nov. 2013;310:1842. 10.1001/jama.2013.280319 . Brandes AA, Vastola F, Monfardini S. Reoperation in Recurrent High-Grade Gliomas. Am J Clin Oncol. Aug. 1999;22(4):387–90. 10.1097/00000421-199908000-00013 . Prasanna P, Patel J, Partovi S, Madabhushi A, Tiwari P. Radiomic features from the peritumoral brain parenchyma on treatment-naïve multiparametric MR imaging predict long versus short-term survival in glioblastoma multiforme: Preliminary findings, Eur Radiol, vol. 27, no. 10, pp. 4188–4197, Oct. 2017, 10.1007/s00330-016-4637-3 . Le NQK, Hung TNK, Do DT, Lam LHT, Dang LH, Huynh T-T. Radiomics-based machine learning model for efficiently classifying transcriptome subtypes in glioblastoma patients from MRI. Comput Biol Med. May 2021;132:104320. 10.1016/j.compbiomed.2021.104320 . Jain R, et al. Outcome prediction in patients with glioblastoma by using imaging, clinical, and genomic biomarkers: Focus on the nonenhancing component of the tumor. Radiology. 2014;272(2):484–93. 10.1148/radiol.14131691 . Kumar V, et al. Radiomics: the process and the challenges. Magn Reson Imaging. Nov. 2012;30(9):1234–48. 10.1016/j.mri.2012.06.010 . Lambin P et al. Mar., Radiomics: Extracting more information from medical images using advanced feature analysis, Eur J Cancer, vol. 48, no. 4, pp. 441–446, 2012, 10.1016/j.ejca.2011.11.036 . Coroller TP et al. Mar., CT-based radiomic signature predicts distant metastasis in lung adenocarcinoma, Radiotherapy and Oncology, vol. 114, no. 3, pp. 345–350, 2015, 10.1016/j.radonc.2015.02.015 . Parmar C, et al. Radiomic feature clusters and Prognostic Signatures specific for Lung and Head & Neck cancer. Sci Rep. Jun. 2015;5(1):11044. 10.1038/srep11044 . Gerlinger M et al. Mar., Intratumor Heterogeneity and Branched Evolution Revealed by Multiregion Sequencing, New England Journal of Medicine, vol. 366, no. 10, pp. 883–892, 2012, 10.1056/NEJMoa1113205 . Sottoriva A et al. Mar., Intratumor heterogeneity in human glioblastoma reflects cancer evolutionary dynamics, Proceedings of the National Academy of Sciences, vol. 110, no. 10, pp. 4009–4014, 2013, 10.1073/pnas.1219747110 . Liu S, et al. Relationship between necrotic patterns in glioblastoma and patient survival: Fractal dimension and lacunarity analyses using magnetic resonance imaging. Sci Rep. Dec. 2017;7(1). 10.1038/s41598-017-08862-6 . Kickingereder P et al. Sep., Radiomic Profiling of Glioblastoma: Identifying an Imaging Predictor of Patient Survival with Improved Performance over Established Clinical and Radiologic Risk Models, Radiology, vol. 280, no. 3, pp. 880–889, 2016, 10.1148/radiol.2016160845 . Zhang X, et al. Survival-relevant high-risk subregion identification for glioblastoma patients: the MRI-based multiple instance learning approach. Eur Radiol. Oct. 2020;30(10):5602–10. 10.1007/s00330-020-06912-8 . Kickingereder P, et al. Radiomic subtyping improves disease stratification beyond key molecular, clinical, and standard imaging characteristics in patients with glioblastoma. Neuro Oncol. May 2018;20(6):848–57. 10.1093/neuonc/nox188 . Ishwaran H, Kogalur UB, Blackstone EH, Lauer MS. Random survival forests. Ann Appl Stat. Sep. 2008;2(3). 10.1214/08-AOAS169 . Suchorska B, et al. Contrast enhancement is a prognostic factor in IDH1/2 mutant, but not in wild-type WHO grade II/III glioma as confirmed by machine learning. Eur J Cancer. Jan. 2019;107:15–27. 10.1016/j.ejca.2018.10.019 . Lao J, et al. A Deep Learning-Based Radiomics Model for Prediction of Survival in Glioblastoma Multiforme. Sci Rep. Sep. 2017;7(1):10353. 10.1038/s41598-017-10649-8 . Traverso A, Wee L, Dekker A, Gillies R. Repeatability and Reproducibility of Radiomic Features: A Systematic Review. Int J Radiation Oncology*Biology*Physics. Nov. 2018;102(4):1143–58. 10.1016/j.ijrobp.2018.05.053 . Sun Y, et al. Efficacy and indications of gamma knife radiosurgery for recurrent low-and high-grade glioma. BMC Cancer. Jan. 2024;24(1). 10.1186/s12885-023-11772-8 . Li Q, et al. A Fully-Automatic Multiparametric Radiomics Model: Towards Reproducible and Prognostic Imaging Signature for Prediction of Overall Survival in Glioblastoma Multiforme. Sci Rep. Dec. 2017;7(1). 10.1038/s41598-017-14753-7 . Zhou H et al. Jun., MRI features predict survival and molecular markers in diffuse lower-grade gliomas, Neuro Oncol, vol. 19, no. 6, pp. 862–870, 2017, 10.1093/neuonc/now256 . Boonzaier NR, Larkin TJ, Matys T, van der Hoorn A, Yan J-L, Price SJ. Multiparametric MR Imaging of Diffusion and Perfusion in Contrast-enhancing and Nonenhancing Components in Patients with Glioblastoma, Radiology, vol. 284, no. 1, pp. 180–190, Jul. 2017, 10.1148/radiol.2017160150 . Nicolasjilwan M et al. Jul., Addition of MR imaging features and genetic biomarkers strengthens glioblastoma survival prediction in TCGA patients, Journal of Neuroradiology, vol. 42, no. 4, pp. 212–221, 2015, 10.1016/j.neurad.2014.02.006 . Additional Declarations No competing interests reported. Supplementary Files SupplementFile.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4647708","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":326698013,"identity":"723121cb-ced2-4b4f-b75d-a12ffafd1915","order_by":0,"name":"Linlin Li","email":"","orcid":"","institution":"Shenyang University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Linlin","middleName":"","lastName":"Li","suffix":""},{"id":326698014,"identity":"bcf23924-38a7-4a1d-8dce-e10fd347b34e","order_by":1,"name":"Ying Yan","email":"","orcid":"","institution":"General Hospital of Northern Theater Command","correspondingAuthor":false,"prefix":"","firstName":"Ying","middleName":"","lastName":"Yan","suffix":""},{"id":326698015,"identity":"d9878813-27a4-4d57-8ded-e325e5ae871f","order_by":2,"name":"Jiaxin Zhang","email":"","orcid":"","institution":"Shenyang University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Jiaxin","middleName":"","lastName":"Zhang","suffix":""},{"id":326698016,"identity":"4fae3454-74a9-48c6-9f23-5231d1cea577","order_by":3,"name":"Zhiru Lv","email":"","orcid":"","institution":"Dalian Women and Children Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Zhiru","middleName":"","lastName":"Lv","suffix":""},{"id":326698017,"identity":"4b0e7136-9679-45e8-a35d-059458f98d8b","order_by":4,"name":"Bing Liu","email":"","orcid":"","institution":"Shenyang University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Bing","middleName":"","lastName":"Liu","suffix":""},{"id":326698018,"identity":"17f9b61e-4363-4342-912a-5d7f16049af1","order_by":5,"name":"Guiyuan Tong","email":"","orcid":"","institution":"Shenyang University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Guiyuan","middleName":"","lastName":"Tong","suffix":""},{"id":326698020,"identity":"7d4f0317-ca34-45b9-999b-9b59610500cd","order_by":6,"name":"Zhaofeng Xue","email":"","orcid":"","institution":"Shenyang University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Zhaofeng","middleName":"","lastName":"Xue","suffix":""},{"id":326698022,"identity":"6942ad94-281c-41a0-bce4-68751852f459","order_by":7,"name":"Ying Sun","email":"","orcid":"","institution":"General Hospital of Northern Theater Command","correspondingAuthor":false,"prefix":"","firstName":"Ying","middleName":"","lastName":"Sun","suffix":""},{"id":326698023,"identity":"d550d3b3-823d-4358-a095-5018e66f1423","order_by":8,"name":"Xinzhuo Zhao","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA40lEQVRIiWNgGAWjYDACCRBhAEaMDyBCCcRrYYZQxGmB6GKTIEqL/Ozmh495Cu7Ym0skH6v82faHgZ89x4Dh5w7cWhjnHDM2nGHwjNlyRlrabd42AwbJnjcGjL1ncGthlkgwk/hgcJjN4EaO2W1GoBYgw4CZsQ23FjaJ9G8SCQaHeUBaCn8CtdgT0sIjkQO2RQKkhQHkMAMJAlokJHKKgX45bGBw5lmyNM85Yx6JM88KDvbi0SI/I33jY54/h+0Njicf/PijTE6Ovz1544OfeLRguhREHCBBwygYBaNgFIwCLAAATMVJsnzuh0kAAAAASUVORK5CYII=","orcid":"","institution":"Shenyang University of Technology","correspondingAuthor":true,"prefix":"","firstName":"Xinzhuo","middleName":"","lastName":"Zhao","suffix":""}],"badges":[],"createdAt":"2024-06-27 10:17:05","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4647708/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4647708/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":61005849,"identity":"234c26cd-4f00-4547-94ca-a20c84cf22c6","added_by":"auto","created_at":"2024-07-24 13:46:47","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":155386,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of the inclusion and exclusion criteria.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4647708/v1/4fb27bc6d8d358c3be1fe673.jpg"},{"id":61005831,"identity":"5885af73-9939-48d3-a4ad-7a5e521d9bee","added_by":"auto","created_at":"2024-07-24 13:46:46","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":208550,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlowchart of the radiomics model for survival analysis. \u003c/strong\u003eI)\u003cstrong\u003e \u003c/strong\u003eThe segmentation of neoplasms regions from T1-weighted FLAIR images. II) Image features consisted of shape, textural from wavelets, LoG, and gradient filter. III) Significant correlated features with OS were retained employing the lasso cox regression arithmetic. IV) Kaplan-Meier analysis and nomogram model were done for presenting the relationship of radiomics feature and OS, and visualizing the way of predicting survival in the model.\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4647708/v1/083d26814ea002da3e0b0ec8.jpg"},{"id":61005837,"identity":"b8ad0ed4-e595-418d-b593-1827e8730efd","added_by":"auto","created_at":"2024-07-24 13:46:46","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":56261,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFrequency of features retained after 300 LASSO cycles. \u003c/strong\u003eThe x-axis represents the index of features. The y-axis represents the features’ frequency. The gold color suggests the features included in the proposed model.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-4647708/v1/9db4d80f9cefcba8b0cc190d.png"},{"id":61007961,"identity":"7d4ebbd8-6cae-405c-b15e-7a8ec6d33c79","added_by":"auto","created_at":"2024-07-24 14:02:47","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":139431,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eK‒M survival analysis according to the radio-score. \u003c/strong\u003eAll patients were successfully divided into low-risk and high-risk groups with significant differences in survival on both training (a) and test sets (b).\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-4647708/v1/b968d302c216da166b3b8a76.png"},{"id":61005852,"identity":"5bd8d4d6-fbfe-498e-91af-e213055bb3c1","added_by":"auto","created_at":"2024-07-24 13:46:47","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":320357,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eNomogram and calibration curves for the C-model and RC-model. \u003c/strong\u003eThe nomogram (A) shows the prediction model with clinical factors (C-model) alone, and (C) is its calibration curve; the nomogram (B) shows the RC-model that integrated image biomarkers with clinical factors, and (D) is its calibration curve. According to nomograms, the score of every risk factor was calculated separately. Then the total points were obtained to predict the survival probability of 1, 2, and 3 years. Calibration curves showed that the RC-model (D) gave more accurate results for patients with 2 - (red curve) and 3-year survival (green curve) than the C-model (C). The calibration curves of (C) proved a better consistency than (D) between the survival prediction and the observed outcomes.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-4647708/v1/245316754562cda138a13085.png"},{"id":61006757,"identity":"0b80b76b-a827-4cef-b4b1-bbfef6805413","added_by":"auto","created_at":"2024-07-24 13:54:47","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":105150,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSix-feature heatmaps from 1-year and 3-year patient ROIs.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4647708/v1/13a98e7fe3d4e89c205e6e56.jpg"},{"id":62863819,"identity":"6b902ec9-a975-4901-ab06-75f61366b2a9","added_by":"auto","created_at":"2024-08-20 11:00:58","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1702387,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4647708/v1/e8c34d01-aa55-4712-b6e2-5bde0fad67dc.pdf"},{"id":61005851,"identity":"9bda0d53-9641-454a-90cd-748516fb3c68","added_by":"auto","created_at":"2024-07-24 13:46:47","extension":"docx","order_by":9,"title":"","display":"","copyAsset":false,"role":"supplement","size":61970,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementFile.docx","url":"https://assets-eu.researchsquare.com/files/rs-4647708/v1/25705af439cccaffe831399c.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Radiomics improves the prognosis assessment of glioma recurrences: Focus on reliability analysis of MRI features","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eGlioma, which originates from glial cells [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e], is a malignant central nervous system (CNA) tumor [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The recurrence rate of malignant gliomas after initial treatment is at least 70% [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. The median survival time for patients with malignant gliomas is approximately one year [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. There is an urgent need to explore what factors influence the survival of patients with recurrent glioma. Traditionally, histopathology obtained from surgery and biopsy is considered the gold standard for cancer diagnosis and prognosis assessment [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. However, the error rate of cancer pathology is as high as 23%[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. An optimal method is expected to characterize the phenotype of the tumor. We aimed to investigate whether imaging features could improve the accuracy of recurrent glioma survival prediction by comparing only clinical factors.\u003c/p\u003e \u003cp\u003eRadiomics is a growing trend in cancer diagnosis and prognosis. As a noninvasive method, it enables the transformation of imaging data into high-throughput quantitative imaging features [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e][\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], which are significantly related to disease prognosis [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e][\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. In view of tumor heterogeneity, Gerlinger et al. [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] and Sottoriva et al. [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] divided whole tumor regions into different subregions to explore tumor heterogeneity. Liu S et al. [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] extracted the necrotic area of a tumor and quantified the geometric shape and complexity of the necrotic lesion with fractal dimensions to quantitatively evaluate the tumor. Jain et al. [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] reported that cerebral blood volume in peritumoral edema lesions was a risk factor significantly associated with survival. Kickingereder et al. [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] extracted 11 radiomic features and divided glioblastoma patients into high- and low-risk groups while simultaneously predicting pseudoprogression and survival. For the artificially defined tumor subregions, [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] K-clustering was used to classify 294 tumor subregions from 104 glioblastomas, and all subregions were classified into high-risk and low-risk groups. Patients with high-risk subregions in the whole tumor region were considered to have a shorter survival period.\u003c/p\u003e \u003cp\u003eIn survival analysis, various kinds of models have been adopted to predict patient prognosis. Philipp et al. [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] extracted 1043 features from different MRI sequences, tested the repeatability of the features by hypothesis and then conducted Cox regression using the penalty likelihood method. Ishwaran et al. [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] proposed random survival forests (RSFs) in 2008. In addition to carrying out survival analysis and variable screening on high-dimensional data, RSF can also prevent overfitting through random sampling. Suchorska et al. [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] used random survival forests (RSFs) in machine learning to reevaluate the predictive value of image markers related to the survival period in MR images. Lao et al. [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] used transfer learning to extract features related to glioma survival using a pretrained convolutional neural network (CNN).\u003c/p\u003e \u003cp\u003eRadiomics features are sensitive to various factors, such as image acquisition, image preprocessing, and software used to extract radiomics features [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], which leads to poor repeatability and reproducibility. Consequently, we developed a prognosis prediction model that links radiomics and clinical signatures to prognostic outcomes in patients with recurrent glioma. To improve the model's generalizability and reliability, three-level feature selection methods (Kruskal‒Wallis test, Pearson correlation coefficient, and LASSO) were utilized. K‒M survival analysis was performed to verify the effect of the radio-score on survival. Finally, predictive radiomic feature maps were generated, which can aid physicians in obtaining quantitative imaging information.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Patients\u003c/h2\u003e \u003cp\u003eIn this retrospective study, the patient cohort from a hospital consisted of 88 patients who experienced recurrence between September 2010 and December 2023. This study passed the medical ethics review of the Branch of Medical Ethics Committee, and as it was a retrospective study, the requirement for informed consent was waived. All workflows were in line with standard regulations and guidelines. The recurrence criteria in this study consisted of radiological and clinical standards. Radiological criteria rely on imaging examinations to determine tumor recurrence, typically manifesting as new abnormal signals or enhanced areas at the primary tumor site. Clinical criteria assess whether patients develop new neurological symptoms or experience disease progression.\u003c/p\u003e \u003cp\u003eTwo common radiotherapy modalities are often used in the clinical treatment of glioblastoma in hospitals: Gamma Knife Radiosurgery (GKRS) and conventional linear accelerator radiotherapy. However, no definitive conclusions have been drawn about the difference in the impact of the two treatments on the survival of glioblastoma patients. Given the extensive use of GKRS in our collaborating hospital and the results of our previous study [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] showing that it can significantly prolong OS and PFS, especially in patients with small recurrent lesions, short intervals, high KPS scores and multiple GKRS treatments, coupled with its lower radiation-induced damage to healthy tissue, we adopted a controlled variable approach and selected data from recurrent glioblastoma patients treated with GKRS for our study.\u003c/p\u003e \u003cp\u003eThe inclusion criteria were as follows: (a) histopathologically confirmed, recurrent, grade I-IV glioma; (b) clinical variables and corresponding follow-up records; and (c) treatment with gamma knife radiosurgery (GKRS).\u003c/p\u003e \u003cp\u003ePatients were excluded if (a) MRI sequences were absent when gliomas recurred (n\u0026thinsp;=\u0026thinsp;8). Finally, a total of 80 patients were randomly divided into two groups at a ratio of 3:1. The complete inclusion and exclusion criteria used in this study are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003ePosition for Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Flowchart of the inclusion and exclusion criteria.\u003c/p\u003e \u003cp\u003eThe clinical characteristics of the patients, including demographics (sex, age), pre-GK treatment specifics (initial surgical time, multiple craniotomies, surgery-to-GKRS interval, adjuvant treatments after initial surgery, KPS scale), GKRS treatment parameters (number of targets, volume of PTV, maximum dose, marginal dose, central does, peritumoral does, multiple GKRS, concurrent/adjuvant chemotherapy), recurrence interval, OS, number of GKRS procedures, clinical notes, radiological images, and reports, and telephone follow-up, were recorded.\u003c/p\u003e \u003cp\u003eAll T1-weighted FLAIR images were obtained in the routine clinical workup with two MR scanners from 1.5T Signal HDxt and 3.0T Discovery MR 750 W (GE Healthcare, Fairfield, Connecticut, USA). The parameters were repetition time (TR)/echo time (TE), 2317 ms/11 ms; inversion time (TI), 860 ms; 3-mm-thick sections; field of view (FOV), 225 \u0026times; 225 mm2 for the GE Signal; TR/TE, 4.044 ms/1.834 ms; and 1.5-mm slice thickness for the GE Discovery MR 750 W.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Image feature extraction\u003c/h2\u003e \u003cp\u003eOne radiologist initially delineated the treatment target area, which was subsequently reviewed and confirmed by two senior doctors. The voxels were resampled to 1 mm \u0026times; 1 mm \u0026times; 1 mm using trilinear interpolation [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. The detailed parameter settings are listed in Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e. The flowchart of the survival analysis in this study is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. A total of 1300 image features were extracted from the region of interest (ROI) on T1-weighted FLAIR images with an open-source Python tool named PyRadiomics 2.2.2. The feature pool contains I) first-order features, Ⅱ) shape and size features, and Ⅲ) textural features. Supplementary Table S2 shows the details of the radiomic features.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePosition for\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. \u003cb\u003eFlowchart of the radiomics model for survival analysis.\u003c/b\u003e\u003c/p\u003e \u003cp\u003eThe reproducibility and repeatability of radiomic features are critical problems introduced by image acquisition, preprocessing, and feature extraction. A three-step procedure reduced feature dimensionality. First, features dependent on the imaging scanner used were removed using Kruskal‒Wallis tests (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05). Second, correlated features (correlation\u0026thinsp;\u0026gt;\u0026thinsp;0.9) were eliminated by removing one. Third, LASSO regression with cross-validation was repeated 300 times with different seeds to obtain stable feature selection. The final features met two criteria: 1) were selected in most repetitions, and 2) had a C-index\u0026thinsp;\u0026gt;\u0026thinsp;0.7. In each repetition, 10-fold cross-validation was used to determine λ to minimize error, and LASSO regression was used to select features by setting irrelevant coefficients to zero. The selected features formed the final predictive model, which maintained its predictive accuracy. The third step intermediate results are provided in Supplementary Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eA radiomics score was then computed for each patient by a linear combination of selected features weighted by their respective coefficients. The formula for the combined radiomics features (referred to as the radio-score) is as follows:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\sum _{i}^{m}{x}_{i}{w}_{i}= Radio\\_score \\left(1\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({w}_{i}\\)\u003c/span\u003e\u003c/span\u003e is the weight of the retained features \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({x}_{i}\\)\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Statistical analysis\u003c/h2\u003e \u003cp\u003eDifferences in clinical characteristics between the training and test cohorts were assessed using independent sample t tests (for discrete variables) and the Mann‒Whitney-Wilcoxon test (for continuous variables). The patients in both the training and test sets were divided into high-risk and low-risk groups. The classification cutoff was determined based on the radio-score obtained using X-tile software (Yale University School of Medicine, New Haven, CT, USA). The association between the radio-score and overall survival (OS) was assessed using univariate Cox regression analysis. This was followed by Kaplan‒Meier survival analysis to examine the correlation between the radio-score and OS in the training set, which was then validated in the test set. The log-rank test was used to determine whether the Kaplan‒Meier survival curves were statistically significant.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Assessment and explanation of the model\u003c/h2\u003e \u003cp\u003eUnivariate Cox regression analysis was first performed to identify clinical risk factors significantly associated with survival in patients with gliomas. A nomogram was then constructed by incorporating the radio-score and significant clinical risk factors identified by univariate Cox regression analysis. Calibration curves were generated to assess the predictive accuracy of the nomogram by comparing the concordance between the predicted and observed outcomes. To facilitate interpretation and clinical application, we not only visualized the feature maps comprising the radio-score but also compared the predictive probabilities of OS obtained from the clinical model (including only clinical risk factors, C-model) versus the combined model (including radio-score and clinical risk factors, RC-model).\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e summarizes the clinical risk factors. Statistical analyses revealed that the median survival of patients with glioma recurrence was approximately one year, which is similar to the survival of patients with glioblastoma [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. No significant difference was observed between the training and test datasets (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.099\u0026ndash;0.922).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe clinical factors of the patients who experienced recurrence are listed in the training and test datasets.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eCharacteristic\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eTraining set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003eTest set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRange\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMedian\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMean\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003eSD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRange\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMedian\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMean\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003eSD\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (year)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e17\u0026ndash;78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e51.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e50.28\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e13.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e24\u0026ndash;73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e59.85\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e14.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.906*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOS (day)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15-4680\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e255\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e583.75\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e835.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e60-2385\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e307.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e681.45\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e731.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.621*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVolume\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.6-124.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e28.98\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e29.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2.80-106.30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e18.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e28.10\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e26.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.903*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePeritumoral dose\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7.5\u0026ndash;24.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e13.40\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e3.44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e9\u0026ndash;20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e13.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e13.52\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e3.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.892*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCentral dose\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e13.5\u0026ndash;42\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e24.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e25.89\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e6.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e18\u0026ndash;40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e24.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e26.04\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e5.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.922*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRecurrence interval\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30-6330\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e450\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e888.5\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e1155.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e30-2940\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e397.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e769.5\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e859.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.627*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKPS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e40\u0026ndash;90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e69.15\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e15.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e40\u0026ndash;90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e67.5\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\pm\\)\u003c/span\u003e\u003c/span\u003e15.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.681*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.201\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003e26 (43.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e12 (60.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003e34 (56.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e8 (40.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStatus (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.099\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003e17 (28.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e2 (10.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDead\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003e43(71.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e18 (90.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTimes of GKRS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.683\u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOne\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003e42 (70.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e13 (65.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMore\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003e18 (30.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003e7 (35.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"8\"\u003e* and \u0026dagger;indicate that the p-value was calculated using the log-rank test and Mann-Withney-Wilcoxon test.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003ePosition for\u003c/b\u003e Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. \u003cb\u003eThe clinical factors of the patients who experienced recurrence are listed in the training and test datasets.\u003c/b\u003e\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Validation of radiomic features\u003c/h2\u003e \u003cp\u003eIrrelevant and highly correlated radiomic features were removed after the selection step. After the Kruskal‒Wallis test, 1091 imaging features were robust to the impact of the two scanner parameters. All 527 individual imaging features remained after PCC analysis. These remaining features were input into the LASSO model. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e shows the frequency of feature selection across 300 LASSO repetitions. Finally, six features (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({f}_{1}-\\)\u003c/span\u003e\u003c/span\u003e \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({f}_{6}\\)\u003c/span\u003e\u003c/span\u003e) with the highest prognostic values were selected based on frequency, feature numbers, and the C-index, as shown in the formula below. The weight coefficients of the selected features and the formula for calculating the radio-score are:\u003c/p\u003e \u003cp\u003e \u003cem\u003eRadio-score\u0026thinsp;=\u0026thinsp;original-first order-10Percentile \u0026times; (0.018)\u003c/em\u003e (2)\u003c/p\u003e \u003cp\u003e \u003cem\u003e+ original-glcm-ClusterShade \u0026times; (-0.121)\u003c/em\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003e+ log.sigma.4.0.mm.3D-gldm-SmallDependenceHighGrayLevelEmphasis \u0026times; (0.106)\u003c/em\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003e+ wavelet. LHL-glcm-DifferenceVariance \u0026times; (0.067)\u003c/em\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003e+ wavelet. LLL-glcm-ClusterShade \u0026times; (-0.100)\u003c/em\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003e+ wavelet. LLL-gldm-LargeDependenceHighGrayLevelEmphasis \u0026times; (0.021)\u003c/em\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003ePosition for\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. \u003cb\u003eFrequency of features retained after 300 LASSO cycles.\u003c/b\u003e\u003c/p\u003e \u003cp\u003eTo investigate whether the radio-score has a significant prognostic value for survival in patients with gliomas, KM survival analysis was performed to compare the high-risk (radio-score\u0026thinsp;\u0026gt;\u0026thinsp;cutoff) and low-risk (radio-score\u0026thinsp;\u0026lt;\u0026thinsp;cutoff) groups. An optimal cutoff of -0.1 was used in the training set (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.0001) and test set (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.0068), as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. The C-indexes of the radio-scores in the training and test sets were 0.751 (95% CI, 0.711 to 0.830; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.0001) and 0.687 (95% CI, 0.506 to 0.866; \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.018), respectively, indicating good discriminative ability.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePosition for\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. \u003cb\u003eK‒M survival analysis according to the radio-score.\u003c/b\u003e\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Construction and assessment of the nomogram\u003c/h2\u003e \u003cp\u003eMultivariate analysis demonstrated that the radiomic signature (radio-score) derived from MRI was a significant independent predictor of overall survival in recurrent glioma patients (hazard ratio 45). The RC model achieved improved predictive performance (C-index of 0.787) compared to the C-model (C-index of 0.734). A higher radio-score, greater tumor volume, shorter recurrence interval, and greater age were associated with significantly worse survival. In conclusion, MRI radiomics enables superior recurrence risk stratification and survival prediction in recurrent glioma patients when combined with clinical variables. These imaging biomarkers may inform individualized prognostication and personalized management for this patient population. The results of univariate Cox regression are shown in Supplementary Table S3 and Table S4. The prognostic value of the identified risk factors in the multivariable Cox regression analysis can be found in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMultivariable prognostic value of the model in the training and test datasets.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFactor\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003eP-\u003c/em\u003evalue\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHazard Ratio\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e-value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHazard Ratio\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003etrain\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003etest\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eradio-score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.0001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e45 (8.7, 240)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e45 (1.6, 1300)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eage\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.01\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.96 (0.94, 0.99)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.022\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.94(0.89, 0.99)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eVolume\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.072\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1, 1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.013\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1, 1.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRecurrence interval\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (1, 1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.088\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1 (1, 1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003ePosition for\u003c/b\u003e Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. \u003cb\u003eMultivariable prognostic value of the model in the training and test datasets.\u003c/b\u003e\u003c/p\u003e \u003cp\u003eNomograms were developed for visualizing the RC model incorporating radiomics and clinical factors compared to the C model incorporating clinical factors alone. The calibration curves demonstrated improved agreement between the predicted and observed OS with the RC model versus the C model, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. The proposed RC model offered superior prognostic performance and more accurate survival prediction than the use of clinical factors alone.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePosition for\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. \u003cb\u003eNomogram and calibration curves for the C-model and RC-model.\u003c/b\u003e\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Radiomics feature visualization\u003c/h2\u003e \u003cp\u003eRadiomic phenotypes of 1- and 3-year survivors (Patients A and B) were analyzed. The radiomic feature values and corresponding maps are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e. First-order statistics and texture features quantifying intensity distribution skew, asymmetry, heterogeneity, and fluctuation were greater for Patient A, reflecting greater genomic instability. Patient radiomic phenotypes aligned with cohort patterns, validating the feature selection. In particular, Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents the six feature values of patients at one and three years and the mean across the dataset. Obviously, except for \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({f}_{6}\\)\u003c/span\u003e\u003c/span\u003e, the mean values of the other five features are consistent with those of the special cases, indicating that this feature has statistical significance on the whole dataset.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe six feature values from two patients with 1- and 3-year OS and the mean across the entire dataset.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003ef\u003c/em\u003e\u003csub\u003e1\u003c/sub\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003ef\u003c/em\u003e\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003ef\u003c/em\u003e\u003csub\u003e3\u003c/sub\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cem\u003ef\u003c/em\u003e\u003csub\u003e4\u003c/sub\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cem\u003ef\u003c/em\u003e\u003csub\u003e5\u003c/sub\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cem\u003ef\u003c/em\u003e\u003csub\u003e6\u003c/sub\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e1-year OS\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-0.029\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-0.601\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.059\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.288\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.315\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.632\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c8\" namest=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003emean\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-0.051\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.060\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.083\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-1.131\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.028\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c8\" namest=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003e3-year OS\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-1.632\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.830\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-1.575\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.143\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.288\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-0.372\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003emean\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-0.047\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.006\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-0.103\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-0.007\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-0.089\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.044\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003ePosition for\u003c/b\u003e Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. \u003cb\u003eThe six feature values from two patients with 1- and 3-year OS and the mean across the entire dataset.\u003c/b\u003e\u003c/p\u003e \u003cp\u003e \u003cb\u003ePosition for\u003c/b\u003e Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e. \u003cb\u003eSix-feature heatmaps from 1-year and 3-year patient ROIs.\u003c/b\u003e\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThe proposed RC model integrates radiomics and clinical data to predict survival in recurrent glioma patients. By incorporating age, tumor volume, and recurrence interval as clinical prognostic factors, the model with radiomic features achieved greater accuracy than did the model with clinical parameters alone. A robust radiomic signature was derived through a three-step feature selection process, enhancing reproducibility. K‒M analysis validated the radio-score as a significant predictor of OS. Calibration curves confirmed the superior survival prediction ability of the RC model compared to that of clinical factors. The presented intuitive radiomic feature maps provide practical insight into habitat quantification. This study demonstrated that combining radiomics and clinical factors creates an accurate, generalizable tool for individualized outcome prediction in recurrent glioma patients. Further validation in independent cohorts could support clinical implementation.\u003c/p\u003e \u003cp\u003eWhile the three-step selection method yielded a stable radiomic signature, only a small proportion of the models achieved a C-index above 0.7 across hundreds of repeated LASSO experiments. All signature features except \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({f}_{1}\\)\u003c/span\u003e\u003c/span\u003e were frequently reproduced, demonstrating the method's effectiveness at identifying reliable prognostic markers in recurrent glioma. In addition, we presented maps of six radiomic features and explained the relationships among the feature values, feature maps, and survival status of patients. No compelling feature map of 2-year surviving patients was presented due to similar survival rates to those of the 1- and 3-year surviving patients in this limited sample. Dividing the ROI into peritumoral, enhanced, and necrotic subregions may unveil more nuanced texture phenotypes. In summary, the presented selection methodology derived a signature with potential generalizability, pending validation in external cohorts. The generation of intuitive feature maps holds promise as a technique for elucidating the prognostic value and clinical utility of radiomic biomarkers.\u003c/p\u003e \u003cp\u003eThis study has certain limitations. The radiomic features were derived solely from T1-weighted FLAIR images and were partly constrained by clinical examination protocols. The incorporation of multimodal MRI could enable more comprehensive image analysis and outcome prediction [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e][\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Additionally, the limited sample size precluded robust univariable Cox regression analysis, as the test set failed to identify risk factors similar to those in the training set. Larger, multicenter datasets are needed to fully validate the proposed model.\u003c/p\u003e \u003cp\u003eFurther research could integrate genetic information and refine tumor subregion analysis. In this study, we hypothesized that radiomic features may capture genetic heterogeneity. An analysis of additional genetic subtypes could elucidate this relationship. Previous studies [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] have demonstrated that most glioma recurrences occur in the peritumoral region. Segmenting the tumor into necrotic, enhanced, and edematous subregions may enable the extraction of more precise radiomic biomarkers.\u003c/p\u003e \u003cp\u003eIn conclusion, our study revealed that integrating radiomics into a predictive model with clinical factors could provide an accurate survival assessment for glioma recurrence. We developed a relatively stable and highly predictive model by facilitating a multistep selection feature. The six imaging features were individually interpreted using a common language, so we hoped that this approach would facilitate the recognition of these numeric radiomic signatures by physicians. In the future, more studies are still needed to explore imaging markers with diagnostic or survival relevance for recurrent gliomas.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eIn conclusion, this study demonstrated that integrating radiomic features with clinical factors can improve the prognostic accuracy for recurrent glioma patients. The multistep selection process yielded a relatively robust and predictive model. Visualizing signature imaging phenotypes through intuitive habitat maps may facilitate the clinical interpretation and application of these numeric radiomic biomarkers. Further research is warranted to identify additional diagnostic or prognostic imaging markers in this patient population.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eData availability\u003c/p\u003e\n\u003cp\u003eThe original image data and analysis\u0026nbsp;results\u0026nbsp;presented in the study are included in the article/supplementary material. Further inquiries can be\u0026nbsp;directed to the corresponding author.\u003c/p\u003e\n\u003cp\u003eConflict of interest statement\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003eEthics approval and consent to participate\u003c/p\u003e\n\u003cp\u003eThis study is a retrospective analysis of existing data collected as part of routine clinical practice. As it does not involve any direct interaction with participants and uses anonymized data. Informed consent was waived by the Medical Ethics Review of the Branch of the Medical Ethics Committee of the General Hospital of Northern Theater Command [Approval Number: Y (2020) 089], as this retrospective study utilized anonymized data from routine clinical practice, which does not require direct patient interaction. The study was conducted in accordance with the Declaration of Helsinki and all relevant national and institutional guidelines.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFunding Declaration\u003c/p\u003e\n\u003cp\u003eThis research was supported by the National Natural Science Foundation of China (Grant Number: 62101357), Liaoning Province Science and Technology Joint Fund (Grant Number: 2023-BSBA-256), and 2023 Liaoning Province Artificial Intelligence Innovation Development Plan Project (Grant Number: 2023JH26/10200013). The funders had no role in study design, data collection, data analysis, data interpretation, or writing of the report. Open Access funding was provided by the National Natural Science Foundation of China (Grant Number: 62101357), which covered the Article Processing Charges (APCs) for this publication.\u0026rdquo;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eGore S, Chougule T, Jagtap J, Saini J, Ingalhalikar M. A Review of Radiomics and Deep Predictive Modeling in Glioma Characterization, Acad Radiol, vol. 28, no. 11, pp. 1599\u0026ndash;1621, Nov. 2021, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.acra.2020.06.016\u003c/span\u003e\u003cspan address=\"10.1016/j.acra.2020.06.016\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOmuro A. Glioblastoma and Other Malignant Gliomas. JAMA. Nov. 2013;310:1842. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jama.2013.280319\u003c/span\u003e\u003cspan address=\"10.1001/jama.2013.280319\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrandes AA, Vastola F, Monfardini S. Reoperation in Recurrent High-Grade Gliomas. Am J Clin Oncol. Aug. 1999;22(4):387\u0026ndash;90. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/00000421-199908000-00013\u003c/span\u003e\u003cspan address=\"10.1097/00000421-199908000-00013\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePrasanna P, Patel J, Partovi S, Madabhushi A, Tiwari P. Radiomic features from the peritumoral brain parenchyma on treatment-na\u0026iuml;ve multiparametric MR imaging predict long versus short-term survival in glioblastoma multiforme: Preliminary findings, Eur Radiol, vol. 27, no. 10, pp. 4188\u0026ndash;4197, Oct. 2017, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00330-016-4637-3\u003c/span\u003e\u003cspan address=\"10.1007/s00330-016-4637-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLe NQK, Hung TNK, Do DT, Lam LHT, Dang LH, Huynh T-T. Radiomics-based machine learning model for efficiently classifying transcriptome subtypes in glioblastoma patients from MRI. Comput Biol Med. May 2021;132:104320. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.compbiomed.2021.104320\u003c/span\u003e\u003cspan address=\"10.1016/j.compbiomed.2021.104320\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJain R, et al. Outcome prediction in patients with glioblastoma by using imaging, clinical, and genomic biomarkers: Focus on the nonenhancing component of the tumor. Radiology. 2014;272(2):484\u0026ndash;93. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1148/radiol.14131691\u003c/span\u003e\u003cspan address=\"10.1148/radiol.14131691\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKumar V, et al. Radiomics: the process and the challenges. Magn Reson Imaging. Nov. 2012;30(9):1234\u0026ndash;48. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.mri.2012.06.010\u003c/span\u003e\u003cspan address=\"10.1016/j.mri.2012.06.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLambin P et al. Mar., Radiomics: Extracting more information from medical images using advanced feature analysis, Eur J Cancer, vol. 48, no. 4, pp. 441\u0026ndash;446, 2012, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ejca.2011.11.036\u003c/span\u003e\u003cspan address=\"10.1016/j.ejca.2011.11.036\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCoroller TP et al. Mar., CT-based radiomic signature predicts distant metastasis in lung adenocarcinoma, Radiotherapy and Oncology, vol. 114, no. 3, pp. 345\u0026ndash;350, 2015, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.radonc.2015.02.015\u003c/span\u003e\u003cspan address=\"10.1016/j.radonc.2015.02.015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParmar C, et al. Radiomic feature clusters and Prognostic Signatures specific for Lung and Head \u0026amp; Neck cancer. Sci Rep. Jun. 2015;5(1):11044. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/srep11044\u003c/span\u003e\u003cspan address=\"10.1038/srep11044\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGerlinger M et al. Mar., Intratumor Heterogeneity and Branched Evolution Revealed by Multiregion Sequencing, New England Journal of Medicine, vol. 366, no. 10, pp. 883\u0026ndash;892, 2012, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1056/NEJMoa1113205\u003c/span\u003e\u003cspan address=\"10.1056/NEJMoa1113205\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSottoriva A et al. Mar., Intratumor heterogeneity in human glioblastoma reflects cancer evolutionary dynamics, Proceedings of the National Academy of Sciences, vol. 110, no. 10, pp. 4009\u0026ndash;4014, 2013, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1073/pnas.1219747110\u003c/span\u003e\u003cspan address=\"10.1073/pnas.1219747110\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu S, et al. Relationship between necrotic patterns in glioblastoma and patient survival: Fractal dimension and lacunarity analyses using magnetic resonance imaging. Sci Rep. Dec. 2017;7(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-017-08862-6\u003c/span\u003e\u003cspan address=\"10.1038/s41598-017-08862-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKickingereder P et al. Sep., Radiomic Profiling of Glioblastoma: Identifying an Imaging Predictor of Patient Survival with Improved Performance over Established Clinical and Radiologic Risk Models, Radiology, vol. 280, no. 3, pp. 880\u0026ndash;889, 2016, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1148/radiol.2016160845\u003c/span\u003e\u003cspan address=\"10.1148/radiol.2016160845\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang X, et al. Survival-relevant high-risk subregion identification for glioblastoma patients: the MRI-based multiple instance learning approach. Eur Radiol. Oct. 2020;30(10):5602\u0026ndash;10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00330-020-06912-8\u003c/span\u003e\u003cspan address=\"10.1007/s00330-020-06912-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKickingereder P, et al. Radiomic subtyping improves disease stratification beyond key molecular, clinical, and standard imaging characteristics in patients with glioblastoma. Neuro Oncol. May 2018;20(6):848\u0026ndash;57. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/neuonc/nox188\u003c/span\u003e\u003cspan address=\"10.1093/neuonc/nox188\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIshwaran H, Kogalur UB, Blackstone EH, Lauer MS. Random survival forests. Ann Appl Stat. Sep. 2008;2(3). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1214/08-AOAS169\u003c/span\u003e\u003cspan address=\"10.1214/08-AOAS169\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuchorska B, et al. Contrast enhancement is a prognostic factor in IDH1/2 mutant, but not in wild-type WHO grade II/III glioma as confirmed by machine learning. Eur J Cancer. Jan. 2019;107:15\u0026ndash;27. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ejca.2018.10.019\u003c/span\u003e\u003cspan address=\"10.1016/j.ejca.2018.10.019\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLao J, et al. A Deep Learning-Based Radiomics Model for Prediction of Survival in Glioblastoma Multiforme. Sci Rep. Sep. 2017;7(1):10353. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-017-10649-8\u003c/span\u003e\u003cspan address=\"10.1038/s41598-017-10649-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTraverso A, Wee L, Dekker A, Gillies R. Repeatability and Reproducibility of Radiomic Features: A Systematic Review. Int J Radiation Oncology*Biology*Physics. Nov. 2018;102(4):1143\u0026ndash;58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijrobp.2018.05.053\u003c/span\u003e\u003cspan address=\"10.1016/j.ijrobp.2018.05.053\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun Y, et al. Efficacy and indications of gamma knife radiosurgery for recurrent low-and high-grade glioma. BMC Cancer. Jan. 2024;24(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12885-023-11772-8\u003c/span\u003e\u003cspan address=\"10.1186/s12885-023-11772-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi Q, et al. A Fully-Automatic Multiparametric Radiomics Model: Towards Reproducible and Prognostic Imaging Signature for Prediction of Overall Survival in Glioblastoma Multiforme. Sci Rep. Dec. 2017;7(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-017-14753-7\u003c/span\u003e\u003cspan address=\"10.1038/s41598-017-14753-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou H et al. Jun., MRI features predict survival and molecular markers in diffuse lower-grade gliomas, Neuro Oncol, vol. 19, no. 6, pp. 862\u0026ndash;870, 2017, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/neuonc/now256\u003c/span\u003e\u003cspan address=\"10.1093/neuonc/now256\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoonzaier NR, Larkin TJ, Matys T, van der Hoorn A, Yan J-L, Price SJ. Multiparametric MR Imaging of Diffusion and Perfusion in Contrast-enhancing and Nonenhancing Components in Patients with Glioblastoma, Radiology, vol. 284, no. 1, pp. 180\u0026ndash;190, Jul. 2017, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1148/radiol.2017160150\u003c/span\u003e\u003cspan address=\"10.1148/radiol.2017160150\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNicolasjilwan M et al. Jul., Addition of MR imaging features and genetic biomarkers strengthens glioblastoma survival prediction in TCGA patients, Journal of Neuroradiology, vol. 42, no. 4, pp. 212\u0026ndash;221, 2015, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.neurad.2014.02.006\u003c/span\u003e\u003cspan address=\"10.1016/j.neurad.2014.02.006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Glioma recurrence, prognosis, MRI, radiomic features","lastPublishedDoi":"10.21203/rs.3.rs-4647708/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4647708/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e \u003cp\u003eTo investigate whether imaging biomarkers could improve the efficacy of recurrent glioma survival prediction compared with that of the established clinical factors model.\u003c/p\u003e\u003ch2\u003eMethod\u003c/h2\u003e \u003cp\u003eThe clinical information of 80 patients was recorded in detail along with the radiomic features of the tumor region on recurrent MR images. An overall survival (OS) prediction model was proposed that combines clinical information and radiomic features. To improve the model\u0026rsquo;s generalizability and reliability, three-level feature selection methods (Kruskal‒Wallis test, Pearson correlation coefficient, and LASSO) were utilized. Finally, feature maps were constructed to explain the radiomic features.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eSix radiomic features and three clinical factors were identified to have prognostic value for recurrent glioma. The model combining radiomics features and clinical factors achieved better predictive performance (C-index\u0026thinsp;=\u0026thinsp;0.787) than the clinical-based model (C-index\u0026thinsp;=\u0026thinsp;0.734). KM survival curves showed clear differences between the high- and low-risk OS groups, with C-indexes of 0.751 (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.0001) and 0.687 (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.018), respectively.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eRadiomics features improve overall survival prediction for recurrent glioma patients.\u003c/p\u003e","manuscriptTitle":"Radiomics improves the prognosis assessment of glioma recurrences: Focus on reliability analysis of MRI features","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-07-24 13:46:42","doi":"10.21203/rs.3.rs-4647708/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4674d517-a2cb-41a5-a5a8-c97e20800c6c","owner":[],"postedDate":"July 24th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-02-12T12:23:43+00:00","versionOfRecord":[],"versionCreatedAt":"2024-07-24 13:46:42","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4647708","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4647708","identity":"rs-4647708","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}