Preoperative Prediction of Persistent Type II Endoleaks After Endovascular Aortic Repair Using Multiregional Perianeurysmal Computed Tomography Angiography Radiomics: A Multicenter Study

Preoperative Prediction of Persistent Type II Endoleaks After Endovascular Aortic Repair Using Multiregional Perianeurysmal Computed Tomography Angiography Radiomics: A Multicenter Study | 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 Preoperative Prediction of Persistent Type II Endoleaks After Endovascular Aortic Repair Using Multiregional Perianeurysmal Computed Tomography Angiography Radiomics: A Multicenter Study Xinlei Yu, Guihan Lin, Weiyue Chen, Weiming Hu, Cheng Ma, Zhuohang Shi, and 12 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9018408/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Background: Preoperative identification of patients with persistent type II endoleaks (T2ELs) after endovascular aortic repair (EVAR) for abdominal aortic aneurysm (AAA) could improve individualized management. Current approaches based on conventional anatomical factors show limited predictive accuracy. Methods: Consecutive patients with AAA undergoing EVAR at three centers were retrospectively included. Radiomic characteristics were extracted from preoperative computed tomography angiographic images, including the aneurysm sac and five concentric perianeurysmal zones positioned 2–10 mm from the outer wall. After feature selection, six radiomic models employing a support vector machine classifier were developed and subsequently compared. This optimal radiomic signature was then combined with significant clinical predictors to formulate a combined model. The model’s performance was evaluated using area under the curve (AUC), calibration plots, and decision curve analysis, while its interpretability was assessed using SHapley Additive exPlanations (SHAP) analysis. Results: The radiomic model combining features from the aneurysm intra-sac and the 6-mm perianeurysmal region demonstrated superior predictive accuracy, with AUCs of 0.910, 0.907, 0.886, and 0.859 in the training, internal validation, and two external test sets, respectively. The maximum aneurysm diameter and thrombus area were identified as the independent clinical predictors. The combined model further improved discrimination, achieving AUCs of 0.954, 0.933, 0.924, and 0.896 in the corresponding cohorts, along with excellent calibration and clinical net benefit. The SHAP analysis explained its predictions both locally and globally. Conclusion: A combined model that merges perianeurysmal radiomic features with essential clinical factors offers a precise and non-invasive approach for preoperative T2EL risk stratification following EVAR, thereby facilitating personalized surveillance protocols. Abdominal aortic aneurysm Type II endoleak Radiomics Perivascular adipose tissue Machine learning Predictive model Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Abdominal aortic aneurysm (AAA) is a condition with a significant life-threatening potential, characterized by a regional dilation of the abdominal aorta with a diagnostic threshold diameter of ≥ 3.0 cm( 1 ). Its principal risk stems from asymptomatic expansion and eventual rupture, with rupture mortality rates exceeding 80%( 2 ). Endovascular aortic repair (EVAR) has become a mainstream, minimally invasive intervention for AAA( 3 ). Nevertheless, its long-term efficacy is frequently compromised by endoleaks, which are the continuous flow of blood inside the aneurysm intra-sac that is external to the stent graft. Among them, type II endoleaks (T2ELs) are the most prevalent subtype, characterized by retrograde perfusion of the sac from patent branch arteries, such as the inferior mesenteric or lumbar arteries. Studies indicate that persistent T2ELs maintain pressurization within the sac, elevating the risks of continued sac expansion and rupture( 4 ). Therefore, accurately identifying patients at a high risk for persistent T2ELs before surgery is of critical importance. It enables the development of tailored surveillance protocols and timely prophylactic interventions, thereby improving the long-term success of EVAR( 5 ). The major imaging method for both the preliminary assessment and postoperative surveillance of AAA is computed tomography angiography (CTA), which provides in-depth anatomical information( 6 ). Current preoperative prediction of T2ELs relies predominantly on specific anatomical factors derived from CTA, including aneurysm neck angulation and the number of patent lumbar arteries( 7 , 8 ). However, models based on these individual parameters exhibit limited predictive efficacy and yield inconsistent results( 9 ). Emerging evidence indicates that inflammatory activity and neovascularization in the perivascular adipose tissue (PVAT) adjacent to the AAA are pivotal in sustaining branch arterial patency and contributing to the persistence of T2ELs( 10 ). In AAA research, computed tomography (CT) attenuation values of PVAT have been correlated with local inflammatory activity and aneurysm progression rates( 7 , 9 ). Nevertheless, their independent predictive value for T2ELs remains uncertain, and attenuation values alone may be inadequate in capturing the complex heterogeneity of perianeurysmal tissue( 11 ). Radiomics, an emerging field that decodes tissue heterogeneity by extracting high-dimensional quantitative features from medical images, presents a promising solution( 12 ). This approach can both characterize the lesion itself( 13 ) and decipher critical information from the perilesional microenvironment related to disease progression( 8 ). Its utility has been demonstrated in assessing peritumoral environments for prognostic prediction and identifying coronary perivascular inflammation for cardiovascular risk stratification( 14 ). This study seeks to extend this paradigm to AAA. It proposes a novel methodological framework for the systematic extraction of radiomic features not only from the aneurysm sac but also from concentric perivascular tissue regions at multiple radial distances from the outer wall. This strategy is designed to capture the spatial biological heterogeneity of the perianeurysmal environment( 15 ). We hypothesize the following: 1) Models incorporating features from the perivascular environment will outperform models based solely on the aneurysm sac, and 2) models integrating the most predictive features from the aneurysm sac and the optimal perianeurysmal region will achieve the highest performance in predicting T2ELs post-EVAR. The purpose of this study is to develop and validate a non-invasive, robust predictive tool that could enhance risk stratification and personalize management for patients with AAA undergoing EVAR by incorporating multiregional perianeurysmal radiomics. Materials and Methods 2.1 Study Population This retrospective study was approved by the Institutional Review Board (Approval No. : [2025(I)-314-01], with a waiver for informed consent. We enrolled consecutive patients with AAA who underwent elective EVAR from January 2015 to January 2024 at three centers: Lishui Hospital of Zhejiang University (Center 1), The Second Affiliated Hospital of Wenzhou Medical University (Center 2), and The Third Affiliated Hospital of Wenzhou Medical University (Center 3). In Fig. 1 , a flowchart that illustrates the process of selecting patients is presented. The inclusion criteria were as follows: (a) preoperative CTA performed within 1 month prior to EVAR and (b) availability of follow-up CTA data at least 6 months postoperatively to confirm T2EL status. The detailed imaging criteria applied for the diagnosis of T2ELs are provided in Appendix S5. The exclusion criteria included the following: (a) non-atherosclerotic aneurysms (e.g., aortic dissection or rupture); (b) prior aortic surgery; (c) complex EVAR techniques (e.g., fenestrated or branched stent grafts); (d) follow-up imaging confirming type I, III, IV, or V endoleaks; and (e) CTA slice thickness of > 1 mm or poor image quality precluding analysis( 3 , 16 ). Finally, 148 eligible patients from Center 1 were randomly assigned to a training set (n = 104) and an internal validation set (n = 44) in a 7:3 ratio. A total of 110 patients from Center 2 and 96 patients from Center 3 were utilized as External Test Sets 1 and 2, respectively. Demographic information, clinical characteristics, and morphological aneurysm data at baseline were retrospectively obtained from electronic medical records and preoperative CTA for all included patients. The detailed definitions of all collected variables, including clinical characteristics (e.g., age, body mass index, and comorbidities) and quantitative anatomical aneurysm parameters, are provided in Appendix S1 . 2.2 CTA Data Collection CTA examinations were performed using Siemens SOMATOM Force scanners at Centers 1 and 2 and a Toshiba Aquilion ONE scanner at Center 3. Patients were scanned in the supine position during quiet respiration. The scan encompassed the diaphragmatic dome through the pubic symphysis to guarantee comprehensive assessment of the abdominal aorta and iliac arteries( 7 ). The detailed acquisition and reconstruction parameters for each scanner are provided in Appendix S2 ( Table S3 ). 2.3 Image Segmentation and Feature Extraction Before feature extraction, all images were resampled to a consistent isotropic voxel size of 1 × 1 × 1 mm 3 employing B-spline interpolation. Grayscale strength was standardized to a range of 0–255 and subsequently underwent Z-score normalization to reduce inter-scanner intensity discrepancies. Two seasoned radiologists, unaware of clinical outcomes, independently executed image segmentation utilizing 3D Slicer (version 5.0.3). The aneurysm boundary was initially delineated slice by slice by two attending radiologists, with 5 and 10 years of experience in vascular imaging, respectively, to define the aneurysm sac volume of interest (VOI). Voxels within a range of − 195 to − 45 Hounsfield units were included to primarily capture adipose tissue. Subsequently, five concentric perianeurysmal VOIs were generated using the “Margin” tool in 3D Slicer via perianeurysmal adipose tissue defined by thresholding at distances of 2, 4, 6, 8, and 10 mm. An example of AAA and PVAT segmentation is illustrated in Figure S1 . After segmentation, radiomic characteristics were collected from each VOI using the PyRadiomics software package. For each VOI (the aneurysm sac and each of the five perianeurysmal regions), a standardized set of 939 radiomic features was extracted per patient, encompassing shape, first-order statistics, and texture feature classes. All extracted features subsequently underwent Z-score normalization to ensure comparability across the cohort( 17 ). Radiomic feature extraction is outlined in Appendix S3 . Intra-class correlation coefficients (ICCs) were computed for both within- and between-observer variations using segmentations from the two radiologists, with specifics outlined in Appendix S6 , to evaluate feature stability. Only characteristics with an ICC greater than 0.80 were preserved for further radiomic investigation( 18 ). 2.4 Feature Selection and Model Construction Feature selection was conducted in three sequential steps to prevent overfitting and identify the most relevant predictors: 1) retention of features demonstrating a statistically significant difference between the T2EL and non-T2EL groups using independent two-sample t-tests; 2) removal of highly redundant features via Spearman correlation analysis (if |r| > 0.9 for any feature pair, only one was retained); and 3) subsequent dimensionality reduction and identification of the most predictive characteristics utilizing a least absolute shrinkage and selection operator regression model with five-fold cross-validation. Based on the selected feature subsets, six distinct prediction models were constructed and compared using a support vector machine (SVM) classifier. The specific hyperparameters and implementation details of the SVM are provided in Appendix S8 . These included one model based solely on intra-sac features (Intra model) and five fusion models (collectively termed Intra + Peri models), each created by combining sac features with features from one specific perianeurysmal region (2, 4, 6, 8, or 10 mm). All models were built using the training set with identical hyperparameter settings, and their performance was evaluated on the independent validation sets( 16 , 19 ). 2.5 Model Development, Evaluation, and Interpretability Clinical and combined models were constructed using logistic regression and the SVM, respectively. In developing the clinical model, variables having a P value below 0.1 in the single-variable analysis were included in the subsequent multivariable logistic regression analysis. The radiomic models, comprising the Intra model and five Intra + Peri fusion models, were constructed using the SVM, with the optimal radiomic signature later amalgamated with clinical predictors to create the comprehensive combination model. All prediction models underwent thorough evaluation across the training, internal validation, and two external validation cohorts. Key performance parameters, including area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), were calculated. Calibration curves were utilized to evaluate predictive accuracy, whereas decision curve analysis (DCA) was applied to measure clinical beneficial effects across likelihood thresholds. Model interpretation was improved via SHapley Additive exPlanations (SHAP) analysis to elucidate feature contributions. The technique for the radiomic workflow of this investigation is illustrated in Fig. 2 . 2.6 Statistical Analysis Statistical analyses were performed using SPSS (version 27.0.1.0), R software (version 4.2.2), and MedCalc (version 22.001). Normally distributed continuous variables were expressed as means ± standard deviations, while non-normally distributed variables were reported as medians (interquartile ranges). Categorical variables were presented as frequencies (percentages). Continuous variables were compared between groups using independent two-sample t-tests, whereas categorical variables were analyzed using the chi-square test or Fisher’s exact test, as appropriate. Statistical significance was defined as a two-sided P value of less than 0.05( 19 , 20 ). Results 3.1 Patient Characteristics and Clinical Predictors The clinical and imaging features of the participants are delineated in Table 2 , and the cohort analysis is provided in Table S1 . The training set included 104 patients, while the internal validation set contained 44 patients; the two external validation sets comprised 110 and 96 patients, respectively. In all cohorts, the baseline demographics and comorbidities, such as sex, age, previous smoking, hypertension, and hyperlipidemia, exhibited no significant differences between the participants with and without T2ELs. However, significant differences were noted in specific imaging parameters: The participants who developed T2ELs had a larger maximum AAA diameter but a smaller thrombus area than those without T2ELs. The univariate and multivariate logistic regression analyses were performed to identify the independent clinical predictors of T2ELs in the training set ( Table 1 ). The maximum AAA diameter (95% confidence interval [CI] = 1.03–1.27, P = 0.009) and the thrombus area (95% CI = 0.55–0.89, P = 0.047) were found to be separate predictors in the multivariate analysis. This suggests that patients with a larger aneurysm diameter and a smaller thrombus area are at a higher risk for postoperative T2ELs. Based on these two variables, a clinical predictive model was established for the subsequent comparative analysis. 3.2 Radiomic Model Construction Following feature selection, six distinct radiomic models were constructed for comparative evaluation: the Intra model and the five Intra + Peri models (Intra + Peri 2 mm to Intra + Peri 10 mm). The feature importance coefficients for the six radiomic models after LASSO regression are presented in Figure S2. The correlation among the selected features for each model is visualized in the heatmaps provided in Figure 3 . The specific features retained after the selection process are detailed in Table S2 . The predictive performance of the six radiomic models, alongside the clinical model, was rigorously evaluated across all cohorts. The comprehensive performance parameters, including AUC, accuracy, sensitivity, specificity, PPV, and NPV, are succinctly described in Table 3 . Figure 4a displays the comparative receiver operating characteristic (ROC) curves for all models. 3.3 Comparative Model Performance and Optimal Model Selection The comparative analysis indicated that the Intra + Peri 6 mm model exhibited the demonstrated good discrimination in prediction. In the training set, it attained an AUC of 0.910 (95% CI = 0.845–0.960), markedly surpassing the standalone Intra model (AUC = 0.782) and the clinical model (AUC = 0.709). This performance advantage was consistently maintained across all validation cohorts, with AUCs of 0.907 (internal validation), 0.886 (External Validation 1), and 0.859 (External Validation 2). A multimetric radar chart visualization for the training cohort ( Figure. 4b ) further confirmed the balanced excellence of the model, showing consistently high values across all evaluated metrics—accuracy, sensitivity, specificity, PPV, and NPV—indicating its robust performance profile beyond discriminatory power alone. In contrast, the models incorporating features from the smaller (2 and 4 mm) and larger (8 and 10 mm) perianeurysmal regions exhibited lower and more variable AUCs ( Figure. 4a ), suggesting that the 6-mm region optimally captures the imaging heterogeneity most relevant to T2EL pathogenesis. Thus, the Intra + Peri 6 mm radiomic signature was selected as the superior radiomic model for further integration with clinical predictions. 3.4 Comprehensive Combined Model Performance and Clinical Utility The optimal Intra + Peri 6 mm radiomic signature was integrated with the independent clinical predictors (maximum aneurysm diameter and thrombus area) to build the combined model and thereby leverage multimodal information. The combined model achieved better performance, surpassing both the standalone clinical model and the optimal radiomic model ( Table 3 ). The model exhibited an accuracy of 92.3%, a sensitivity of 89.7%, and a specificity of 93.3% in the training set, with an AUC of 0.954 (95% CI = 0.913–0.986). Its generalizability was exceptional, as evidenced by AUCs of 0.933, 0.924, and 0.896 in the internal validation and two external validation sets, respectively. The Radar chart ( Figure. 4b ) visually affirmed the superior discriminatory ability of the combined model across all datasets. The clinical efficacy of the models was assessed via the DCA. As depicted in Figure 4c , the combined model exhibited a greater net beneficial effect across a wide range of threshold probabilities in comparison with the independent clinical model, the Intra + Peri 6 mm radiomic model, and the approaches of treating all or no patients. This indicates that employing the combined model for preoperative decision-making would lead to better clinical outcomes by more accurately identifying patients who would truly benefit from intensified surveillance or intervention while avoiding unnecessary procedures for low-risk patients. 3.5 Model Interpretation and Visualization All features were examined using the SHAP analysis to improve the interpretability of the combined model. The results were visualized as an integrated SHAP summary plot combining feature importance ranking and beeswarm distribution ( Figure. 5a ), a SHAP decision plot ( Figure. 5b ), and a SHAP heatmap ( Figure. 5c ). Both the importance ranking and beeswarm plot identified the intra‑aneurysmal radiomic feature Intra_square_glcm_Idn, the 6‑mm perianeurysmal feature Peri6mm_logarithm_glszm_LargeAreaHighGrayLevelEmphasis and Peri6mm_gradient_glcm_Correlation as the top three contributors for predicting T2ELs post-EVAR. The waterfall plot of a representative patient shown in Figure 5d further illustrates how individual features contribute incrementally to the final prediction probability. Figure 6 displays the radiomic feature heatmaps of the key predictive features overlaid on the corresponding CTA images for two representative patients in the training cohort—one with and one without a T2EL. The visualization highlights distinct spatial heterogeneity patterns within the PVAT and intra-sac, providing a visual correlation between radiomic signatures and clinical outcomes. Discussion Precise preoperative classification of patients at an elevated risk for T2ELs is essential for customizing postoperative monitoring and facilitating early intervention, thus improving the long-term effectiveness of EVAR( 21 ). This study leveraged preoperative CTA to quantitatively evaluate the predictive value of radiomic features derived from both the aneurysm sac and perianeurysmal regions of varying extents. Our findings demonstrate that a radiomic signature fusing features from the aneurysm sac and the PVAT within a 6-mm margin from the aneurysm wall yields optimal and stable predictive performance. The combined model, which further integrates this optimal radiomic signature with key clinical predictors (maximum aneurysm diameter and thrombus area), demonstrates robust predictive performance across multiple validation cohorts. In recent years, radiomic analysis of PVAT has garnered significant attention in the field of vascular diseases. It enables non-invasive characterization of the biological state of the perivascular microenvironment, offering a novel perspective for predicting the progression after EVAR. Previous studies have established a link between PVAT features and AAA outcomes. Huang et al. created a CTA-based radiomic model to forecast post-EVAR sac development, designating the PVAT region of interest as adipose tissue within 10 mm of the aneurysm wall. Their combined model showed optimal performance, indicating that radiomic PVAT features correlate with EVAR prognosis. Nevertheless, their study did not investigate whether this 10-mm boundary was optimal, nor was it specifically designed for predicting T2ELs( 22 ). Other research has also found that PVAT imaging characteristics within 5-mm perianeurysmal rims can yield biological information related to inflammation and progression( 6 ). However, no previous research has systematically examined the additional predictive value of perianeurysmal radiomics at varying radial extents for T2ELs after EVAR in patients with AAA. This study extracted imaging features at the three-dimensional level and, for the first time, systematically derived features from multiple consecutive distances ranging from 2 to 10 mm outside the aneurysm wall. The study then constructed a series of Intra + Peri fusion models. Unlike approaches relying solely on a single region, our work incorporated both intra-aneurysmal and perianeurysmal biological information, which enabled the identification of the most predictive spatial extent and consequently improved the model performance. Notably, the model was tested using an internal validation set as well as two separate external validation sets to determine its generalizability. Our findings confirm that the incorporation of PVAT features in radiomic models yields superior predictive value compared with the Intra model. The Intra + Peri 6 mm model attained an AUC of 0.910 in the training cohort and sustained AUCs exceeding 0.907 and 0.859 in the internal and external validation cohorts, respectively. This model exhibited optimal and consistent performance across all datasets. The 6-mm zone likely represents a critical biological interface where inflammatory processes from the aneurysm wall extend into the surrounding fat, promoting collateral vessel formation and thereby enriching radiomic signatures relevant to T2EL pathogenesis. In contrast, proximal perianeurysmal zones (e.g., 2–4 mm) are susceptible to feature instability due to partial-volume effects and wall contamination, whereas distal zones (e.g., 8–10 mm) risk diluting the specific radiomic signal by including extraneous anatomical noise. Together, these observations underscore the importance of carefully defining the perianeurysmal region of interest for extracting discriminative imaging biomarkers. The Intra + Peri 6 mm model thus provides a more reliable tool for preoperative risk stratification than models based on the aneurysm intra-sac or on other radial distances. A set of 15 radiomic features was selected for the final model. Twelve features originated from the intra-sac region, while three were derived from the 6-mm perianeurysmal adipose tissue. The majority of these features represented texture-based and gray-level co-occurrence matrix-derived metrics, capturing subtle spatial heterogeneity within the aneurysm intra-sac and perivascular environment that is not discernible through conventional visual assessment. This class of features offers enhanced sensitivity to microstructural variations that may reflect underlying pathophysiological processes. Among the selected features, Peri_6mm_exponential_glszm_ZoneVariance, extracted from the PVAT, emerged as particularly representative. This measure, derived from the gray-level size zone matrix through exponential transformation, assesses variations in the dimensions of homogeneous patches in the image. Elevated values indicate pronounced spatial heterogeneity in perianeurysmal adipose tissue, which may correspond to disordered distributions of inflammatory foci, fibrotic areas, or neovascular clusters. This microstructural disorder aligns with a more active inflammatory environment and has been previously linked to worse clinical outcomes, hence reinforcing the involvement of PVAT in the pathophysiology of T2ELs. Our multivariate analysis revealed the maximum aneurysm diameter and thrombus size as the independent predictors of T2ELs, which were then utilized to develop the clinical model. A larger aneurysm diameter typically implies more complex geometry and a greater number of potential patent side branches( 23 – 25 ), whereas a smaller thrombus area may reflect a more hemodynamically active sac environment with reduced thrombogenic propensity( 19 , 26 ), which are both established risk factors for T2ELs. Finally, we integrated the optimally selected radiomic feature set from the Intra + Peri 6 mm model with these clinical predictors to develop the combined predictive model. This integrated model demonstrated excellent performance across all cohorts. Specifically, in the training set, it achieved an AUC of 0.954 (95% CI = 0.913–0.986), an accuracy of 92.3%, a sensitivity of 89.7%, and a specificity of 93.3%. This performance was maintained in the internal validation set (AUC = 0.933) and the two external validation sets (AUC = 0.924 and 0.896, respectively), hence affirming the robust generalizability of the model. Furthermore, the DCA demonstrated that the combined model exhibited enhanced calibration curve and provided a superior net clinical benefit across a broad spectrum of probability thresholds when compared with the radiomic or clinical model used independently. These results not only validate the value of radiomic features in characterizing both the perianeurysmal microenvironment and intra-luminal heterogeneity but also indicate that their combination with key clinical variables enables a more comprehensive and accurate individualized risk assessment. The SHAP analysis was employed to graphically illustrate the direction and size of each predictive feature’s contribution to the T2EL risk, hence enhancing model interpretability. The SHAP framework underscores the significance and impact of principal predictors within the integrated model. Through case-based illustrations, we demonstrated how specific features contribute to individual risk assessments. The derived Shapley values were subsequently used to compute personalized prediction probabilities. This approach renders the model a non-invasive tool that couples high predictive accuracy with transparent interpretability, thereby facilitating the preoperative identification of high-risk patients and guiding tailored postoperative surveillance or preventive interventions. This study possesses several constraints that must be recognized when interpreting the findings. First, the retrospective design may have led to biased selections. Prospective, multicenter trials are warranted to validate the predictive performance and clinical applicability of our model in broader patient populations. Second, the radiomic features were derived solely from preoperative arterial-phase CTA. Future studies could incorporate multiphasic or dynamic CTA acquisitions, which may capture hemodynamic patterns and further improve predictive accuracy. Third, manual segmentation, although yielding great spatial precision, is laborious and prone to inter-observer variability. The prospective integration of automated or semi-automatic segmentation techniques into the workflow might enhance the repeatability and effectiveness of radiomic feature extraction. Conclusion This multicenter research developed and proved the value of a non-invasive prediction instrument for persistent T2ELs following EVAR using perianeurysmal radiomics alongside clinical indicators. We found that the models including 6-mm PVAT characteristics surpassed those based exclusively on the aneurysm sac, with the combined model attaining the maximum predictive efficacy. This technique can assist in identifying high-risk patients for customized surveillance while minimizing wasteful operations in low-risk individuals. Abbreviations AAA Abdominal Aortic Aneurysm T2EL Type II Endoleak EVAR Endovascular Aortic Repair CTA Computed Tomography Angiography PVAT Perivascular Adipose Tissue VOI Volume of Interest ICC Intra-class Correlation Coefficient LASSO Least Absolute Shrinkage and Selection Operator SVM Support Vector Machine AUC Area Under the Curve CI Confidence Interval DCA Decision Curve Analysis SHAP SHapley Additive exPlanations PPV Positive Predictive Value NPV Negative Predictive Value ROC Receiver Operating Characteristic GLCM Gray Level Co-occurrence Matrix GLSZM Gray Level Size Zone Matrix BMI Body Mass Index OR Odds Ratio Declarations Ethical Approval This study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). Institutional Review Board approval was obtained from the Second Affiliated Hospital of Wenzhou Medical University, the Third Affiliated Hospital of Wenzhou Medical University, and the Fifth Affiliated Hospital of Wenzhou Medical University (Approval No. 2025(I)-314-01). Written informed consent was waived by the ethics committees due to the retrospective nature of this study. 2.Consent for publication All authors named in this manuscript consented to its publication and take full responsibility for its content. All patients whose CTA images are used in the manuscript figures have signed informed consent forms and agreed to publication. All authors read and approved the final manuscript. 3.Availability of data and materials The datasets used and/or analyzed in the current study are available from the corresponding author upon reasonable request. Conflict of Interest The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article. Funding This research was funded by the Key Project of Joint Construction by Zhejiang Medicine and Health Science and Technology Project (Grant No.2026KY1522 to Feipeng Lin). Author Contributions Conception and design of the work: Xinlei Yu, Guihan Lin, Weiyue Chen, Weiqian Chen, Weiming Hu; Data collection: Weiming Hu, Cheng Ma, Zhuohang Shi, Jinhong Sun, Jie Zhang, Jianhua Wu, Xiongying Yi, Hua Yang; Imaging segmentation and radiomic feature extraction: Xinlei Yu, Guihan Lin, Lei Xu, Yongjun Chen; Data analysis: Xinlei Yu, Guihan Lin, Weiyue Chen, Changsheng Shi; Data interpretation: Guihan Lin, Weiyue Chen, Feipeng Lin, Weiqian Chen, Donglin Li; Drafting the manuscript: Xinlei Yu, Weiming Hu, Jianhua Wu, Guihan Lin; Critical revision of the manuscript: Guihan Lin, Weiyue Chen, Weiqian Chen, Donglin Li; Final approval of the version to be published: Weiqian Chen,Guihan Lin, Weiyue Chen, Donglin Li, Jiansong Ji. All authors have read and agreed to the published version of the manuscript. Statistics and Biometry One of the authors has significant statistical expertise in the design and implementation of the statistical analysis for this study. Informed Consent Written informed consent was waived by the Institutional Review Board due to the retrospective nature of the study and full anonymization of patient clinical and imaging data. Study subjects or cohorts overlap Some study subjects or cohorts have not been previously reported. Methodology Methodology: retrospective observational multicenter study References Debono S, Tzolos E, Syed MBJ, Nash J, Fletcher AJ, Dweck MR, et al. CT Attenuation of Periaortic Adipose Tissue in Abdominal Aortic Aneurysms. Radiology: Cardiothoracic Imaging. 2024;6(1). Miceli F, Dajci A, Di Girolamo A, Nardis P, Ascione M, Cangiano R, et al. Early and Mid-Term Outcomes of Isolated Type 2 Endoleak Refractory to an Embolization Procedure. Journal of Clinical Medicine. 2025;14(2). Charalambous S, Klontzas ME, Kontopodis N, Ioannou CV, Perisinakis K, Maris TG, et al. Radiomics and machine learning to predict aggressive type 2 endoleaks after endovascular aneurysm repair: a proof of concept. Acta Radiologica. 2021;63(9):1293-9. Liu H, Mo X, Zhang F, Qing K-X, Xiao Y. Numerical study on the biomechanics mechanism of Type II endoleak after EVAR for Abdominal Aortic Aneurysm. PLOS One. 2025;20(5). Liu B, Tang X, He N, Chen Z. Development and validation of a nomogram to predict the risk of type II endoleak after endovascular aneurysm repair. Frontiers in Cardiovascular Medicine. 2025;12. Ginzburg D, Nowak S, Attenberger U, Luetkens J, Sprinkart AM, Kuetting D. Computer tomography-based assessment of perivascular adipose tissue in patients with abdominal aortic aneurysms. Scientific Reports. 2024;14(1). Lv R, Hu G, Zhang S, Zhang Z, Chen J, Wang K, et al. Assessing abdominal aortic aneurysm growth using radiomic features of perivascular adipose tissue after endovascular repair. Insights into Imaging. 2024;15(1). Zhang S, Chang N, Yu X, Kang B, Tan R. Association between perivascular fat density on CT angiography and abdominal aortic aneurysm progression. BMC Medical Imaging. 2025;25(1). Yamaguchi M, Yonetsu T, Hoshino M, Sugiyama T, Kanaji Y, Yasui Y, et al. Clinical Significance of Increased Computed Tomography Attenuation of Periaortic Adipose Tissue in Patients With Abdominal Aortic Aneurysms. Circulation Journal. 2021;85(12):2172-80. Grodecki K, Geers J, Kwiecinski J, Lin A, Slipczuk L, Slomka PJ, et al. Phenotyping atherosclerotic plaque and perivascular adipose tissue: signalling pathways and clinical biomarkers in atherosclerosis. Nature Reviews Cardiology. 2025;22(6):443-55. Gaibazzi N, Sartorio D, Tuttolomondo D, Napolitano F, Siniscalchi C, Borrello B, et al. Attenuation of peri-vascular fat at computed tomography to measure inflammation in ascending aorta aneurysms. European Journal of Preventive Cardiology. 2020;28(8):e23-e5. Zhang W, Zhang W, Li X, Cao X, Yang G, Zhang H. Predicting Tumor Perineural Invasion Status in High-Grade Prostate Cancer Based on a Clinical–Radiomics Model Incorporating T2-Weighted and Diffusion-Weighted Magnetic Resonance Images. Cancers. 2022;15(1). Nadendla RR, Settipalli S, Bagchi S, Das G, Chakraborty T, Hasan T, et al. Perivascular Adipose Tissue: Implications for Cardiometabolic Diseases. Biomedical and Pharmacology Journal. 2025;18(3):1765-74. Zhang S, Yu X, Gu H, Kang B, Guo N, Wang X. Identification of high-risk carotid plaque by using carotid perivascular fat density on computed tomography angiography. European Journal of Radiology. 2022;150. Mamopoulos AT, Freyhardt P, Touloumtzidis A, Zapenko A, Katoh M, Gäbel G. Quantification of periaortic adipose tissue in contrast-enhanced CT angiography: technical feasibility and methodological considerations. The International Journal of Cardiovascular Imaging. 2022;38(7):1621-33. Mo J, Liu Q, Wang K, Huang L, Yao C. Prediction of persistent type II endoleak after endovascular aortic repair using machine learning based on preoperative clinical data and radiomic. Vascular Investigation and Therapy. 2025;8(1):31-8. Liu H-F, Wang M, Wang Q, Lu Y, Lu Y-J, Sheng Y, et al. Multiparametric MRI-based intratumoral and peritumoral radiomics for predicting the pathological differentiation of hepatocellular carcinoma. Insights into Imaging. 2024;15(1). Kauffmann C, Tang A, Therasse É, Giroux M-F, Elkouri S, Melanson P, et al. Measurements and detection of abdominal aortic aneurysm growth: Accuracy and reproducibility of a segmentation software. European Journal of Radiology. 2012;81(8):1688-94. Wang Y, Zhou M, Ding Y, Li X, Zhou Z, Shi Z, et al. Development and Comparison of Multimodal Models for Preoperative Prediction of Outcomes After Endovascular Aneurysm Repair. Frontiers in Cardiovascular Medicine. 2022;9. Lareyre F, Guzzi L, Nasr B, Alouane A, Goffart S, Chierici A, et al. Imaging Characterisation of Peripheral Artery Disease: A Scoping Review on Current Classifications and New Insights Brought by Artificial Intelligence. EJVES Vascular Forum. 2025;64:87-95. Kaladji A, Daoudal A, Duménil A, Göksu C, Cardon A, Clochard E, et al. Predictive Models of Complications after Endovascular Aortic Aneurysm Repair. Annals of Vascular Surgery. 2017;40:19-27. Huang S, Liu D, Deng K, Shu C, Wu Y, Zhou Z. A computed tomography angiography-based radiomics model for prognostic prediction of endovascular abdominal aortic repair. International Journal of Cardiology. 2025;429. Kouvelos GN, Spanos K, Nana P, Koutsias S, Rousas N, Giannoukas A, et al. Large Diameter (≥29 mm) Proximal Aortic Necks Are Associated with Increased Complication Rates after Endovascular Repair for Abdominal Aortic Aneurysm. Annals of Vascular Surgery. 2019;60:70-5. Montelione N, Sirignano P, d'Adamo A, Stilo F, Mansour W, Capoccia L, et al. Comparison of Outcomes Following EVAR Based on Aneurysm Diameter and Volume and Their Postoperative Variations. Annals of Vascular Surgery. 2021;74:183-93. Kitagawa A, Mastracci TM, von Allmen R, Powell JT. The role of diameter versus volume as the best prognostic measurement of abdominal aortic aneurysms. Journal of Vascular Surgery. 2013;58(1):258-65. Riveros F, Martufi G, Gasser TC, Rodriguez-Matas JF. On the Impact of Intraluminal Thrombus Mechanical Behavior in AAA Passive Mechanics. Annals of Biomedical Engineering. 2015;43(9):2253-64. Tables Tables 1 to 3 are available in the supplementary files section Additional Declarations No competing interests reported. Supplementary Files supplementmaterials.docx tables.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 06 Apr, 2026 Reviewers agreed at journal 05 Apr, 2026 Reviewers invited by journal 02 Apr, 2026 Editor invited by journal 06 Mar, 2026 Editor assigned by journal 05 Mar, 2026 Submission checks completed at journal 05 Mar, 2026 First submitted to journal 03 Mar, 2026 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-9018408","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":618345598,"identity":"4e22a1d9-c014-488d-b090-edee4c7b8fd0","order_by":0,"name":"Xinlei Yu","email":"","orcid":"","institution":"Lishui Hospital, School of Medicine, Zhejiang University","correspondingAuthor":false,"prefix":"","firstName":"Xinlei","middleName":"","lastName":"Yu","suffix":""},{"id":618345599,"identity":"9ce0c603-de67-49f8-83f6-7004fd7e69f2","order_by":1,"name":"Guihan Lin","email":"","orcid":"","institution":"Lishui Hospital, School of Medicine, Zhejiang University","correspondingAuthor":false,"prefix":"","firstName":"Guihan","middleName":"","lastName":"Lin","suffix":""},{"id":618345600,"identity":"19222cfb-ed03-4a17-9ab4-f1af505d9100","order_by":2,"name":"Weiyue Chen","email":"","orcid":"","institution":"Lishui Hospital, School of Medicine, Zhejiang University","correspondingAuthor":false,"prefix":"","firstName":"Weiyue","middleName":"","lastName":"Chen","suffix":""},{"id":618345601,"identity":"ca08ddd2-4b21-411f-8a44-b9597b5d7881","order_by":3,"name":"Weiming Hu","email":"","orcid":"","institution":"Lishui Central Hospital, The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Weiming","middleName":"","lastName":"Hu","suffix":""},{"id":618345602,"identity":"230165e5-5ddc-47ef-8aa7-5b3b139b4f57","order_by":4,"name":"Cheng Ma","email":"","orcid":"","institution":"Lishui Central Hospital, The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Cheng","middleName":"","lastName":"Ma","suffix":""},{"id":618345603,"identity":"d404a50c-d270-4ba3-bbf4-f4f2fe15f0a5","order_by":5,"name":"Zhuohang Shi","email":"","orcid":"","institution":"Lishui Central Hospital, The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zhuohang","middleName":"","lastName":"Shi","suffix":""},{"id":618345604,"identity":"d9c7fcde-c930-4296-8301-f1ad438efe01","order_by":6,"name":"Feipeng Lin","email":"","orcid":"","institution":"Lishui Central Hospital, The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Feipeng","middleName":"","lastName":"Lin","suffix":""},{"id":618345605,"identity":"0c941da5-8650-4e84-9cd7-c258ab09a0c6","order_by":7,"name":"Jinhong Sun","email":"","orcid":"","institution":"Lishui Central Hospital, The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jinhong","middleName":"","lastName":"Sun","suffix":""},{"id":618345606,"identity":"d23af74f-cd0b-423a-a84a-f67311337632","order_by":8,"name":"Jie Zhang","email":"","orcid":"","institution":"Lishui Central Hospital, The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jie","middleName":"","lastName":"Zhang","suffix":""},{"id":618345607,"identity":"a99d22fe-5aeb-418d-9704-20a710ee3ecb","order_by":9,"name":"Jianhua Wu","email":"","orcid":"","institution":"Lishui Central Hospital, The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jianhua","middleName":"","lastName":"Wu","suffix":""},{"id":618345608,"identity":"e09da293-b527-4436-bfd7-09fefe15aa17","order_by":10,"name":"Xiongying Yi","email":"","orcid":"","institution":"Lishui Central Hospital, The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Xiongying","middleName":"","lastName":"Yi","suffix":""},{"id":618345609,"identity":"e18a5ce1-c2ad-4b59-8017-85dedc2e16c0","order_by":11,"name":"Hua Yang","email":"","orcid":"","institution":"Lishui Central Hospital, The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Hua","middleName":"","lastName":"Yang","suffix":""},{"id":618345610,"identity":"635b885b-e2f3-447c-be5f-b3291310c3a5","order_by":12,"name":"Lei Xu","email":"","orcid":"","institution":"The Second Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Lei","middleName":"","lastName":"Xu","suffix":""},{"id":618345611,"identity":"b0437279-571b-48fa-9c5e-6ebd8a8473de","order_by":13,"name":"Changsheng Shi","email":"","orcid":"","institution":"Ruian People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Changsheng","middleName":"","lastName":"Shi","suffix":""},{"id":618345612,"identity":"f408801a-bff6-45f9-bd7b-c9814253fc62","order_by":14,"name":"Yongjun Chen","email":"","orcid":"","institution":"Lishui People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yongjun","middleName":"","lastName":"Chen","suffix":""},{"id":618345613,"identity":"e0207d07-3985-49c5-a96c-6a5d68fdb92c","order_by":15,"name":"Jiansong Ji","email":"","orcid":"","institution":"Lishui Hospital, School of Medicine, Zhejiang University","correspondingAuthor":false,"prefix":"","firstName":"Jiansong","middleName":"","lastName":"Ji","suffix":""},{"id":618345614,"identity":"67c92734-2df5-429f-8038-4f20582478ae","order_by":16,"name":"Donglin Li","email":"","orcid":"","institution":"The First Affiliated Hospital of Zhejiang University,School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Donglin","middleName":"","lastName":"Li","suffix":""},{"id":618345615,"identity":"87b8cc7e-7d38-4336-9acf-0026f968e519","order_by":17,"name":"Weiqian Chen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzklEQVRIiWNgGAWjYBACPgbmAwcSKiTk+NkbGx98IEYLGwNb4oEHZ2yMJXsONxvOIE4Lj/HBh21piRtupLdJcxClhf+AwYHEtsPGkjMfNkgzMNjJ6TYQ0iKRkHAg4dxhOX7pxAbjAoZkY7MDBLUwAL1fBrRldmJD8gyGA4nbCGrhP9hwIIHtcOKGmwcbDvMQpYUhmeFAAtj7jI3NxGmRSANqAQdyYjPjDAMi/MLPf/7zxx/gqDz+/MeHCjs5glrQgAFpykfBKBgFo2AU4AAAFOZKUcaNQYsAAAAASUVORK5CYII=","orcid":"","institution":"Lishui Hospital, School of Medicine, Zhejiang University","correspondingAuthor":true,"prefix":"","firstName":"Weiqian","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2026-03-03 09:12:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9018408/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9018408/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106725495,"identity":"0896352e-12e9-459d-b984-f8363547c2ef","added_by":"auto","created_at":"2026-04-12 18:33:00","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":625124,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of Patient Selection in This Multicenter Study. EVAR, endovascular aortic repair; CTA, computed tomography angiography; T2EL, type II endoleak.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-9018408/v1/598996ca3887261bc785d7bd.png"},{"id":106725498,"identity":"830b72c8-b225-4d7a-97f0-e8450d44cb0a","added_by":"auto","created_at":"2026-04-12 18:33:02","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":5025520,"visible":true,"origin":"","legend":"\u003cp\u003eDetailed Flowchart Including CTA Scanning, VOI Segmentation With Perianeurysmal Region Dilation, Feature Extraction and Selection, and Radiomic Model Construction and Evaluation.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-9018408/v1/187b511e66a073db4ced3a2f.png"},{"id":106725333,"identity":"fb23454a-4e73-409c-8be9-7b983cfbb030","added_by":"auto","created_at":"2026-04-12 18:32:29","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":5950336,"visible":true,"origin":"","legend":"\u003cp\u003eFeature Correlation Heatmaps. Pairwise Spearman correlation coefficients of radiomic features selected via LASSO regression are shown for each of the six VOIs: aneurysm sac and peri-sac regions at 2-, 4-, 6-, 8-, and 10-mm distances.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-9018408/v1/11f38c14080bbf2bd9009390.png"},{"id":106725381,"identity":"89c2050f-b349-4b6f-8e30-e46529503551","added_by":"auto","created_at":"2026-04-12 18:32:40","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":8186333,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance Comparison of the Prediction Models. (a) ROC curves for the eight distinct prediction models evaluating type II endoleak risk across the training, internal validation, External Test 1, and External Test 2 cohorts. (b) Radar chart visualization of predictive performance metrics for the eight models in the four cohorts. (c) Calibration curves for the clinical model, radiomic model, and SVM combined model across the four cohorts. ROC, receiver operating characteristic; AUC, area under the curve.\u003c/p\u003e","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-9018408/v1/b27814b41bad0b403e0e696b.png"},{"id":106725702,"identity":"b4713e2f-2f63-4612-aa25-5ec62d89d6e0","added_by":"auto","created_at":"2026-04-12 18:33:31","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":6913502,"visible":true,"origin":"","legend":"\u003cp\u003eModel Interpretability Analysis Using SHAP. (a) Integrated SHAP summary plot combining feature importance ranking (left bar chart) with beeswarm distribution of individual SHAP values (right panel, red: positive impact, blue: negative impact). (b) SHAP decision plot illustrating how features cumulatively drive the prediction probability from the baseline. (c) SHAP heatmap showing the direction and magnitude of feature impacts across the entire cohort. (d) SHAP waterfall plot for a representative patient, detailing how each feature contributes incrementally to the final predicted probability of T2ELs. SHAP, SHapley Additive exPlanations.\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-9018408/v1/1a3d3f3aec2d8a08461d023e.png"},{"id":106725383,"identity":"57b9f5d7-6cc5-4155-9ded-78625753aa58","added_by":"auto","created_at":"2026-04-12 18:32:41","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":6954656,"visible":true,"origin":"","legend":"\u003cp\u003eExamples of Radiomic Feature Visualization in Patients With or Without T2ELs. (a) A 68‑year‑old man with a persistent T2EL after EVAR. The left image shows the preoperative arterial‑phase CTA scan at the maximum axial plane of the AAA. The middle image displays the corresponding segmented aneurysm sac (red contour) and the 6‑mm perianeurysmal adipose region (green contour). The right image presents the heatmap of the key perianeurysmal radiomic feature exponential_glszm_ZoneVariance, with red areas reflecting higher feature values (greater heterogeneity) and blue areas indicating lower values (more homogeneous tissue). (b) A 65‑year‑old man in the non‑T2EL group, with no endoleak detected during follow‑up. The images in (b) correspond to the same representations shown in (a).\u003c/p\u003e","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-9018408/v1/689e2330eef7de81ea9c828a.png"},{"id":106727241,"identity":"ec79a211-adf2-48c7-be71-040c61c92cc2","added_by":"auto","created_at":"2026-04-12 18:38:25","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":41961325,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9018408/v1/e91eea94-e21d-4cb4-a30a-b292437bbae4.pdf"},{"id":106725698,"identity":"95e445bc-d6ec-46c4-8429-19dde21e87d3","added_by":"auto","created_at":"2026-04-12 18:33:30","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":8451938,"visible":true,"origin":"","legend":"","description":"","filename":"supplementmaterials.docx","url":"https://assets-eu.researchsquare.com/files/rs-9018408/v1/05890fede5053c7da232fbd0.docx"},{"id":106725390,"identity":"1ee8076a-cd8c-49b6-9671-b7de81598cb6","added_by":"auto","created_at":"2026-04-12 18:32:42","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":40679,"visible":true,"origin":"","legend":"","description":"","filename":"tables.docx","url":"https://assets-eu.researchsquare.com/files/rs-9018408/v1/ba5c0b61c457e5e924b5828c.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Preoperative Prediction of Persistent Type II Endoleaks After Endovascular Aortic Repair Using Multiregional Perianeurysmal Computed Tomography Angiography Radiomics: A Multicenter Study","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAbdominal aortic aneurysm (AAA) is a condition with a significant life-threatening potential, characterized by a regional dilation of the abdominal aorta with a diagnostic threshold diameter of \u0026ge;\u0026thinsp;3.0 cm(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). Its principal risk stems from asymptomatic expansion and eventual rupture, with rupture mortality rates exceeding 80%(\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e). Endovascular aortic repair (EVAR) has become a mainstream, minimally invasive intervention for AAA(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). Nevertheless, its long-term efficacy is frequently compromised by endoleaks, which are the continuous flow of blood inside the aneurysm intra-sac that is external to the stent graft. Among them, type II endoleaks (T2ELs) are the most prevalent subtype, characterized by retrograde perfusion of the sac from patent branch arteries, such as the inferior mesenteric or lumbar arteries. Studies indicate that persistent T2ELs maintain pressurization within the sac, elevating the risks of continued sac expansion and rupture(\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). Therefore, accurately identifying patients at a high risk for persistent T2ELs before surgery is of critical importance. It enables the development of tailored surveillance protocols and timely prophylactic interventions, thereby improving the long-term success of EVAR(\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe major imaging method for both the preliminary assessment and postoperative surveillance of AAA is computed tomography angiography (CTA), which provides in-depth anatomical information(\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). Current preoperative prediction of T2ELs relies predominantly on specific anatomical factors derived from CTA, including aneurysm neck angulation and the number of patent lumbar arteries(\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). However, models based on these individual parameters exhibit limited predictive efficacy and yield inconsistent results(\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). Emerging evidence indicates that inflammatory activity and neovascularization in the perivascular adipose tissue (PVAT) adjacent to the AAA are pivotal in sustaining branch arterial patency and contributing to the persistence of T2ELs(\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e). In AAA research, computed tomography (CT) attenuation values of PVAT have been correlated with local inflammatory activity and aneurysm progression rates(\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). Nevertheless, their independent predictive value for T2ELs remains uncertain, and attenuation values alone may be inadequate in capturing the complex heterogeneity of perianeurysmal tissue(\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eRadiomics, an emerging field that decodes tissue heterogeneity by extracting high-dimensional quantitative features from medical images, presents a promising solution(\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). This approach can both characterize the lesion itself(\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e) and decipher critical information from the perilesional microenvironment related to disease progression(\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). Its utility has been demonstrated in assessing peritumoral environments for prognostic prediction and identifying coronary perivascular inflammation for cardiovascular risk stratification(\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). This study seeks to extend this paradigm to AAA. It proposes a novel methodological framework for the systematic extraction of radiomic features not only from the aneurysm sac but also from concentric perivascular tissue regions at multiple radial distances from the outer wall. This strategy is designed to capture the spatial biological heterogeneity of the perianeurysmal environment(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). We hypothesize the following: 1) Models incorporating features from the perivascular environment will outperform models based solely on the aneurysm sac, and 2) models integrating the most predictive features from the aneurysm sac and the optimal perianeurysmal region will achieve the highest performance in predicting T2ELs post-EVAR. The purpose of this study is to develop and validate a non-invasive, robust predictive tool that could enhance risk stratification and personalize management for patients with AAA undergoing EVAR by incorporating multiregional perianeurysmal radiomics.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Study Population\u003c/h2\u003e \u003cp\u003eThis retrospective study was approved by the Institutional Review Board (Approval No. : [2025(I)-314-01], with a waiver for informed consent. We enrolled consecutive patients with AAA who underwent elective EVAR from January 2015 to January 2024 at three centers: Lishui Hospital of Zhejiang University (Center 1), The Second Affiliated Hospital of Wenzhou Medical University (Center 2), and The Third Affiliated Hospital of Wenzhou Medical University (Center 3). In Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, a flowchart that illustrates the process of selecting patients is presented. The inclusion criteria were as follows: (a) preoperative CTA performed within 1 month prior to EVAR and (b) availability of follow-up CTA data at least 6 months postoperatively to confirm T2EL status. The detailed imaging criteria applied for the diagnosis of T2ELs are provided in Appendix S5. The exclusion criteria included the following: (a) non-atherosclerotic aneurysms (e.g., aortic dissection or rupture); (b) prior aortic surgery; (c) complex EVAR techniques (e.g., fenestrated or branched stent grafts); (d) follow-up imaging confirming type I, III, IV, or V endoleaks; and (e) CTA slice thickness of \u0026gt;\u0026thinsp;1 mm or poor image quality precluding analysis(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e). Finally, 148 eligible patients from Center 1 were randomly assigned to a training set (n\u0026thinsp;=\u0026thinsp;104) and an internal validation set (n\u0026thinsp;=\u0026thinsp;44) in a 7:3 ratio. A total of 110 patients from Center 2 and 96 patients from Center 3 were utilized as External Test Sets 1 and 2, respectively.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eDemographic information, clinical characteristics, and morphological aneurysm data at baseline were retrospectively obtained from electronic medical records and preoperative CTA for all included patients. The detailed definitions of all collected variables, including clinical characteristics (e.g., age, body mass index, and comorbidities) and quantitative anatomical aneurysm parameters, are provided in \u003cb\u003eAppendix S1\u003c/b\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 CTA Data Collection\u003c/h2\u003e \u003cp\u003eCTA examinations were performed using Siemens SOMATOM Force scanners at Centers 1 and 2 and a Toshiba Aquilion ONE scanner at Center 3. Patients were scanned in the supine position during quiet respiration. The scan encompassed the diaphragmatic dome through the pubic symphysis to guarantee comprehensive assessment of the abdominal aorta and iliac arteries(\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). The detailed acquisition and reconstruction parameters for each scanner are provided in \u003cb\u003eAppendix S2\u003c/b\u003e (\u003cb\u003eTable S3\u003c/b\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Image Segmentation and Feature Extraction\u003c/h2\u003e \u003cp\u003eBefore feature extraction, all images were resampled to a consistent isotropic voxel size of 1 \u0026times; 1 \u0026times; 1 mm\u003csup\u003e3\u003c/sup\u003e employing B-spline interpolation. Grayscale strength was standardized to a range of 0\u0026ndash;255 and subsequently underwent Z-score normalization to reduce inter-scanner intensity discrepancies.\u003c/p\u003e \u003cp\u003eTwo seasoned radiologists, unaware of clinical outcomes, independently executed image segmentation utilizing 3D Slicer (version 5.0.3). The aneurysm boundary was initially delineated slice by slice by two attending radiologists, with 5 and 10 years of experience in vascular imaging, respectively, to define the aneurysm sac volume of interest (VOI). Voxels within a range of \u0026minus;\u0026thinsp;195 to \u0026minus;\u0026thinsp;45 Hounsfield units were included to primarily capture adipose tissue. Subsequently, five concentric perianeurysmal VOIs were generated using the \u0026ldquo;Margin\u0026rdquo; tool in 3D Slicer via perianeurysmal adipose tissue defined by thresholding at distances of 2, 4, 6, 8, and 10 mm. An example of AAA and PVAT segmentation is illustrated in \u003cb\u003eFigure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e.\u003c/p\u003e \u003cp\u003eAfter segmentation, radiomic characteristics were collected from each VOI using the PyRadiomics software package. For each VOI (the aneurysm sac and each of the five perianeurysmal regions), a standardized set of 939 radiomic features was extracted per patient, encompassing shape, first-order statistics, and texture feature classes. All extracted features subsequently underwent Z-score normalization to ensure comparability across the cohort(\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). Radiomic feature extraction is outlined in \u003cb\u003eAppendix S3\u003c/b\u003e. Intra-class correlation coefficients (ICCs) were computed for both within- and between-observer variations using segmentations from the two radiologists, with specifics outlined in \u003cb\u003eAppendix S6\u003c/b\u003e, to evaluate feature stability. Only characteristics with an ICC greater than 0.80 were preserved for further radiomic investigation(\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Feature Selection and Model Construction\u003c/h2\u003e \u003cp\u003eFeature selection was conducted in three sequential steps to prevent overfitting and identify the most relevant predictors: 1) retention of features demonstrating a statistically significant difference between the T2EL and non-T2EL groups using independent two-sample t-tests; 2) removal of highly redundant features via Spearman correlation analysis (if |r| \u0026gt; 0.9 for any feature pair, only one was retained); and 3) subsequent dimensionality reduction and identification of the most predictive characteristics utilizing a least absolute shrinkage and selection operator regression model with five-fold cross-validation.\u003c/p\u003e \u003cp\u003eBased on the selected feature subsets, six distinct prediction models were constructed and compared using a support vector machine (SVM) classifier. The specific hyperparameters and implementation details of the SVM are provided in \u003cb\u003eAppendix S8\u003c/b\u003e. These included one model based solely on intra-sac features (Intra model) and five fusion models (collectively termed Intra\u0026thinsp;+\u0026thinsp;Peri models), each created by combining sac features with features from one specific perianeurysmal region (2, 4, 6, 8, or 10 mm). All models were built using the training set with identical hyperparameter settings, and their performance was evaluated on the independent validation sets(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Model Development, Evaluation, and Interpretability\u003c/h2\u003e \u003cp\u003eClinical and combined models were constructed using logistic regression and the SVM, respectively. In developing the clinical model, variables having a \u003cem\u003eP\u003c/em\u003e value below 0.1 in the single-variable analysis were included in the subsequent multivariable logistic regression analysis. The radiomic models, comprising the Intra model and five Intra\u0026thinsp;+\u0026thinsp;Peri fusion models, were constructed using the SVM, with the optimal radiomic signature later amalgamated with clinical predictors to create the comprehensive combination model.\u003c/p\u003e \u003cp\u003eAll prediction models underwent thorough evaluation across the training, internal validation, and two external validation cohorts. Key performance parameters, including area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), were calculated. Calibration curves were utilized to evaluate predictive accuracy, whereas decision curve analysis (DCA) was applied to measure clinical beneficial effects across likelihood thresholds. Model interpretation was improved via SHapley Additive exPlanations (SHAP) analysis to elucidate feature contributions. The technique for the radiomic workflow of this investigation is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Statistical Analysis\u003c/h2\u003e \u003cp\u003eStatistical analyses were performed using SPSS (version 27.0.1.0), R software (version 4.2.2), and MedCalc (version 22.001). Normally distributed continuous variables were expressed as means\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviations, while non-normally distributed variables were reported as medians (interquartile ranges). Categorical variables were presented as frequencies (percentages). Continuous variables were compared between groups using independent two-sample t-tests, whereas categorical variables were analyzed using the chi-square test or Fisher\u0026rsquo;s exact test, as appropriate. Statistical significance was defined as a two-sided \u003cem\u003eP\u003c/em\u003e value of less than 0.05(\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003e3.1 Patient Characteristics and Clinical Predictors\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe clinical and imaging features of the participants are delineated in \u003cstrong\u003eTable 2\u003c/strong\u003e, and the cohort analysis is provided in \u003cstrong\u003eTable S1\u003c/strong\u003e. The training set included 104 patients, while the internal validation set contained 44 patients; the two external validation sets comprised 110 and 96 patients, respectively. In all cohorts, the baseline demographics and comorbidities, such as sex, age, previous smoking, hypertension, and hyperlipidemia, exhibited no significant differences between the participants with and without T2ELs. However, significant differences were noted in specific imaging parameters: The participants who developed T2ELs had a larger maximum AAA diameter but a smaller thrombus area than those without T2ELs.\u003c/p\u003e\n\u003cp\u003eThe univariate and multivariate logistic regression analyses were performed to identify the independent clinical predictors of T2ELs in the training set (\u003cstrong\u003eTable 1\u003c/strong\u003e). The maximum AAA diameter (95% confidence interval [CI] = 1.03–1.27, \u003cem\u003eP\u003c/em\u003e = 0.009) and the thrombus area (95% CI = 0.55–0.89, \u003cem\u003eP\u003c/em\u003e = 0.047) were found to be separate predictors in the multivariate analysis. This suggests that patients with a larger aneurysm diameter and a smaller thrombus area are at a higher risk for postoperative T2ELs. Based on these two variables, a clinical predictive model was established for the subsequent comparative analysis.\u003c/p\u003e\n\u003ch3\u003e3.2 Radiomic Model Construction\u003c/h3\u003e\n\u003cp\u003eFollowing feature selection, six distinct radiomic models were constructed for comparative evaluation: the Intra model and the five Intra + Peri models (Intra + Peri 2 mm to Intra + Peri 10 mm). The feature importance coefficients for the six radiomic models after LASSO regression are presented in Figure S2. The correlation among the selected features for each model is visualized in the heatmaps provided in \u003cstrong\u003eFigure 3\u003c/strong\u003e. The specific features retained after the selection process are detailed in \u003cstrong\u003eTable S2\u003c/strong\u003e. The predictive performance of the six radiomic models, alongside the clinical model, was rigorously evaluated across all cohorts. The comprehensive performance parameters, including AUC, accuracy, sensitivity, specificity, PPV, and NPV, are succinctly described in \u003cstrong\u003eTable 3\u003c/strong\u003e.\u003cstrong\u003e\u0026nbsp;Figure 4a\u0026nbsp;\u003c/strong\u003edisplays the comparative receiver operating characteristic (ROC) curves for all models.\u003c/p\u003e\n\u003ch3\u003e3.3 Comparative Model Performance and Optimal Model Selection\u003c/h3\u003e\n\u003cp\u003eThe comparative analysis indicated that the Intra + Peri 6 mm model exhibited the demonstrated good discrimination in prediction. In the training set, it attained an AUC of 0.910 (95% CI = 0.845–0.960), markedly surpassing the standalone Intra model (AUC = 0.782) and the clinical model (AUC = 0.709). This performance advantage was consistently maintained across all validation cohorts, with AUCs of 0.907 (internal validation), 0.886 (External Validation 1), and 0.859 (External Validation 2).\u003c/p\u003e\n\u003cp\u003eA multimetric radar chart visualization for the training cohort (\u003cstrong\u003eFigure. 4b\u003c/strong\u003e) further confirmed the balanced excellence of the model, showing consistently high values across all evaluated metrics—accuracy, sensitivity, specificity, PPV, and NPV—indicating its robust performance profile beyond discriminatory power alone. In contrast, the models incorporating features from the smaller (2 and 4 mm) and larger (8 and 10 mm) perianeurysmal regions exhibited lower and more variable AUCs (\u003cstrong\u003eFigure. 4a\u003c/strong\u003e), suggesting that the 6-mm region optimally captures the imaging heterogeneity most relevant to T2EL pathogenesis. Thus, the Intra + Peri 6 mm radiomic signature was selected as the superior radiomic model for further integration with clinical predictions.\u003c/p\u003e\n\u003ch3\u003e3.4 Comprehensive Combined Model Performance and Clinical Utility\u003c/h3\u003e\n\u003cp\u003eThe optimal Intra + Peri 6 mm radiomic signature was integrated with the independent clinical predictors (maximum aneurysm diameter and thrombus area) to build the\u0026nbsp;combined model and thereby leverage multimodal information. The combined model achieved better performance, surpassing both the standalone clinical model and the optimal radiomic model (\u003cstrong\u003eTable 3\u003c/strong\u003e). The model exhibited an accuracy of 92.3%, a sensitivity of 89.7%, and a specificity of 93.3% in the training set, with an AUC of 0.954 (95% CI = 0.913–0.986). Its generalizability was exceptional, as evidenced by AUCs of 0.933, 0.924, and 0.896 in the internal validation and two external validation sets, respectively. The Radar chart (\u003cstrong\u003eFigure. 4b\u003c/strong\u003e) visually affirmed the superior discriminatory ability of the combined model across all datasets.\u003c/p\u003e\n\u003cp\u003eThe clinical efficacy of the models was assessed via the DCA. As depicted in \u003cstrong\u003eFigure 4c\u003c/strong\u003e, the combined model exhibited a greater net beneficial effect across a wide range of threshold probabilities in comparison with the independent clinical model, the Intra + Peri 6 mm radiomic model, and the approaches of treating all or no patients. This indicates that employing the combined model for preoperative decision-making would lead to better clinical outcomes by more accurately identifying patients who would truly benefit from intensified surveillance or intervention while avoiding unnecessary procedures for low-risk patients.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.5 Model Interpretation and Visualization\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll features were examined using the SHAP analysis to improve the interpretability of the combined model. The results were visualized as an integrated SHAP summary plot combining feature importance ranking and beeswarm distribution (\u003cstrong\u003eFigure. 5a\u003c/strong\u003e), a SHAP decision plot (\u003cstrong\u003eFigure. 5b\u003c/strong\u003e), and a SHAP heatmap (\u003cstrong\u003eFigure. 5c\u003c/strong\u003e). Both the importance ranking and beeswarm plot identified the intra‑aneurysmal radiomic feature\u0026nbsp;Intra_square_glcm_Idn, the 6‑mm perianeurysmal feature\u0026nbsp;Peri6mm_logarithm_glszm_LargeAreaHighGrayLevelEmphasis and\u0026nbsp;Peri6mm_gradient_glcm_Correlation as the top three contributors for predicting T2ELs post-EVAR. The waterfall plot of a representative patient shown in \u003cstrong\u003eFigure 5d\u0026nbsp;\u003c/strong\u003efurther illustrates how individual features contribute incrementally to the final prediction probability.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 6\u003c/strong\u003e displays the radiomic feature heatmaps of the key predictive features overlaid on the corresponding CTA images for two representative patients in the training cohort—one with and one without a T2EL. The visualization highlights distinct spatial heterogeneity patterns within the PVAT and intra-sac, providing a visual correlation between radiomic signatures and clinical outcomes.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003ePrecise preoperative classification of patients at an elevated risk for T2ELs is essential for customizing postoperative monitoring and facilitating early intervention, thus improving the long-term effectiveness of EVAR(\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). This study leveraged preoperative CTA to quantitatively evaluate the predictive value of radiomic features derived from both the aneurysm sac and perianeurysmal regions of varying extents. Our findings demonstrate that a radiomic signature fusing features from the aneurysm sac and the PVAT within a 6-mm margin from the aneurysm wall yields optimal and stable predictive performance. The combined model, which further integrates this optimal radiomic signature with key clinical predictors (maximum aneurysm diameter and thrombus area), demonstrates robust predictive performance across multiple validation cohorts.\u003c/p\u003e \u003cp\u003eIn recent years, radiomic analysis of PVAT has garnered significant attention in the field of vascular diseases. It enables non-invasive characterization of the biological state of the perivascular microenvironment, offering a novel perspective for predicting the progression after EVAR. Previous studies have established a link between PVAT features and AAA outcomes. Huang et al. created a CTA-based radiomic model to forecast post-EVAR sac development, designating the PVAT region of interest as adipose tissue within 10 mm of the aneurysm wall. Their combined model showed optimal performance, indicating that radiomic PVAT features correlate with EVAR prognosis. Nevertheless, their study did not investigate whether this 10-mm boundary was optimal, nor was it specifically designed for predicting T2ELs(\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e). Other research has also found that PVAT imaging characteristics within 5-mm perianeurysmal rims can yield biological information related to inflammation and progression(\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). However, no previous research has systematically examined the additional predictive value of perianeurysmal radiomics at varying radial extents for T2ELs after EVAR in patients with AAA.\u003c/p\u003e \u003cp\u003eThis study extracted imaging features at the three-dimensional level and, for the first time, systematically derived features from multiple consecutive distances ranging from 2 to 10 mm outside the aneurysm wall. The study then constructed a series of Intra\u0026thinsp;+\u0026thinsp;Peri fusion models. Unlike approaches relying solely on a single region, our work incorporated both intra-aneurysmal and perianeurysmal biological information, which enabled the identification of the most predictive spatial extent and consequently improved the model performance. Notably, the model was tested using an internal validation set as well as two separate external validation sets to determine its generalizability.\u003c/p\u003e \u003cp\u003eOur findings confirm that the incorporation of PVAT features in radiomic models yields superior predictive value compared with the Intra model. The Intra\u0026thinsp;+\u0026thinsp;Peri 6 mm model attained an AUC of 0.910 in the training cohort and sustained AUCs exceeding 0.907 and 0.859 in the internal and external validation cohorts, respectively. This model exhibited optimal and consistent performance across all datasets. The 6-mm zone likely represents a critical biological interface where inflammatory processes from the aneurysm wall extend into the surrounding fat, promoting collateral vessel formation and thereby enriching radiomic signatures relevant to T2EL pathogenesis. In contrast, proximal perianeurysmal zones (e.g., 2\u0026ndash;4 mm) are susceptible to feature instability due to partial-volume effects and wall contamination, whereas distal zones (e.g., 8\u0026ndash;10 mm) risk diluting the specific radiomic signal by including extraneous anatomical noise. Together, these observations underscore the importance of carefully defining the perianeurysmal region of interest for extracting discriminative imaging biomarkers. The Intra\u0026thinsp;+\u0026thinsp;Peri 6 mm model thus provides a more reliable tool for preoperative risk stratification than models based on the aneurysm intra-sac or on other radial distances.\u003c/p\u003e \u003cp\u003eA set of 15 radiomic features was selected for the final model. Twelve features originated from the intra-sac region, while three were derived from the 6-mm perianeurysmal adipose tissue. The majority of these features represented texture-based and gray-level co-occurrence matrix-derived metrics, capturing subtle spatial heterogeneity within the aneurysm intra-sac and perivascular environment that is not discernible through conventional visual assessment. This class of features offers enhanced sensitivity to microstructural variations that may reflect underlying pathophysiological processes. Among the selected features, Peri_6mm_exponential_glszm_ZoneVariance, extracted from the PVAT, emerged as particularly representative. This measure, derived from the gray-level size zone matrix through exponential transformation, assesses variations in the dimensions of homogeneous patches in the image. Elevated values indicate pronounced spatial heterogeneity in perianeurysmal adipose tissue, which may correspond to disordered distributions of inflammatory foci, fibrotic areas, or neovascular clusters. This microstructural disorder aligns with a more active inflammatory environment and has been previously linked to worse clinical outcomes, hence reinforcing the involvement of PVAT in the pathophysiology of T2ELs.\u003c/p\u003e \u003cp\u003eOur multivariate analysis revealed the maximum aneurysm diameter and thrombus size as the independent predictors of T2ELs, which were then utilized to develop the clinical model. A larger aneurysm diameter typically implies more complex geometry and a greater number of potential patent side branches(\u003cspan additionalcitationids=\"CR24\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e), whereas a smaller thrombus area may reflect a more hemodynamically active sac environment with reduced thrombogenic propensity(\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e), which are both established risk factors for T2ELs. Finally, we integrated the optimally selected radiomic feature set from the Intra\u0026thinsp;+\u0026thinsp;Peri 6 mm model with these clinical predictors to develop the combined predictive model. This integrated model demonstrated excellent performance across all cohorts. Specifically, in the training set, it achieved an AUC of 0.954 (95% CI\u0026thinsp;=\u0026thinsp;0.913\u0026ndash;0.986), an accuracy of 92.3%, a sensitivity of 89.7%, and a specificity of 93.3%. This performance was maintained in the internal validation set (AUC\u0026thinsp;=\u0026thinsp;0.933) and the two external validation sets (AUC\u0026thinsp;=\u0026thinsp;0.924 and 0.896, respectively), hence affirming the robust generalizability of the model. Furthermore, the DCA demonstrated that the combined model exhibited enhanced calibration curve and provided a superior net clinical benefit across a broad spectrum of probability thresholds when compared with the radiomic or clinical model used independently. These results not only validate the value of radiomic features in characterizing both the perianeurysmal microenvironment and intra-luminal heterogeneity but also indicate that their combination with key clinical variables enables a more comprehensive and accurate individualized risk assessment.\u003c/p\u003e \u003cp\u003eThe SHAP analysis was employed to graphically illustrate the direction and size of each predictive feature\u0026rsquo;s contribution to the T2EL risk, hence enhancing model interpretability. The SHAP framework underscores the significance and impact of principal predictors within the integrated model. Through case-based illustrations, we demonstrated how specific features contribute to individual risk assessments. The derived Shapley values were subsequently used to compute personalized prediction probabilities. This approach renders the model a non-invasive tool that couples high predictive accuracy with transparent interpretability, thereby facilitating the preoperative identification of high-risk patients and guiding tailored postoperative surveillance or preventive interventions.\u003c/p\u003e \u003cp\u003eThis study possesses several constraints that must be recognized when interpreting the findings. First, the retrospective design may have led to biased selections. Prospective, multicenter trials are warranted to validate the predictive performance and clinical applicability of our model in broader patient populations. Second, the radiomic features were derived solely from preoperative arterial-phase CTA. Future studies could incorporate multiphasic or dynamic CTA acquisitions, which may capture hemodynamic patterns and further improve predictive accuracy. Third, manual segmentation, although yielding great spatial precision, is laborious and prone to inter-observer variability. The prospective integration of automated or semi-automatic segmentation techniques into the workflow might enhance the repeatability and effectiveness of radiomic feature extraction.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis multicenter research developed and proved the value of a non-invasive prediction instrument for persistent T2ELs following EVAR using perianeurysmal radiomics alongside clinical indicators. We found that the models including 6-mm PVAT characteristics surpassed those based exclusively on the aneurysm sac, with the combined model attaining the maximum predictive efficacy. This technique can assist in identifying high-risk patients for customized surveillance while minimizing wasteful operations in low-risk individuals.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAAA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAbdominal Aortic Aneurysm\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eT2EL\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eType II Endoleak\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eEVAR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eEndovascular Aortic Repair\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCTA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eComputed Tomography Angiography\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePVAT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePerivascular Adipose Tissue\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eVOI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eVolume of Interest\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eICC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eIntra-class Correlation Coefficient\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLASSO\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLeast Absolute Shrinkage and Selection Operator\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSVM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSupport Vector Machine\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eArea Under the Curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eConfidence Interval\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDCA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDecision Curve Analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSHAP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSHapley Additive exPlanations\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePPV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePositive Predictive Value\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNPV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNegative Predictive Value\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eReceiver Operating Characteristic\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGLCM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGray Level Co-occurrence Matrix\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGLSZM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGray Level Size Zone Matrix\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eBMI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eBody Mass Index\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eOR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eOdds Ratio\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003col\u003e\n \u003cli\u003e\u003cstrong\u003eEthical Approval\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eThis study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). Institutional Review Board approval was obtained from the Second Affiliated Hospital of Wenzhou Medical University, the Third Affiliated Hospital of Wenzhou Medical University, and the Fifth Affiliated Hospital of Wenzhou Medical University (Approval No. 2025(I)-314-01). Written informed consent was waived by the ethics committees due to the retrospective nature of this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.Consent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors named in this manuscript consented to its publication and take full responsibility for its content. All patients whose CTA images are used in the manuscript figures have signed informed consent forms and agreed to publication. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.Availability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets used and/or analyzed in the current study are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003col start=\"4\"\u003e\n \u003cli\u003e\u003cstrong\u003eConflict of Interest\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eThe authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article.\u003c/p\u003e\n\u003col start=\"5\"\u003e\n \u003cli\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eThis research was funded by the Key Project of Joint Construction by Zhejiang Medicine and Health Science and Technology Project (Grant No.2026KY1522 to Feipeng Lin).\u003c/p\u003e\n\u003col start=\"6\"\u003e\n \u003cli\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eConception and design of the work: Xinlei Yu, Guihan Lin, Weiyue Chen, Weiqian Chen, Weiming Hu;\u003c/p\u003e\n\u003cp\u003eData collection: Weiming Hu, Cheng Ma, Zhuohang Shi, Jinhong Sun, Jie Zhang, Jianhua Wu, Xiongying Yi, Hua Yang;\u003c/p\u003e\n\u003cp\u003eImaging segmentation and radiomic feature extraction: Xinlei Yu, Guihan Lin, Lei Xu, Yongjun Chen;\u003c/p\u003e\n\u003cp\u003eData analysis: Xinlei Yu, Guihan Lin, Weiyue Chen, Changsheng Shi;\u003c/p\u003e\n\u003cp\u003eData interpretation: Guihan Lin, Weiyue Chen, Feipeng Lin, Weiqian Chen, Donglin Li;\u003c/p\u003e\n\u003cp\u003eDrafting the manuscript: Xinlei Yu, Weiming Hu, Jianhua Wu, Guihan Lin;\u003c/p\u003e\n\u003cp\u003eCritical revision of the manuscript: Guihan Lin, Weiyue Chen, Weiqian Chen, Donglin Li;\u003c/p\u003e\n\u003cp\u003eFinal approval of the version to be published: Weiqian Chen,Guihan Lin, Weiyue Chen, Donglin Li, Jiansong Ji.\u003c/p\u003e\n\u003cp\u003eAll authors have read and agreed to the published version of the manuscript.\u003c/p\u003e\n\u003col start=\"7\"\u003e\n \u003cli\u003e\u003cstrong\u003eStatistics and Biometry\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eOne of the authors has significant statistical expertise in the design and implementation of the statistical analysis for this study.\u003c/p\u003e\n\u003col start=\"8\"\u003e\n \u003cli\u003e\u003cstrong\u003eInformed Consent\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eWritten informed consent was waived by the Institutional Review Board due to the retrospective nature of the study and full anonymization of patient clinical and imaging data.\u003c/p\u003e\n\u003col start=\"9\"\u003e\n \u003cli\u003e\u003cstrong\u003eStudy subjects or cohorts overlap\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eSome study subjects or cohorts have not been previously reported.\u003c/p\u003e\n\u003col start=\"10\"\u003e\n \u003cli\u003e\u003cstrong\u003e\u0026nbsp;Methodology\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eMethodology:\u0026nbsp;\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eretrospective\u003c/li\u003e\n \u003cli\u003eobservational\u003c/li\u003e\n \u003cli\u003emulticenter study\u003c/li\u003e\n\u003c/ul\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eDebono S, Tzolos E, Syed MBJ, Nash J, Fletcher AJ, Dweck MR, et al. CT Attenuation of Periaortic Adipose Tissue in Abdominal Aortic Aneurysms. Radiology: Cardiothoracic Imaging. 2024;6(1).\u003c/li\u003e\n \u003cli\u003eMiceli F, Dajci A, Di Girolamo A, Nardis P, Ascione M, Cangiano R, et al. Early and Mid-Term Outcomes of Isolated Type 2 Endoleak Refractory to an Embolization Procedure. Journal of Clinical Medicine. 2025;14(2).\u003c/li\u003e\n \u003cli\u003eCharalambous S, Klontzas ME, Kontopodis N, Ioannou CV, Perisinakis K, Maris TG, et al. Radiomics and machine learning to predict aggressive type 2 endoleaks after endovascular aneurysm repair: a proof of concept. Acta Radiologica. 2021;63(9):1293-9.\u003c/li\u003e\n \u003cli\u003eLiu H, Mo X, Zhang F, Qing K-X, Xiao Y. Numerical study on the biomechanics mechanism of Type II endoleak after EVAR for Abdominal Aortic Aneurysm. PLOS One. 2025;20(5).\u003c/li\u003e\n \u003cli\u003eLiu B, Tang X, He N, Chen Z. Development and validation of a nomogram to predict the risk of type II endoleak after endovascular aneurysm repair. Frontiers in Cardiovascular Medicine. 2025;12.\u003c/li\u003e\n \u003cli\u003eGinzburg D, Nowak S, Attenberger U, Luetkens J, Sprinkart AM, Kuetting D. Computer tomography-based assessment of perivascular adipose tissue in patients with abdominal aortic aneurysms. Scientific Reports. 2024;14(1).\u003c/li\u003e\n \u003cli\u003eLv R, Hu G, Zhang S, Zhang Z, Chen J, Wang K, et al. Assessing abdominal aortic aneurysm growth using radiomic features of perivascular adipose tissue after endovascular repair. Insights into Imaging. 2024;15(1).\u003c/li\u003e\n \u003cli\u003eZhang S, Chang N, Yu X, Kang B, Tan R. Association between perivascular fat density on CT angiography and abdominal aortic aneurysm progression. BMC Medical Imaging. 2025;25(1).\u003c/li\u003e\n \u003cli\u003eYamaguchi M, Yonetsu T, Hoshino M, Sugiyama T, Kanaji Y, Yasui Y, et al. Clinical Significance of Increased Computed Tomography Attenuation of Periaortic Adipose Tissue in Patients With Abdominal Aortic Aneurysms. Circulation Journal. 2021;85(12):2172-80.\u003c/li\u003e\n \u003cli\u003eGrodecki K, Geers J, Kwiecinski J, Lin A, Slipczuk L, Slomka PJ, et al. Phenotyping atherosclerotic plaque and perivascular adipose tissue: signalling pathways and clinical biomarkers in atherosclerosis. Nature Reviews Cardiology. 2025;22(6):443-55.\u003c/li\u003e\n \u003cli\u003eGaibazzi N, Sartorio D, Tuttolomondo D, Napolitano F, Siniscalchi C, Borrello B, et al. Attenuation of peri-vascular fat at computed tomography to measure inflammation in ascending aorta aneurysms. European Journal of Preventive Cardiology. 2020;28(8):e23-e5.\u003c/li\u003e\n \u003cli\u003eZhang W, Zhang W, Li X, Cao X, Yang G, Zhang H. Predicting Tumor Perineural Invasion Status in High-Grade Prostate Cancer Based on a Clinical\u0026ndash;Radiomics Model Incorporating T2-Weighted and Diffusion-Weighted Magnetic Resonance Images. Cancers. 2022;15(1).\u003c/li\u003e\n \u003cli\u003eNadendla RR, Settipalli S, Bagchi S, Das G, Chakraborty T, Hasan T, et al. Perivascular Adipose Tissue: Implications for Cardiometabolic Diseases. Biomedical and Pharmacology Journal. 2025;18(3):1765-74.\u003c/li\u003e\n \u003cli\u003eZhang S, Yu X, Gu H, Kang B, Guo N, Wang X. Identification of high-risk carotid plaque by using carotid perivascular fat density on computed tomography angiography. European Journal of Radiology. 2022;150.\u003c/li\u003e\n \u003cli\u003eMamopoulos AT, Freyhardt P, Touloumtzidis A, Zapenko A, Katoh M, G\u0026auml;bel G. Quantification of periaortic adipose tissue in contrast-enhanced CT angiography: technical feasibility and methodological considerations. The International Journal of Cardiovascular Imaging. 2022;38(7):1621-33.\u003c/li\u003e\n \u003cli\u003eMo J, Liu Q, Wang K, Huang L, Yao C. Prediction of persistent type II endoleak after endovascular aortic repair using machine learning based on preoperative clinical data and radiomic. Vascular Investigation and Therapy. 2025;8(1):31-8.\u003c/li\u003e\n \u003cli\u003eLiu H-F, Wang M, Wang Q, Lu Y, Lu Y-J, Sheng Y, et al. Multiparametric MRI-based intratumoral and peritumoral radiomics for predicting the pathological differentiation of hepatocellular carcinoma. Insights into Imaging. 2024;15(1).\u003c/li\u003e\n \u003cli\u003eKauffmann C, Tang A, Therasse \u0026Eacute;, Giroux M-F, Elkouri S, Melanson P, et al. Measurements and detection of abdominal aortic aneurysm growth: Accuracy and reproducibility of a segmentation software. European Journal of Radiology. 2012;81(8):1688-94.\u003c/li\u003e\n \u003cli\u003eWang Y, Zhou M, Ding Y, Li X, Zhou Z, Shi Z, et al. Development and Comparison of Multimodal Models for Preoperative Prediction of Outcomes After Endovascular Aneurysm Repair. Frontiers in Cardiovascular Medicine. 2022;9.\u003c/li\u003e\n \u003cli\u003eLareyre F, Guzzi L, Nasr B, Alouane A, Goffart S, Chierici A, et al. Imaging Characterisation of Peripheral Artery Disease: A Scoping Review on Current Classifications and New Insights Brought by Artificial Intelligence. EJVES Vascular Forum. 2025;64:87-95.\u003c/li\u003e\n \u003cli\u003eKaladji A, Daoudal A, Dum\u0026eacute;nil A, G\u0026ouml;ksu C, Cardon A, Clochard E, et al. Predictive Models of Complications after Endovascular Aortic Aneurysm Repair. Annals of Vascular Surgery. 2017;40:19-27.\u003c/li\u003e\n \u003cli\u003eHuang S, Liu D, Deng K, Shu C, Wu Y, Zhou Z. A computed tomography angiography-based radiomics model for prognostic prediction of endovascular abdominal aortic repair. International Journal of Cardiology. 2025;429.\u003c/li\u003e\n \u003cli\u003eKouvelos GN, Spanos K, Nana P, Koutsias S, Rousas N, Giannoukas A, et al. Large Diameter (\u0026ge;29 mm) Proximal Aortic Necks Are Associated with Increased Complication Rates after Endovascular Repair for Abdominal Aortic Aneurysm. Annals of Vascular Surgery. 2019;60:70-5.\u003c/li\u003e\n \u003cli\u003eMontelione N, Sirignano P, d\u0026apos;Adamo A, Stilo F, Mansour W, Capoccia L, et al. Comparison of Outcomes Following EVAR Based on Aneurysm Diameter and Volume and Their Postoperative Variations. Annals of Vascular Surgery. 2021;74:183-93.\u003c/li\u003e\n \u003cli\u003eKitagawa A, Mastracci TM, von Allmen R, Powell JT. The role of diameter versus volume as the best prognostic measurement of abdominal aortic aneurysms. Journal of Vascular Surgery. 2013;58(1):258-65.\u003c/li\u003e\n \u003cli\u003eRiveros F, Martufi G, Gasser TC, Rodriguez-Matas JF. On the Impact of Intraluminal Thrombus Mechanical Behavior in AAA Passive Mechanics. Annals of Biomedical Engineering. 2015;43(9):2253-64.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables 1 to 3 are available in the supplementary files section\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Abdominal aortic aneurysm, Type II endoleak, Radiomics, Perivascular adipose tissue, Machine learning, Predictive model","lastPublishedDoi":"10.21203/rs.3.rs-9018408/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9018408/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground: \u003c/strong\u003ePreoperative identification of patients with persistent type II endoleaks (T2ELs) after endovascular aortic repair (EVAR) for abdominal aortic aneurysm (AAA) could improve individualized management. Current approaches based on conventional anatomical factors show limited predictive accuracy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods: \u003c/strong\u003eConsecutive patients with AAA undergoing EVAR at three centers were retrospectively included. Radiomic characteristics were extracted from preoperative computed tomography angiographic images, including the aneurysm sac and five concentric perianeurysmal zones positioned 2–10 mm from the outer wall. After feature selection, six radiomic models employing a support vector machine classifier were developed and subsequently compared. This optimal radiomic signature was then combined with significant clinical predictors to formulate a combined model. The model’s performance was evaluated using area under the curve (AUC), calibration plots, and decision curve analysis, while its interpretability was assessed using SHapley Additive exPlanations (SHAP) analysis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults: \u003c/strong\u003eThe radiomic model combining features from the aneurysm intra-sac and the 6-mm perianeurysmal region demonstrated superior predictive accuracy, with AUCs of 0.910, 0.907, 0.886, and 0.859 in the training, internal validation, and two external test sets, respectively. The maximum aneurysm diameter and thrombus area were identified as the independent clinical predictors. The combined model further improved discrimination, achieving AUCs of 0.954, 0.933, 0.924, and 0.896 in the corresponding cohorts, along with excellent calibration and clinical net benefit. The SHAP analysis explained its predictions both locally and globally.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion: \u003c/strong\u003eA combined model that merges perianeurysmal radiomic features with essential clinical factors offers a precise and non-invasive approach for preoperative T2EL risk stratification following EVAR, thereby facilitating personalized surveillance protocols.\u003c/p\u003e","manuscriptTitle":"Preoperative Prediction of Persistent Type II Endoleaks After Endovascular Aortic Repair Using Multiregional Perianeurysmal Computed Tomography Angiography Radiomics: A Multicenter Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-09 14:40:28","doi":"10.21203/rs.3.rs-9018408/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-04-06T15:33:54+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"28431566442250364021301287986431787104","date":"2026-04-05T16:06:12+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-02T09:26:06+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-06T18:55:13+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-05T10:28:52+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-05T10:27:06+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Imaging","date":"2026-03-03T09:02:49+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"ce61ce1a-66da-4614-a2e7-e2b5ba3de4c1","owner":[],"postedDate":"April 9th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-09T14:40:36+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-09 14:40:28","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9018408","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9018408","identity":"rs-9018408","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}