Radiomics signature and deep learning signature of intrathrombus and perithrombus for prediction of Malignant Cerebral Edema after Acute Ischemic Stroke: a multicenter CT study

Radiomics signature and deep learning signature of intrathrombus and perithrombus for prediction of Malignant Cerebral Edema after Acute Ischemic Stroke: a multicenter CT 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 Radiomics signature and deep learning signature of intrathrombus and perithrombus for prediction of Malignant Cerebral Edema after Acute Ischemic Stroke: a multicenter CT study Shuhao Wang, Xiaoli Gu, Haiqi Wang, Yangyang Nan, Jia Zhou, Xiaoyu Xu, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6948356/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Malignant cerebral edema (MCE) exerts a detrimental impact on clinical outcomes, underscoring the critical need for the development of more precise and objective predictive models. To accurately assess the predictive ability of radiomics and deep learning (DL) features in intrathrombus and perithrombus regions for the risk of MCE after acute ischemic stroke (AIS). Materials and Methods A retrospective study was conducted, enrolling 406 AIS patients who underwent admission CT before endovascular thrombectomy (EVT). Patients from Center A were randomly divided into training and testing sets at a 7:3 ratio, while those from Center B and Center C formed the external validation cohort. Regions of interest (ROIs) of thrombus and perithrombus were manually delineated and automatically expanded in margin by one pixel. 428 radiomic features were extracted from CT images of intrathrombus and perithrombus regions, and 128 DL features were obtained by inputting these images into a VGG16 architecture. Following features fusion, least absolute shrinkage and selection operator (LASSO) regression was employed for dimensionality reduction. Eleven machine learning classifiers were used for model development. Models’ performance was evaluated using Matthews correlation coefficient (MCC) and area under the receiver operating characteristic curve (AUC), with AUC differences tested using DeLong’s method. Results MCE occurred in 49 patients (12.1%). In the validation cohort, the logistic regression (LR) models demonstrated discriminative performance with perithrombus (LR-peri: MCC = 0.857, AUC = 0.891), intrathrombus, (LR-intra: MCC = 0.328, AUC = 0.626), and combined (LR-combined: MCC = 0.41, AUC = 0.869) models. The LR-combined model exhibited a significantly superior predictive capacity to that of LR-intra ( p < 0.05). Conclusions Perithrombus features can improve the prediction of MCE after AIS, which in turn enables the optimization of medical resource allocation. Clinical relevance statement: Emphasis is placed on the critical significance of radiomics extracted from the area in and around the thrombus in predicting MCE after AIS, which has far-reaching significance for improving patient prognosis. Radiomics Malignant cerebral edema Acute Ischemic Stroke Retrospective studies Computed Tomography Angiography Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Key Points • Machine learning models related to thrombosis can effectively predict the occurrence of MCE after AIS. • The proposed LR-peri radiomics model reached a higher area under the curve (AUC: 0.891, 95% CI: 0.762 - 1.000). • Its application provides a beneficial approach for formulating personalized treatment strategies for patients with AIS. Introduction Stroke ranks as the second leading cause of death globally, and MCE is one of the severe complications, with an incidence rate of approximately 10% [ 1 ]. Cytotoxic edema usually peaks 3 to 4 days after brain injury, but reperfusion of necrotic tissue can cause malignant edema within the first 24 hours. Decompressive craniectomy within 48 hours improves outcomes and reduces mortality in large-scale infarctions, but unnecessary surgery is highly invasive [ 2 – 4 ]. Thus, early and accurate prediction of complications is essential. CT is the first-line imaging modality for stroke patients on admission and can predict ischemic brain tissue progression. Wen X et al. conducted a study on predicting MCE by extracting the CT radiomics features of the middle cerebral artery (MCA) blood supply area from Non-Contrast Computed Tomography (NCCT) images of patients with cerebral infarction. Shi J et al. demonstrated that the combined Alberta stroke program early CT score and net water uptake (ASPECTS–NWU) could serve as a quantitative predictor of MCE after MCA territory large vessel occlusion, with a moderate positive correlation with the grade of brain edema, indicating that quantitative measurements of ASPECT score, net water uptake, and enhancement ratio based on CT imaging are effective predictive factors for MCE [ 5 , 6 ]. Prior studies have primarily focused on the infarct core, with less research focusing on the impact of the culprit thrombus and surrounding tissue on post-stroke edema. In ischemic stroke, disruption of the blood-brain barrier leads to vasogenic edema, hemorrhagic transformation, and increased mortality. This pathological process is influenced by thrombus characteristics, as research indicates that thrombi with low red blood cell content, high fibrin levels, and elevated extracellular DNA are less likely to achieve first-pass recanalization (FPR). Some studies have also confirmed that the serum inflammatory factor levels and BBB disruption after AIS are associated with the occurrence of vasogenic cerebral edema [ 7 ]. It is reasonable to assume that the characteristics of the thrombus and surrounding brain tissue can more precisely reflect the inflammatory response and BBB disruption in post-stroke brain tissue [ 8 ]. Moreover, extracting high-dimensional quantitative radiomic features and deep learning features from medical images to construct machine learning predictive models has its advantages of reducing physician subjective judgment factors and improving accuracy [ 9 – 11 ]. Thrombus and perithrombus radiomic features can predict the origin and prognosis of thrombi. For example, according to our team’s previous research, it was found that 1) thrombus radiomic features could predict the origin and composition of stroke thrombi, and 2) the logistic regression model combining radiomic features from both inside and around the thrombus could effectively assess clinical prognosis after EVT [ 12 , 13 ]. However, our previous studies did not involve deep learning features. DL features refer to high-dimensional data representations automatically extracted by multi-layer neural networks, which can effectively capture complex patterns and structures in the data [ 14 ]. The application of deep learning in the field of stroke covers multiple aspects, from the detection of acute cerebral infarction, lesion segmentation, ASPECTS quantification, to prognostic prediction [ 15 – 18 ]. This study aims to fill the gap in the research on predicting MCE using radiomic and deep learning features of the thrombus and its surrounding areas. It deeply explores the possibility of predicting the development of AIS into MCE based on CT thrombus region features, providing strong support for neurosurgeons to formulate personalized treatment strategies. Materials and methods Patients This retrospective study adhered to the 1964 Declaration of Helsinki and its amendments, and was approved by our local ethics committee. Written informed consent was waived by the Institutional Review Board. We analyzed data from stroke patients admitted to three centers (A: ***, B: ***, and C: ***) between December 2018 and December 2023. Inclusion criteria: (1) Acute stroke due to anterior-circulation Large-vessel occlusion (LVO); (2) Admission non-contrast computed tomography (NCCT) and computed tomography angiography showing visible thrombus; (3) EVT performed immediately after CT; (4) Follow-up images of sufficient quality taken within 24 h post-EVT; (5) Complete demographic and clinical data. Exclusion criteria: (1) Inadequate imaging clarity due to motion or metal artifacts; (2) Incomplete clinical records. Figure 1 shows a flowchart of the patient selection process. The clinical data collected from medical and follow-up records included age, gender, National Institutes of Health Stroke Scale (NIHSS) scores, lesion location, comorbidities (atrial fibrillation, hypertension, hyperlipidemia, diabetes mellitus, and coronary heart disease), and smoking history. MCE was defined as a midline shift of ≥ 5 millimeters accompanied by signs of local cerebral swelling [ 19 , 20 ], whose identification was based on follow-up imaging. CT scan and thrombosis segmentation NCCT and CTA examinations were performed using multi-detector CT scanners from three manufacturers: Philips (Brilliance ICT), Toshiba (Aquilion ONE/PRIME), and Siemens (Somatom Sensation). Prior to thrombus segmentation, each CT image underwent an intensity normalization process as described in our previous studies. ROIs for thrombus were outlined using ITK-SNAP software (Version 3.6.0; [ITK-SNAP Home]( http://www.itksnap.org/pmwiki/pmwiki.php )). Following segmentation of the intrathrombus areas, perithrombus areas were automatically demarcated by expanding the radius of the initial 1-mm ROIs. To ensure segmentation accuracy, thrombi were segmented by two radiology residents (*** and ***) who reached a consensus after consultation, and their work was reviewed by a radiology attending physician (***). Evaluators were blinded to clinical details. Feature extraction, selection, and model building After the identification of regions within and surrounding the thrombus, radiomic features were derived utilizing the PyRadiomics ( https://pypi.org/project/pyradiomic/ ). The NCCT and CTA images of the area inside and around the thrombus were cropped into fully-covered two-dimensional images. The size of each image was adjusted to 224×224 and then input into the VGG16 model. Finally, 107 radiomic features and 32 deep learning features were extracted from the intrathrombus and perithrombus regions on NCCT and CTA images respectively. The features with p < 0.05 were retained by Mann-Whitney U test, and the features with correlation coefficients greater than 0.9 were eliminated according to the Spearman correlation coefficient to achieve dimensionality reduction. After performing feature screening using the LASSO, which included radiomic features and deep learning features, these features were input into machine learning model using 11 different algorithms: Logistic Regression (LR), Naive Bayes (NB), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), Extra-Tree, XG Boost, Light GBM, Gradient Boosting (GB), AdaBoost, and Multilayer Perceptron (MLP). The five-fold cross-validation method was employed to verify the predictive performance of each model. MCC was used to screen optimal algorithm for constructing machine learning models. The performance of these models were evaluated using the ROC curve, along with the AUC, accuracy, precision, specificity, and other metrics. The DeLong test was used to statistically assess the differences in the predictive performance of the machine learning models (intrathrombus, perithrombus, and combined models). The workflow of this study is shown in Fig. 2 . Statistical analysis Clinical characteristics were analyzed by the t-test, Mann–Whitney U test, or chi-squared test, as appropriate. The model’s predictive performance for MCE following AIS were evaluated by conducting ROC curve analysis to calculate metrics like the AUC, sensitivity, specificity, positive and negative predictive values, and DeLong’s test was employed to compare ROC curves and assess the model’s clinical utility via decision curve analysis (DCA). Results Patient characteristics In this study, 406 AIS patients were respectively selected according to the inclusive criteria. Eligible patients from Center A were randomly assigned to the training group (n=178) and the testing group (n=77) at a 7:3 ratio, while those from Centers B and C formed the validation group (n=151). The incidence of MCE after cerebral infarction was 49 cases (12.1%). Table 1 systematically presents the detailed demographic information of the patients in different cohorts classified according to the occurrence of MCE. There was a significant statistical difference in the National Institutes of Health Stroke Scale (NIHSS) score in the training cohort. However, there was no significant statistical difference of NIHSS scores in the testing and validation cohorts. There was also no significant statistical difference among the three groups in terms of other clinical features. Feature extraction and selection After dimensionality reduction, a total of 12 DL features and 4 radiomic features (including 9 CTA features and 7 NCCT features) were used to construct machine learning models based on intrathrombus image features. Regarding the models established based on perithrombus images, they incorporate 14 DL features and 4 radiomic features (including 10 CTA features and 8 NCCT features). Simultaneously, the models developed according to the features of combined are equipped with 24 DL features and 8 radiomics features (including 16 CTA features and 16 NCCT features). Feature selection is shown in Figure 3 . Figure 3 Model Performance and Comparison The performance of eleven classifiers were evaluated based on MCC values. In the training sets, most classifiers (8 in 11) performed well (MCC>0.7). Figure 4a illustrates the MCC results. Figure 4a When evaluating the performance of eleven models in the validation set for the perithrombus models, LR performed the best (AUC: 0.891, 95% CI: 0.762 - 1.000). For the combined models, LR also showed the optimal performance (AUC: 0.869, 95% CI: 0.778 - 0.9618). For the intrathrombus models, SVM performed well (AUC: 0.733, 95% CI: 0.546 - 0.921). Figure 4b shows the AUC results. In view of all factors, LR, which performed the best in the predictive task, was selected to construct the prediction models. For detailed information about these models, please refer to Table 2 and the supplementary materials. Figure 4b In the validation set, the LR-intra model had the lowest AUC value (0.626), showing relatively weak performance in the task. The AUC value of the LR-combine (0.869) was close to that of the LR-peri model. The comparison between LR-intra and LR-combine showed a significant difference ( p = 0.007), while the comparisons between LR-intra and LR-peri, as well as between the LR-combined and the LR-peri, did not show significant differences ( p = 0.063 and p = 0.788). Figure 4c shows the DeLong test results between SVM and LR models of two categories. During the model validation stage, the DCA indicated that the combined model showed a higher net benefit (0.069), while the perithrombus model had a wider effective threshold probability range (0.08-0.97). Specific cases are shown in Figure 5 . Discussion The common cause of AIS is the sudden blockage of the proximal middle cerebral artery or the distal internal carotid artery. The BBB is damaged, leading to excessive water infiltration into brain tissue, with a mortality rate that may be as high as 80% [ 21 , 22 ]. Currently, clinical diagnosis of MCE relies on observing midline shift or brain herniation via CT, which are usually late signs of the disease. The main goal of our research is to develop a model for more early prediction of MCE in AIS patients after EVT. In our study, the perithrombus model exhibited superior predictive capacity in the validation cohort (MCC = 0.857, AUC = 0.891). The model can serve as a tool for early prediction of AIS complications when applied to clinical scenarios, thereby improving the quality of survival of AIS patients after EVT. Previous developed models mainly extracted image features of the brain infarction area of NCCT images or segmented the MCA supply area of NCCT images. However, it is difficult to accurately delineate the infarcted area on CT images, making it hard to be popularized. Zhang et al. constructed a machine learning model based on the brain infarction area of NCCT images to predict the occurrence of MCE [ 23 ]. The model had an AUC of 0.912, showing good predictive ability. Fu et al. evaluated the predictive ability of the IP–NWU value in the middle cerebral artery supply area of NCCT for MCE. The radiomic model had a maximum AUC of 0.96 [ 24 ]. However, these studies ignored the impact of the responsible blood vessels on the prognosis of brain infarction, and there was little research on deep learning features related to the thrombus area. Additionally, Sarioglu O et al. found that thrombus-based radiomic features could effectively predict the first-pass effect (FPE) in patients with AIS [ 25 ]. Similar findings were also reported by XIONG et al [ 26 ]. When a patient develops AIS, the ischemia and hypoxia of brain tissue caused by the thrombus will damage the BBB, leading to the leakage of macromolecules in the plasma into the brain tissue interstitium due to the increased vascular permeability. The inflammatory response caused by the thrombus will further damage the vascular endothelial cells and exacerbate the vasogenic brain edema. Subsequent reperfusion after revascularization may also aggravate brain injury, thus triggering or exacerbating brain edema. The area surrounding a thrombus includes structures such as vascular wall cells and perivascular fat. It can be seen that the radiomics features around the thrombus can effectively predict MCE and this prediction is interpretable. Li et al. conducted a retrospective analysis of studies related to CT scans before EVT [ 27 ]. The research showed that the LR model combining intrathrombus and perithrombus radiomics features was very effective in predicting the prognosis of thrombectomy, with an AUC value as high as 0.87 in the validation cohort. Lu et al. developed a two-stage deep learning model to identify early occult AIS in NCCT [ 28 ]. However, we haven’t seen any relevant research based on the deep learning features of thrombus so far. Inspired by these studies, we constructed a machine learning prediction model for predicting MCE using radiomics and deep learning features extracted from the thrombus and its surrounding areas in NCCT and CTA. Compared with complex models, LR is favored for its statistical simplicity, interpretability, and robust performance in binary classification tasks. Although complex models, like SVM, KNN, RF, Extra Trees, XG Boost, MLP, GB, NB, and AdaBoost, have advantages in handling high-dimensional data, they are prone to overfitting without a large amount of data and careful adjustment. Since the Light GBM model performed poorly in different cohorts of this experiment, which may be related to factors such as the feature distribution of the data, sample size, noise, and the parameter settings of the model, this model should be avoided in future related research. Whereas the consistent performance of LR in the validation cohort confirms its applicability and reliability in clinical diagnosis. Given its effective generalization ability across different datasets, the choice of LR is reasonable. In the validation cohort, compared with using only the intrathrombus area and the combined area, the radiomics features of the perithrombus area significantly improved the predictive ability for MCE after acute cerebral infarction (AUC: 0.891, 95% CI: 0.761 to 1.000). During model validation, the combined model yielded a higher net benefit (0.069) compared to the perithrombus model, which demonstrated a broader effective threshold probability range (0.08–0.97). Clinicians should therefore adopt a dual consideration of net benefit and threshold probability range in clinical decision-making. For instance, integrating the complementary strengths of the combined and perithrombus models may optimize risk stratification compared with the unstable predictive effect of intrathrombus radiomics features, the radiomic features extracted from the perithrombus area play an important role in the prediction model. In this study, except for the significant statistical difference in the NIHSS index of the training cohort, other clinical variables with or without MCE did not show significant statistical differences in either the training cohort or the validation cohort. Although this study has innovation and important findings, it is still restricted by various factors. The retrospective design inherently limits its causal inference ability, and the relatively insufficient sample size weakens the universality and robustness of the conclusions. Differences in contrast agent selection may interfere with the cross-scenario application of the model. Although the reliability of radiomics feature extraction is high, manual drawing of ROIs may introduce human errors and uncertainties. In the future, fully automated technology urgently needs to be introduced for optimization. The exploitation of deep learning to enhance model precision represents a critical trend. Notwithstanding, challenges including limited multi-center validation, inadequate integration of clinical covariates, and the necessity for advanced hybrid modeling techniques persist. Future investigations should prioritize overcoming these barriers to elevate both research quality and clinical translational value. Conclusion The LR model proposed in this study integrates the NCCT and CTA image features of the perithrombus areas after cerebral infarction through machine learning methods, providing a way to predict MCE. This may help clinicians decide earlier whether to perform decompressive craniectomy or adopt other intensive monitoring measures, ultimately benefiting stroke patients. Abbreviations DL Deep learning MCE Malignant cerebral edema AIS Acute ischemic stroke EVT Endovascular thrombectomy ROIs Regions of interest LASSO Least absolute shrinkage and selection operator MCC Matthews correlation coefficient AUC Area under the receiver operating characteristics curve MCA Middle cerebral artery NCCT Non-Contrast Computed Tomography ASPECTS–NWU Alberta stroke program early CT score and net water uptake FPR First-pass recanalization DCA Decision curve analysis Declarations Ethics approval and consent to participate Study approval was obtained from the ethics committee of Sixth People's Hospital Affiliated to Shanghai Jiao Tong University (registration number: 2020-212). The study was performed in accordance with the relevant guidelines and/or regulations including the Declaration of Helsinki. Informed consent to participate was waived for this retrospective cohort study by the ethics committee of Sixth People's Hospital Affiliated to Shanghai Jiao Tong University. Consent for publication Not applicable . Availability of data and materials The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Competing 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 Supported by National Natural Science Foundation of China No.82402364 Authors' contributions Jiang Jingxuan: Conceptualization, Methodology, Funding acquisition, Writing - Review & Editing Wang Shuhao: Writing- Original draft preparation, Data curation, Formal analysis, Gu Xiaoli: Visualization, Investigation, Resources, Project administration. Wang Haiqi: Supervision, Software, Investigation. Nan Yangyang: Software, Validation. Zhou Jia: Data Curation, Software, Investigation. Xu Xiaoyu: Project administration . Wang Chenqing ： Software. Acknowledgements None References Wen X, Hu X, Xiao Y, Chen J. Radiomics analysis for predicting malignant cerebral edema in patients undergoing endovascular treatment for acute ischemic stroke. dir. 2023;29:402–9. https://doi.org/10.4274/dir.2023.221764 . Cheng Y, Wu S, Wang Y, et al. External Validation and Modification of the EDEMA Score for Predicting Malignant Brain Edema After Acute Ischemic Stroke. Neurocrit Care. 2020;32:104–12. https://doi.org/10.1007/s12028-019-00844-y . Dower A, Mulcahy M, Maharaj M et al. (2023) Surgical Decompression for Malignant Cerebral Edema After Ischemic Stroke: Cochrane Review. Stroke 54:. https://doi.org/10.1161/STROKEAHA.122.042260 Cannarsa GJ, Simard JM. Decompressive Craniectomy for Stroke: Who, When, and How. Neurol Clin. 2022;40:321–36. https://doi.org/10.1016/j.ncl.2021.11.009 . Shi J, Wu H, Dong Z, et al. Automated quantitative lesion water uptake in acute stroke is a predictor of malignant cerebral edema. Eur Radiol. 2022;32:2771–80. https://doi.org/10.1007/s00330-021-08443-2 . Stafford R, Chatzidakis S, Kim ISY, et al. Follow-up ASPECTS improves prediction of potentially lethal malignant edema in patients with large middle cerebral artery stroke. J NeuroIntervent Surg. 2025;17:e83–6. https://doi.org/10.1136/jnis-2023-021145 . Wang M, Pang X, Lu H, Wang X. Clinical role of serum histone deacetylase 4 measurement in acute ischemic stroke: Relation to disease risk, severity, and prognosis. Clin Lab Anal. 2022;36:e24372. https://doi.org/10.1002/jcla.24372 . Meng S, Cao H, Huang Y, et al. ASK1-K716R reduces neuroinflammation and white matter injury via preserving blood–brain barrier integrity after traumatic brain injury. J Neuroinflammation. 2023;20:244. https://doi.org/10.1186/s12974-023-02923-6 . Jan Y-T, Tsai P-S, Huang W-H, et al. Machine learning combined with radiomics and deep learning features extracted from CT images: a novel AI model to distinguish benign from malignant ovarian tumors. Insights Imaging. 2023;14:68. https://doi.org/10.1186/s13244-023-01412-x . Sheng L, Yang C, Chen Y, Song B. Machine Learning Combined with Radiomics Facilitating the Personal Treatment of Malignant Liver Tumors. Biomedicines. 2023;12:58. https://doi.org/10.3390/biomedicines12010058 . Yang J, Cao Y, Zhou F, et al. Combined deep-learning MRI-based radiomic models for preoperative risk classification of endometrial endometrioid adenocarcinoma. Front Oncol. 2023;13:1231497. https://doi.org/10.3389/fonc.2023.1231497 . Jiang J, Wei J, Zhu Y, et al. Clot-based radiomics model for cardioembolic stroke prediction with CT imaging before recanalization: a multicenter study. Eur Radiol. 2022;33:970–80. https://doi.org/10.1007/s00330-022-09116-4 . Li M, Zhou J, Sheng K, et al. Radiomics of intrathrombus and perithrombus regions for Post-EVT intracranial hemorrhage risk Prediction: A multicenter CT study. Eur J Radiol. 2024;178:111653. https://doi.org/10.1016/j.ejrad.2024.111653 . Wang Y, Li N, Chen L, et al. Guidelines, Consensus Statements, and Standards for the Use of Artificial Intelligence in Medicine: Systematic Review. J Med Internet Res. 2023;25:e46089. https://doi.org/10.2196/46089 . Lei L, Zhou Y, Guo X, et al. The value of a deep learning image reconstruction algorithm in whole-brain computed tomography perfusion in patients with acute ischemic stroke. Quant Imaging Med Surg. 2023;13:8173–89. https://doi.org/10.21037/qims-23-547 . Hakim A, Christensen S, Winzeck S, et al. Predicting Infarct Core From Computed Tomography Perfusion in Acute Ischemia With Machine Learning: Lessons From the ISLES Challenge. Stroke. 2021;52:2328–37. https://doi.org/10.1161/STROKEAHA.120.030696 . Ayobi A, Davis A, Chang PD, et al. Deep Learning–Based ASPECTS Algorithm Enhances Reader Performance and Reduces Interpretation Time. AJNR Am J Neuroradiol. 2024. https://doi.org/10.3174/ajnr.A8491 . ajnr;ajnr.A8491v2. Herzog L, Kook L, Hamann J, et al. Deep Learning Versus Neurologists: Functional Outcome Prediction in LVO Stroke Patients Undergoing Mechanical Thrombectomy. Stroke. 2023;54:1761–9. https://doi.org/10.1161/STROKEAHA.123.042496 . Jiang Q, Yu S, Dong X, et al. Predictors and Dynamic Nomogram to Determine the Individual Risk of Malignant Brain Edema After Endovascular Thrombectomy in Acute Ischemic Stroke. J Clin Neurol. 2022;18:298. https://doi.org/10.3988/jcn.2022.18.3.298 . Zhang L, Li J, Yang B, et al. The risk and outcome of malignant brain edema in post-mechanical thrombectomy: acute ischemic stroke by anterior circulation occlusion. Eur J Med Res. 2023;28:435. https://doi.org/10.1186/s40001-023-01414-x . Zheng X, Ren B, Gao Y. Tight junction proteins related to blood-brain barrier and their regulatory signaling pathways in ischemic stroke. Biomed Pharmacother. 2023;165:115272. https://doi.org/10.1016/j.biopha.2023.115272 . Luo L, Hu Q, Yan R, et al. Alpha–Asarone Ameliorates Neuronal Injury After Ischemic Stroke and Hemorrhagic Transformation by Attenuating Blood–Brain Barrier Destruction, Promoting Neurogenesis, and Inhibiting Neuroinflammation. Mol Neurobiol. 2025;62:5252–72. https://doi.org/10.1007/s12035-024-04596-5 . Zhang L, Xie G, Zhang Y, et al. A CT-based machine learning model for using clinical-radiomics to predict malignant cerebral edema after stroke: a two-center study. Front Neurosci. 2024;18:1443486. https://doi.org/10.3389/fnins.2024.1443486 . Fu B, Qi S, Tao L, et al. Image Patch-Based Net Water Uptake and Radiomics Models Predict Malignant Cerebral Edema After Ischemic Stroke. Front Neurol. 2020;11:609747. https://doi.org/10.3389/fneur.2020.609747 . Sarioglu O, Sarioglu FC, Capar AE, et al. Clot-based radiomics features predict first pass effect in acute ischemic stroke. Interv Neuroradiol. 2022;28:160–8. https://doi.org/10.1177/15910199211019176 . Xiong X, Wang J, Ke J, et al. Radiomics-based intracranial thrombus features on preoperative noncontrast CT predicts successful recanalization of mechanical thrombectomy in acute ischemic stroke. Quant Imaging Med Surg. 2023;13:682–94. https://doi.org/10.21037/qims-22-599 . Li M, Jiang J, Gu H, et al. CT-Based Intrathrombus and Perithrombus Radiomics for Prediction of Prognosis after Endovascular Thrombectomy: A Retrospective Study across 2 Centers. AJNR Am J Neuroradiol. 2025;46:681–8. https://doi.org/10.3174/ajnr.A8522 . Lu J, Zhou Y, Lv W, et al. Identification of early invisible acute ischemic stroke in non-contrast computed tomography using two-stage deep-learning model. Theranostics. 2022;12:5564–73. https://doi.org/10.7150/thno.74125 . Tables Table 1 . Baseline clinical characteristics of the patients. Clinical characteristics Training(MCE-) Training(MCE+) p value Test(MCE-) Test(MCE+) p value Validation(MCE-) Validation(MCE+) p value (n=148) (n=30) (n=70) (n=7) (n=139) (n=12) Age 71.72±11.96 69.00±11.88 0.213 62.47±11.74 62.14±17.22 0.946 69.58±11.26 66.42±11.61 0.283 NIHSS 15.08±8.28 10.90±6.26 0.01 13.46±6.24 9.43±4.58 0.101 12.05±5.93 12.92±5.33 0.549 Gender 0.507 1 0.763 Female 67(45.27) 11(36.67) 22(31.43) 2(28.57) 57(41.01) 6(50.00) Male 81(54.73) 19(63.33) 48(68.57) 5(71.43) 82(58.99) 6(50.00) Atrial fibrillation 0.064 0.97 0.539 Absence 89(60.14) 24(80.00) 44(62.86) 5(71.43) 111(79.86) 11(91.67) Presence 59(39.86) 6(20.00) 26(37.14) 2(28.57) 28(20.14) 1(8.33) Smoke 0.592 0.477 0.4 Absence 118(79.73) 22(73.33) 54(77.14) 4(57.14) 113(81.29) 8(66.67) Presence 30(20.27) 8(26.67) 16(22.86) 3(42.86) 26(18.71) 4(33.33) Hypertension 0.969 0.785 0.547 Absence 36(24.32) 8(26.67) 21(30.00) 3(42.86) 51(36.69) 6(50.00) Presence 112(75.68) 22(73.33) 49(70.00) 4(57.14) 88(63.31) 6(50.00) Hyperlipidemia 0.668 0.234 1 Absence 126(85.14) 24(80.00) 65(92.86) 5(71.43) 138(99.28) 12(100.00) Presence 22(14.86) 6(20.00) 5(7.14) 2(28.57) 1(0.72) 0 Diabetes 0.85 0.362 0.247 Absence 103(69.59) 22(73.33) 56(80.00) 4(57.14) 108(77.70) 7(58.33) Presence 45(30.41) 8(26.67) 14(20.00) 3(42.86) 31(22.30) 5(41.67) Coronary disease 1 0.571 1 Absence 119(80.41) 24(80.00) 61(87.14) 5(71.43) 130(93.53) 11(91.67) Presence 29(19.59) 6(20.00) 9(12.86) 2(28.57) 9(6.47) 1(8.33) Location 0.839 0.104 0.613 MCA 105(70.95) 22(73.33) 49(70.00) 5(71.43) 90(64.75) 8(66.67) ICA 31(20.95) 5(16.67) 20(28.57) 1(14.29) 39(28.06) 4(33.33) MCA+ICA 12(8.11) 3(10.00) 1(1.43) 1(14.29) 10(7.19) 0 Tabe 2 . Performance of LR models. Models Intra-thrombus Peri-thrombus Combined Training Test Validation Training Test Validation Training Test Validation Accuracy 0.876 0.87 0.887 0.888 0.831 0.974 0.938 0.753 0.808 AUC 0.968 0.741 0.626 0.976 0.908 0.891 0.993 0.927 0.869 95% CI 0.9458 - 0.9907 0.5063 - 0.9754 0.4325 - 0.8205 0.9563 - 0.9950 0.7826 - 1.0000 0.7615 - 1.0000 0.9854 - 1.0000 0.8473 - 1.0000 0.7776 - 0.9611 Sensitivity 0.967 0.429 0.333 0.933 0.714 0.667 0.967 0.857 0.75 Specificity 0.858 0.914 0.935 0.878 0.843 1 0.932 0.743 0.813 PPV 0.58 0.333 0.308 0.609 0.312 1 0.744 0.25 0.257 NPV 0.992 0.941 0.942 0.985 0.967 0.972 0.993 0.981 0.974 F1 0.725 0.375 0.32 0.737 0.435 0.8 0.841 0.387 0.383 Threshold 0.196 0.376 0.257 0.182 0.12 0.353 0.2364 0.04271 0.09833 MCC 0.71 0.415 0.328 0.718 0.485 0.857 0.836 0.456 0.41 Additional Declarations No competing interests reported. Supplementary Files SupplementaryTables.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6948356","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":494868019,"identity":"22486f21-3ef1-44c6-b73b-233d989211f0","order_by":0,"name":"Shuhao Wang","email":"","orcid":"","institution":"Guanghua Hospital Affiliated to Shanghai University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Shuhao","middleName":"","lastName":"Wang","suffix":""},{"id":494868020,"identity":"785320d5-491b-4f5c-9fcb-a48e4f2aadde","order_by":1,"name":"Xiaoli Gu","email":"","orcid":"","institution":"Guanghua Hospital Affiliated to Shanghai University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Xiaoli","middleName":"","lastName":"Gu","suffix":""},{"id":494868022,"identity":"16137201-d6fd-49f1-b4be-4be4681ec93e","order_by":2,"name":"Haiqi Wang","email":"","orcid":"","institution":"Guanghua Hospital Affiliated to Shanghai University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Haiqi","middleName":"","lastName":"Wang","suffix":""},{"id":494868023,"identity":"40c538ed-1a36-46b1-8c0e-dfcecefe3d7f","order_by":3,"name":"Yangyang Nan","email":"","orcid":"","institution":"Guanghua Hospital Affiliated to Shanghai University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yangyang","middleName":"","lastName":"Nan","suffix":""},{"id":494868025,"identity":"73298be3-eb57-4730-b934-adfbafb530d6","order_by":4,"name":"Jia Zhou","email":"","orcid":"","institution":"Shanghai Jiao Tong University Affiliated Sixth People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jia","middleName":"","lastName":"Zhou","suffix":""},{"id":494868027,"identity":"18215a7e-3524-455a-bc5c-73eb759627a9","order_by":5,"name":"Xiaoyu Xu","email":"","orcid":"","institution":"Shanghai Jiao Tong University Affiliated Sixth People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xiaoyu","middleName":"","lastName":"Xu","suffix":""},{"id":494868029,"identity":"76297f61-3e98-41ef-8565-dbead385105d","order_by":6,"name":"Chenqing Wang","email":"","orcid":"","institution":"Shukun Technology Co., Ltd","correspondingAuthor":false,"prefix":"","firstName":"Chenqing","middleName":"","lastName":"Wang","suffix":""},{"id":494868031,"identity":"6388e62b-41e0-4d5a-8a17-7914573f268c","order_by":7,"name":"Jingxuan Jiang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAy0lEQVRIiWNgGAWjYBACxoYDjA8+GEjwyLM3Nj78QKQWZsMZBRZyhj2Hm40liLSITZrnQ4Uxw430NgEeYtQzN54xNuYxkEhsnPmwjUGCwU5Ot4Ggw44lPpwD1NIundj2oIAh2djsAEEthw8bvAHZMjux3UCC4UDiNsJaDrZJgBzWcBPEIE7L4WOSQC1A7zMSreVYsuEMAwlgICcCA9mACL8Yzjhj+ODDnzpgVB5/+PBDhZ0cEVpQVBgQUA4C8vwNRKgaBaNgFIyCkQ0ACldIOh3w/O8AAAAASUVORK5CYII=","orcid":"","institution":"Shanghai Jiao Tong University Affiliated Sixth People’s Hospital","correspondingAuthor":true,"prefix":"","firstName":"Jingxuan","middleName":"","lastName":"Jiang","suffix":""}],"badges":[],"createdAt":"2025-06-22 08:08:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6948356/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6948356/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":88406895,"identity":"cea25275-d780-4f62-9201-88aa6de00353","added_by":"auto","created_at":"2025-08-06 08:04:48","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":106440,"visible":true,"origin":"","legend":"\u003cp\u003eFlow chart of the patient-selection process.\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6948356/v1/7df100294a715bd16ee0c48e.jpg"},{"id":88406900,"identity":"b2fcbc91-6c2e-4f7c-a7f9-40773d7daa28","added_by":"auto","created_at":"2025-08-06 08:04:48","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":230994,"visible":true,"origin":"","legend":"\u003cp\u003eThe workflow of this study.\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6948356/v1/a01d20e1cb65d56f6f42072f.jpg"},{"id":88406902,"identity":"54db7af5-35c7-4a7a-847a-615320fbcdf3","added_by":"auto","created_at":"2025-08-06 08:04:48","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":106823,"visible":true,"origin":"","legend":"\u003cp\u003eFeature selection process.\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6948356/v1/efc7ae3e355b4340424cd794.jpg"},{"id":88406899,"identity":"8df8e32d-1577-4b7b-abd0-e9d1e84d26e9","added_by":"auto","created_at":"2025-08-06 08:04:48","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":134390,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e4a\u003c/strong\u003eillustrates the MCC results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4b \u003c/strong\u003eshows the AUC results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4c\u003c/strong\u003e shows the DeLong test results between SVM and LR models of two categories.\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6948356/v1/d6ae4831ea6d4fe066d29195.jpg"},{"id":88406905,"identity":"cbaf4269-fd11-4705-954c-a2c40fae0314","added_by":"auto","created_at":"2025-08-06 08:04:48","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":122265,"visible":true,"origin":"","legend":"\u003cp\u003eSpecific cases are shown in\u003cstrong\u003e.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eModel prediction in endovascular thrombectomy (EVT) patients. A 45-year-old female predicted and confirmed with post-EVT MCE (case 1), and a 60-year-old male predicted and found without post-EVT MCE (case 2).\u003c/p\u003e","description":"","filename":"Picture5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6948356/v1/00244a8457ac24ad45ae0b9a.jpg"},{"id":88963354,"identity":"6747784b-6ff9-409f-b8b5-411024360e6b","added_by":"auto","created_at":"2025-08-13 08:33:05","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1534385,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6948356/v1/0db9bab5-a90d-4d52-a046-cee8b7bd9771.pdf"},{"id":88406896,"identity":"581b664e-7423-4999-bc74-3705bd9af5b2","added_by":"auto","created_at":"2025-08-06 08:04:48","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":46807,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTables.docx","url":"https://assets-eu.researchsquare.com/files/rs-6948356/v1/5ccd535c0b27773775365612.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Radiomics signature and deep learning signature of intrathrombus and perithrombus for prediction of Malignant Cerebral Edema after Acute Ischemic Stroke: a multicenter CT study","fulltext":[{"header":"Key Points","content":"\u003cp\u003e• \u0026nbsp;Machine learning models related to thrombosis can effectively predict the occurrence of MCE after AIS.\u003c/p\u003e\n\u003cp\u003e• The proposed LR-peri radiomics model reached a higher area under the curve (AUC: 0.891, 95% CI: 0.762 - 1.000).\u003c/p\u003e\n\u003cp\u003e• Its application provides a beneficial approach for formulating personalized treatment strategies for patients with AIS.\u003c/p\u003e"},{"header":"Introduction","content":"\u003cp\u003eStroke ranks as the second leading cause of death globally, and MCE is one of the severe complications, with an incidence rate of approximately 10% [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Cytotoxic edema usually peaks 3 to 4 days after brain injury, but reperfusion of necrotic tissue can cause malignant edema within the first 24 hours. Decompressive craniectomy within 48 hours improves outcomes and reduces mortality in large-scale infarctions, but unnecessary surgery is highly invasive [\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Thus, early and accurate prediction of complications is essential.\u003c/p\u003e\u003cp\u003eCT is the first-line imaging modality for stroke patients on admission and can predict ischemic brain tissue progression. Wen X et al. conducted a study on predicting MCE by extracting the CT radiomics features of the middle cerebral artery (MCA) blood supply area from Non-Contrast Computed Tomography (NCCT) images of patients with cerebral infarction. Shi J et al. demonstrated that the combined Alberta stroke program early CT score and net water uptake (ASPECTS\u0026ndash;NWU) could serve as a quantitative predictor of MCE after MCA territory large vessel occlusion, with a moderate positive correlation with the grade of brain edema, indicating that quantitative measurements of ASPECT score, net water uptake, and enhancement ratio based on CT imaging are effective predictive factors for MCE [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Prior studies have primarily focused on the infarct core, with less research focusing on the impact of the culprit thrombus and surrounding tissue on post-stroke edema. In ischemic stroke, disruption of the blood-brain barrier leads to vasogenic edema, hemorrhagic transformation, and increased mortality. This pathological process is influenced by thrombus characteristics, as research indicates that thrombi with low red blood cell content, high fibrin levels, and elevated extracellular DNA are less likely to achieve first-pass recanalization (FPR). Some studies have also confirmed that the serum inflammatory factor levels and BBB disruption after AIS are associated with the occurrence of vasogenic cerebral edema [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. It is reasonable to assume that the characteristics of the thrombus and surrounding brain tissue can more precisely reflect the inflammatory response and BBB disruption in post-stroke brain tissue [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Moreover, extracting high-dimensional quantitative radiomic features and deep learning features from medical images to construct machine learning predictive models has its advantages of reducing physician subjective judgment factors and improving accuracy [\u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThrombus and perithrombus radiomic features can predict the origin and prognosis of thrombi. For example, according to our team\u0026rsquo;s previous research, it was found that 1) thrombus radiomic features could predict the origin and composition of stroke thrombi, and 2) the logistic regression model combining radiomic features from both inside and around the thrombus could effectively assess clinical prognosis after EVT [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. However, our previous studies did not involve deep learning features. DL features refer to high-dimensional data representations automatically extracted by multi-layer neural networks, which can effectively capture complex patterns and structures in the data [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. The application of deep learning in the field of stroke covers multiple aspects, from the detection of acute cerebral infarction, lesion segmentation, ASPECTS quantification, to prognostic prediction [\u003cspan additionalcitationids=\"CR16 CR17\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThis study aims to fill the gap in the research on predicting MCE using radiomic and deep learning features of the thrombus and its surrounding areas. It deeply explores the possibility of predicting the development of AIS into MCE based on CT thrombus region features, providing strong support for neurosurgeons to formulate personalized treatment strategies.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003ePatients\u003c/h2\u003e\u003cp\u003e This retrospective study adhered to the 1964 Declaration of Helsinki and its amendments, and was approved by our local ethics committee. Written informed consent was waived by the Institutional Review Board. We analyzed data from stroke patients admitted to three centers (A: ***, B: ***, and C: ***) between December 2018 and December 2023. Inclusion criteria: (1) Acute stroke due to anterior-circulation Large-vessel occlusion (LVO); (2) Admission non-contrast computed tomography (NCCT) and computed tomography angiography showing visible thrombus; (3) EVT performed immediately after CT; (4) Follow-up images of sufficient quality taken within 24 h post-EVT; (5) Complete demographic and clinical data. Exclusion criteria: (1) Inadequate imaging clarity due to motion or metal artifacts; (2) Incomplete clinical records. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows a flowchart of the patient selection process. The clinical data collected from medical and follow-up records included age, gender, National Institutes of Health Stroke Scale (NIHSS) scores, lesion location, comorbidities (atrial fibrillation, hypertension, hyperlipidemia, diabetes mellitus, and coronary heart disease), and smoking history. MCE was defined as a midline shift of \u0026ge;\u0026thinsp;5 millimeters accompanied by signs of local cerebral swelling [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], whose identification was based on follow-up imaging.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eCT scan and thrombosis segmentation\u003c/h3\u003e\n\u003cp\u003eNCCT and CTA examinations were performed using multi-detector CT scanners from three manufacturers: Philips (Brilliance ICT), Toshiba (Aquilion ONE/PRIME), and Siemens (Somatom Sensation). Prior to thrombus segmentation, each CT image underwent an intensity normalization process as described in our previous studies. ROIs for thrombus were outlined using ITK-SNAP software (Version 3.6.0; [ITK-SNAP Home](\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.itksnap.org/pmwiki/pmwiki.php\u003c/span\u003e\u003cspan address=\"http://www.itksnap.org/pmwiki/pmwiki.php\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)). Following segmentation of the intrathrombus areas, perithrombus areas were automatically demarcated by expanding the radius of the initial 1-mm ROIs. To ensure segmentation accuracy, thrombi were segmented by two radiology residents (*** and ***) who reached a consensus after consultation, and their work was reviewed by a radiology attending physician (***). Evaluators were blinded to clinical details.\u003c/p\u003e\n\u003ch3\u003eFeature extraction, selection, and model building\u003c/h3\u003e\n\u003cp\u003eAfter the identification of regions within and surrounding the thrombus, radiomic features were derived utilizing the PyRadiomics (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pypi.org/project/pyradiomic/\u003c/span\u003e\u003cspan address=\"https://pypi.org/project/pyradiomic/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). The NCCT and CTA images of the area inside and around the thrombus were cropped into fully-covered two-dimensional images. The size of each image was adjusted to 224\u0026times;224 and then input into the VGG16 model. Finally, 107 radiomic features and 32 deep learning features were extracted from the intrathrombus and perithrombus regions on NCCT and CTA images respectively.\u003c/p\u003e\u003cp\u003eThe features with \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05 were retained by Mann-Whitney U test, and the features with correlation coefficients greater than 0.9 were eliminated according to the Spearman correlation coefficient to achieve dimensionality reduction. After performing feature screening using the LASSO, which included radiomic features and deep learning features, these features were input into machine learning model using 11 different algorithms: Logistic Regression (LR), Naive Bayes (NB), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), Extra-Tree, XG Boost, Light GBM, Gradient Boosting (GB), AdaBoost, and Multilayer Perceptron (MLP). The five-fold cross-validation method was employed to verify the predictive performance of each model. MCC was used to screen optimal algorithm for constructing machine learning models. The performance of these models were evaluated using the ROC curve, along with the AUC, accuracy, precision, specificity, and other metrics. The DeLong test was used to statistically assess the differences in the predictive performance of the machine learning models (intrathrombus, perithrombus, and combined models). The workflow of this study is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003eStatistical analysis\u003c/h2\u003e\u003cp\u003eClinical characteristics were analyzed by the t-test, Mann\u0026ndash;Whitney U test, or chi-squared test, as appropriate. The model\u0026rsquo;s predictive performance for MCE following AIS were evaluated by conducting ROC curve analysis to calculate metrics like the AUC, sensitivity, specificity, positive and negative predictive values, and DeLong\u0026rsquo;s test was employed to compare ROC curves and assess the model\u0026rsquo;s clinical utility via decision curve analysis (DCA).\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003ePatient characteristics\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this study, 406 AIS patients were respectively selected according to the inclusive criteria. Eligible patients from Center A were randomly assigned to the training group (n=178) and the testing group (n=77) at a 7:3 ratio, while those from Centers B and C formed the validation group (n=151). The incidence of MCE after cerebral infarction was 49 cases (12.1%). \u003cstrong\u003eTable 1\u003c/strong\u003e systematically presents the detailed demographic information of the patients in different cohorts classified according to the occurrence of MCE. There was a significant statistical difference in the National Institutes of Health Stroke Scale (NIHSS) score in the training cohort. \u0026nbsp; However, there was no significant statistical difference of NIHSS scores in the testing and validation cohorts. There was also no significant statistical difference among the three groups in terms of other clinical features.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFeature extraction and selection\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAfter dimensionality reduction, a total of 12 DL features and 4 radiomic features (including 9 CTA features and 7 NCCT features) were used to construct machine learning models based on intrathrombus image features. Regarding the models established based on perithrombus images, they incorporate 14 DL features and 4 radiomic features (including 10 CTA features and 8 NCCT features). Simultaneously, the models developed according to the features of combined are equipped with 24 DL features and 8 radiomics features (including 16 CTA features and 16 NCCT features). Feature selection is shown in\u003cstrong\u003e\u0026nbsp;Figure 3\u003c/strong\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 3\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Performance and Comparison\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe performance of eleven classifiers were evaluated based on MCC values. In the training sets, most classifiers (8 in 11) performed well (MCC\u0026gt;0.7). \u003cstrong\u003eFigure 4a\u003c/strong\u003e illustrates the MCC results.\u003cstrong\u003e\u0026nbsp;Figure 4a\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;When evaluating the performance of eleven models in the validation set for the perithrombus models, LR performed the best (AUC: 0.891, 95% CI: 0.762 - 1.000). For the combined models, LR also showed the optimal performance (AUC: 0.869, 95% CI: 0.778 - 0.9618). For the intrathrombus models, SVM performed well (AUC: 0.733, 95% CI: 0.546 - 0.921). \u003cstrong\u003eFigure 4b\u003c/strong\u003e shows the AUC results. In view of all factors, LR, which performed the best in the predictive task, was selected to construct the prediction models. For detailed information about these models, please refer to \u003cstrong\u003eTable 2\u0026nbsp;\u003c/strong\u003eand the supplementary materials.\u003cstrong\u003e\u0026nbsp;Figure 4b\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn the validation set, the LR-intra model had the lowest AUC value (0.626), showing relatively weak performance in the task. The AUC value of the LR-combine (0.869) was close to that of the LR-peri model.\u0026nbsp;The comparison between LR-intra and LR-combine showed a significant difference (\u003cem\u003ep\u003c/em\u003e = 0.007), while the comparisons between LR-intra and LR-peri, as well as between the LR-combined and the LR-peri, did not show significant differences (\u003cem\u003ep\u003c/em\u003e = 0.063 and \u003cem\u003ep\u003c/em\u003e = 0.788).\u003cstrong\u003e\u0026nbsp;Figure 4c\u003c/strong\u003e shows the DeLong test results between SVM and LR models of two categories. During the model validation stage, the DCA indicated that the combined model showed a higher net benefit (0.069), while the perithrombus model had a wider effective threshold probability range (0.08-0.97). Specific cases are shown in \u003cstrong\u003eFigure 5\u003c/strong\u003e.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe common cause of AIS is the sudden blockage of the proximal middle cerebral artery or the distal internal carotid artery. The BBB is damaged, leading to excessive water infiltration into brain tissue, with a mortality rate that may be as high as 80% [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Currently, clinical diagnosis of MCE relies on observing midline shift or brain herniation via CT, which are usually late signs of the disease. The main goal of our research is to develop a model for more early prediction of MCE in AIS patients after EVT. In our study, the perithrombus model exhibited superior predictive capacity in the validation cohort (MCC\u0026thinsp;=\u0026thinsp;0.857, AUC\u0026thinsp;=\u0026thinsp;0.891). The model can serve as a tool for early prediction of AIS complications when applied to clinical scenarios, thereby improving the quality of survival of AIS patients after EVT.\u003c/p\u003e\u003cp\u003ePrevious developed models mainly extracted image features of the brain infarction area of NCCT images or segmented the MCA supply area of NCCT images. However, it is difficult to accurately delineate the infarcted area on CT images, making it hard to be popularized. Zhang et al. constructed a machine learning model based on the brain infarction area of NCCT images to predict the occurrence of MCE [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. The model had an AUC of 0.912, showing good predictive ability. Fu et al. evaluated the predictive ability of the IP\u0026ndash;NWU value in the middle cerebral artery supply area of NCCT for MCE. The radiomic model had a maximum AUC of 0.96 [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. However, these studies ignored the impact of the responsible blood vessels on the prognosis of brain infarction, and there was little research on deep learning features related to the thrombus area. Additionally, Sarioglu O et al. found that thrombus-based radiomic features could effectively predict the first-pass effect (FPE) in patients with AIS [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Similar findings were also reported by XIONG et al [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. When a patient develops AIS, the ischemia and hypoxia of brain tissue caused by the thrombus will damage the BBB, leading to the leakage of macromolecules in the plasma into the brain tissue interstitium due to the increased vascular permeability. The inflammatory response caused by the thrombus will further damage the vascular endothelial cells and exacerbate the vasogenic brain edema. Subsequent reperfusion after revascularization may also aggravate brain injury, thus triggering or exacerbating brain edema. The area surrounding a thrombus includes structures such as vascular wall cells and perivascular fat. It can be seen that the radiomics features around the thrombus can effectively predict MCE and this prediction is interpretable. Li et al. conducted a retrospective analysis of studies related to CT scans before EVT [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. The research showed that the LR model combining intrathrombus and perithrombus radiomics features was very effective in predicting the prognosis of thrombectomy, with an AUC value as high as 0.87 in the validation cohort. Lu et al. developed a two-stage deep learning model to identify early occult AIS in NCCT [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. However, we haven\u0026rsquo;t seen any relevant research based on the deep learning features of thrombus so far. Inspired by these studies, we constructed a machine learning prediction model for predicting MCE using radiomics and deep learning features extracted from the thrombus and its surrounding areas in NCCT and CTA. Compared with complex models, LR is favored for its statistical simplicity, interpretability, and robust performance in binary classification tasks. Although complex models, like SVM, KNN, RF, Extra Trees, XG Boost, MLP, GB, NB, and AdaBoost, have advantages in handling high-dimensional data, they are prone to overfitting without a large amount of data and careful adjustment. Since the Light GBM model performed poorly in different cohorts of this experiment, which may be related to factors such as the feature distribution of the data, sample size, noise, and the parameter settings of the model, this model should be avoided in future related research. Whereas the consistent performance of LR in the validation cohort confirms its applicability and reliability in clinical diagnosis. Given its effective generalization ability across different datasets, the choice of LR is reasonable.\u003c/p\u003e\u003cp\u003eIn the validation cohort, compared with using only the intrathrombus area and the combined area, the radiomics features of the perithrombus area significantly improved the predictive ability for MCE after acute cerebral infarction (AUC: 0.891, 95% CI: 0.761 to 1.000). During model validation, the combined model yielded a higher net benefit (0.069) compared to the perithrombus model, which demonstrated a broader effective threshold probability range (0.08\u0026ndash;0.97). Clinicians should therefore adopt a dual consideration of net benefit and threshold probability range in clinical decision-making. For instance, integrating the complementary strengths of the combined and perithrombus models may optimize risk stratification compared with the unstable predictive effect of intrathrombus radiomics features, the radiomic features extracted from the perithrombus area play an important role in the prediction model.\u003c/p\u003e\u003cp\u003eIn this study, except for the significant statistical difference in the NIHSS index of the training cohort, other clinical variables with or without MCE did not show significant statistical differences in either the training cohort or the validation cohort. Although this study has innovation and important findings, it is still restricted by various factors. The retrospective design inherently limits its causal inference ability, and the relatively insufficient sample size weakens the universality and robustness of the conclusions. Differences in contrast agent selection may interfere with the cross-scenario application of the model. Although the reliability of radiomics feature extraction is high, manual drawing of ROIs may introduce human errors and uncertainties. In the future, fully automated technology urgently needs to be introduced for optimization. The exploitation of deep learning to enhance model precision represents a critical trend. Notwithstanding, challenges including limited multi-center validation, inadequate integration of clinical covariates, and the necessity for advanced hybrid modeling techniques persist. Future investigations should prioritize overcoming these barriers to elevate both research quality and clinical translational value.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe LR model proposed in this study integrates the NCCT and CTA image features of the perithrombus areas after cerebral infarction through machine learning methods, providing a way to predict MCE. This may help clinicians decide earlier whether to perform decompressive craniectomy or adopt other intensive monitoring measures, ultimately benefiting stroke patients.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eDL \u0026nbsp; \u0026nbsp; \u0026nbsp; Deep learning\u003c/p\u003e\n\u003cp\u003eMCE \u0026nbsp; \u0026nbsp;Malignant cerebral edema\u003c/p\u003e\n\u003cp\u003eAIS \u0026nbsp; \u0026nbsp; \u0026nbsp; Acute ischemic stroke\u003c/p\u003e\n\u003cp\u003eEVT \u0026nbsp; \u0026nbsp; \u0026nbsp;Endovascular thrombectomy\u003c/p\u003e\n\u003cp\u003eROIs \u0026nbsp; \u0026nbsp; Regions of interest\u003c/p\u003e\n\u003cp\u003eLASSO Least absolute shrinkage and selection operator\u003c/p\u003e\n\u003cp\u003eMCC \u0026nbsp; \u0026nbsp; Matthews correlation coefficient\u003c/p\u003e\n\u003cp\u003eAUC \u0026nbsp; \u0026nbsp; Area under the receiver operating characteristics curve\u003c/p\u003e\n\u003cp\u003eMCA \u0026nbsp; \u0026nbsp; Middle cerebral artery\u003c/p\u003e\n\u003cp\u003eNCCT \u0026nbsp; Non-Contrast Computed Tomography\u003c/p\u003e\n\u003cp\u003eASPECTS–NWU Alberta stroke program early CT score and net water uptake\u003c/p\u003e\n\u003cp\u003eFPR \u0026nbsp; \u0026nbsp; \u0026nbsp;First-pass recanalization\u003c/p\u003e\n\u003cp\u003eDCA \u0026nbsp; \u0026nbsp; Decision curve analysis\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eStudy approval was obtained from the ethics committee of Sixth People's Hospital Affiliated to Shanghai Jiao Tong University (registration number: 2020-212). The study was performed in accordance with the relevant guidelines and/or regulations including the Declaration of Helsinki. Informed consent to participate was waived for this retrospective cohort study by the ethics committee of Sixth People's Hospital Affiliated to Shanghai Jiao Tong University.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable .\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interest\u003c/strong\u003e\u003c/p\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\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSupported by National Natural Science Foundation of China No.82402364\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors' contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eJiang Jingxuan: Conceptualization, Methodology, Funding acquisition, Writing - Review \u0026amp; Editing\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eWang Shuhao: Writing- Original draft preparation, Data curation, Formal analysis,\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eGu Xiaoli: Visualization, Investigation, Resources, Project administration.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eWang Haiqi: Supervision, Software, Investigation.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eNan Yangyang: Software, Validation.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eZhou Jia: Data Curation, Software, Investigation.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eXu Xiaoyu: Project administration .\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eWang Chenqing\u003c/em\u003e\u003cem\u003e：\u003c/em\u003e\u003cem\u003eSoftware.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNone\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eWen X, Hu X, Xiao Y, Chen J. Radiomics analysis for predicting malignant cerebral edema in patients undergoing endovascular treatment for acute ischemic stroke. dir. 2023;29:402\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4274/dir.2023.221764\u003c/span\u003e\u003cspan address=\"10.4274/dir.2023.221764\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCheng Y, Wu S, Wang Y, et al. External Validation and Modification of the EDEMA Score for Predicting Malignant Brain Edema After Acute Ischemic Stroke. Neurocrit Care. 2020;32:104\u0026ndash;12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s12028-019-00844-y\u003c/span\u003e\u003cspan address=\"10.1007/s12028-019-00844-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDower A, Mulcahy M, Maharaj M et al. (2023) Surgical Decompression for Malignant Cerebral Edema After Ischemic Stroke: Cochrane Review. Stroke 54:. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/STROKEAHA.122.042260\u003c/span\u003e\u003cspan address=\"10.1161/STROKEAHA.122.042260\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCannarsa GJ, Simard JM. Decompressive Craniectomy for Stroke: Who, When, and How. Neurol Clin. 2022;40:321\u0026ndash;36. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ncl.2021.11.009\u003c/span\u003e\u003cspan address=\"10.1016/j.ncl.2021.11.009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShi J, Wu H, Dong Z, et al. Automated quantitative lesion water uptake in acute stroke is a predictor of malignant cerebral edema. Eur Radiol. 2022;32:2771\u0026ndash;80. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00330-021-08443-2\u003c/span\u003e\u003cspan address=\"10.1007/s00330-021-08443-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStafford R, Chatzidakis S, Kim ISY, et al. Follow-up ASPECTS improves prediction of potentially lethal malignant edema in patients with large middle cerebral artery stroke. J NeuroIntervent Surg. 2025;17:e83\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/jnis-2023-021145\u003c/span\u003e\u003cspan address=\"10.1136/jnis-2023-021145\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang M, Pang X, Lu H, Wang X. Clinical role of serum histone deacetylase 4 measurement in acute ischemic stroke: Relation to disease risk, severity, and prognosis. Clin Lab Anal. 2022;36:e24372. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/jcla.24372\u003c/span\u003e\u003cspan address=\"10.1002/jcla.24372\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMeng S, Cao H, Huang Y, et al. ASK1-K716R reduces neuroinflammation and white matter injury via preserving blood\u0026ndash;brain barrier integrity after traumatic brain injury. J Neuroinflammation. 2023;20:244. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12974-023-02923-6\u003c/span\u003e\u003cspan address=\"10.1186/s12974-023-02923-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJan Y-T, Tsai P-S, Huang W-H, et al. Machine learning combined with radiomics and deep learning features extracted from CT images: a novel AI model to distinguish benign from malignant ovarian tumors. Insights Imaging. 2023;14:68. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s13244-023-01412-x\u003c/span\u003e\u003cspan address=\"10.1186/s13244-023-01412-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSheng L, Yang C, Chen Y, Song B. Machine Learning Combined with Radiomics Facilitating the Personal Treatment of Malignant Liver Tumors. Biomedicines. 2023;12:58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/biomedicines12010058\u003c/span\u003e\u003cspan address=\"10.3390/biomedicines12010058\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang J, Cao Y, Zhou F, et al. Combined deep-learning MRI-based radiomic models for preoperative risk classification of endometrial endometrioid adenocarcinoma. Front Oncol. 2023;13:1231497. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fonc.2023.1231497\u003c/span\u003e\u003cspan address=\"10.3389/fonc.2023.1231497\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJiang J, Wei J, Zhu Y, et al. Clot-based radiomics model for cardioembolic stroke prediction with CT imaging before recanalization: a multicenter study. Eur Radiol. 2022;33:970\u0026ndash;80. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00330-022-09116-4\u003c/span\u003e\u003cspan address=\"10.1007/s00330-022-09116-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi M, Zhou J, Sheng K, et al. Radiomics of intrathrombus and perithrombus regions for Post-EVT intracranial hemorrhage risk Prediction: A multicenter CT study. Eur J Radiol. 2024;178:111653. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ejrad.2024.111653\u003c/span\u003e\u003cspan address=\"10.1016/j.ejrad.2024.111653\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang Y, Li N, Chen L, et al. Guidelines, Consensus Statements, and Standards for the Use of Artificial Intelligence in Medicine: Systematic Review. J Med Internet Res. 2023;25:e46089. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2196/46089\u003c/span\u003e\u003cspan address=\"10.2196/46089\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLei L, Zhou Y, Guo X, et al. The value of a deep learning image reconstruction algorithm in whole-brain computed tomography perfusion in patients with acute ischemic stroke. Quant Imaging Med Surg. 2023;13:8173\u0026ndash;89. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21037/qims-23-547\u003c/span\u003e\u003cspan address=\"10.21037/qims-23-547\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHakim A, Christensen S, Winzeck S, et al. Predicting Infarct Core From Computed Tomography Perfusion in Acute Ischemia With Machine Learning: Lessons From the ISLES Challenge. Stroke. 2021;52:2328\u0026ndash;37. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/STROKEAHA.120.030696\u003c/span\u003e\u003cspan address=\"10.1161/STROKEAHA.120.030696\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAyobi A, Davis A, Chang PD, et al. Deep Learning\u0026ndash;Based ASPECTS Algorithm Enhances Reader Performance and Reduces Interpretation Time. AJNR Am J Neuroradiol. 2024. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3174/ajnr.A8491\u003c/span\u003e\u003cspan address=\"10.3174/ajnr.A8491\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. ajnr;ajnr.A8491v2.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHerzog L, Kook L, Hamann J, et al. Deep Learning Versus Neurologists: Functional Outcome Prediction in LVO Stroke Patients Undergoing Mechanical Thrombectomy. Stroke. 2023;54:1761\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/STROKEAHA.123.042496\u003c/span\u003e\u003cspan address=\"10.1161/STROKEAHA.123.042496\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJiang Q, Yu S, Dong X, et al. Predictors and Dynamic Nomogram to Determine the Individual Risk of Malignant Brain Edema After Endovascular Thrombectomy in Acute Ischemic Stroke. J Clin Neurol. 2022;18:298. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3988/jcn.2022.18.3.298\u003c/span\u003e\u003cspan address=\"10.3988/jcn.2022.18.3.298\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang L, Li J, Yang B, et al. The risk and outcome of malignant brain edema in post-mechanical thrombectomy: acute ischemic stroke by anterior circulation occlusion. Eur J Med Res. 2023;28:435. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s40001-023-01414-x\u003c/span\u003e\u003cspan address=\"10.1186/s40001-023-01414-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZheng X, Ren B, Gao Y. Tight junction proteins related to blood-brain barrier and their regulatory signaling pathways in ischemic stroke. Biomed Pharmacother. 2023;165:115272. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.biopha.2023.115272\u003c/span\u003e\u003cspan address=\"10.1016/j.biopha.2023.115272\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLuo L, Hu Q, Yan R, et al. Alpha\u0026ndash;Asarone Ameliorates Neuronal Injury After Ischemic Stroke and Hemorrhagic Transformation by Attenuating Blood\u0026ndash;Brain Barrier Destruction, Promoting Neurogenesis, and Inhibiting Neuroinflammation. Mol Neurobiol. 2025;62:5252\u0026ndash;72. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s12035-024-04596-5\u003c/span\u003e\u003cspan address=\"10.1007/s12035-024-04596-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang L, Xie G, Zhang Y, et al. A CT-based machine learning model for using clinical-radiomics to predict malignant cerebral edema after stroke: a two-center study. Front Neurosci. 2024;18:1443486. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fnins.2024.1443486\u003c/span\u003e\u003cspan address=\"10.3389/fnins.2024.1443486\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFu B, Qi S, Tao L, et al. Image Patch-Based Net Water Uptake and Radiomics Models Predict Malignant Cerebral Edema After Ischemic Stroke. Front Neurol. 2020;11:609747. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fneur.2020.609747\u003c/span\u003e\u003cspan address=\"10.3389/fneur.2020.609747\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSarioglu O, Sarioglu FC, Capar AE, et al. Clot-based radiomics features predict first pass effect in acute ischemic stroke. Interv Neuroradiol. 2022;28:160\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/15910199211019176\u003c/span\u003e\u003cspan address=\"10.1177/15910199211019176\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXiong X, Wang J, Ke J, et al. Radiomics-based intracranial thrombus features on preoperative noncontrast CT predicts successful recanalization of mechanical thrombectomy in acute ischemic stroke. Quant Imaging Med Surg. 2023;13:682\u0026ndash;94. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21037/qims-22-599\u003c/span\u003e\u003cspan address=\"10.21037/qims-22-599\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi M, Jiang J, Gu H, et al. CT-Based Intrathrombus and Perithrombus Radiomics for Prediction of Prognosis after Endovascular Thrombectomy: A Retrospective Study across 2 Centers. AJNR Am J Neuroradiol. 2025;46:681\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3174/ajnr.A8522\u003c/span\u003e\u003cspan address=\"10.3174/ajnr.A8522\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLu J, Zhou Y, Lv W, et al. Identification of early invisible acute ischemic stroke in non-contrast computed tomography using two-stage deep-learning model. Theranostics. 2022;12:5564\u0026ndash;73. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7150/thno.74125\u003c/span\u003e\u003cspan address=\"10.7150/thno.74125\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e. Baseline clinical characteristics of the patients.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eClinical characteristics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003eTraining(MCE-)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003eTraining(MCE+)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003ep value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003eTest(MCE-)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003eTest(MCE+)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003ep value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003eValidation(MCE-)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003eValidation(MCE+)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 8px;\"\u003e\n \u003cp\u003ep value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e(n=148)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e(n=30)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e(n=70)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e(n=7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e(n=139)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e(n=12)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e71.72\u0026plusmn;11.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e69.00\u0026plusmn;11.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.213\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e62.47\u0026plusmn;11.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e62.14\u0026plusmn;17.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.946\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e69.58\u0026plusmn;11.26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e66.42\u0026plusmn;11.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.283\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eNIHSS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e15.08\u0026plusmn;8.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e10.90\u0026plusmn;6.26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e13.46\u0026plusmn;6.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e9.43\u0026plusmn;4.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.101\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e12.05\u0026plusmn;5.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e12.92\u0026plusmn;5.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.549\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eGender\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.507\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.763\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e67(45.27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e11(36.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e22(31.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e2(28.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e57(41.01)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e6(50.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e81(54.73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e19(63.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e48(68.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e5(71.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e82(58.99)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e6(50.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eAtrial fibrillation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.064\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.539\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eAbsence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e89(60.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e24(80.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e44(62.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e5(71.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e111(79.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e11(91.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003ePresence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e59(39.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e6(20.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e26(37.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e2(28.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e28(20.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e1(8.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eSmoke\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.592\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.477\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eAbsence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e118(79.73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e22(73.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e54(77.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e4(57.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e113(81.29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e8(66.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003ePresence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e30(20.27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e8(26.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e16(22.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e3(42.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e26(18.71)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e4(33.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eHypertension\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.969\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.785\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.547\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eAbsence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e36(24.32)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e8(26.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e21(30.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e3(42.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e51(36.69)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e6(50.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003ePresence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e112(75.68)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e22(73.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e49(70.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e4(57.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e88(63.31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e6(50.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eHyperlipidemia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.668\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.234\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eAbsence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e126(85.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e24(80.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e65(92.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e5(71.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e138(99.28)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e12(100.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003ePresence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e22(14.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e6(20.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e5(7.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e2(28.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e1(0.72)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eDiabetes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.362\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.247\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eAbsence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e103(69.59)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e22(73.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e56(80.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e4(57.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e108(77.70)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e7(58.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003ePresence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e45(30.41)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e8(26.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e14(20.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e3(42.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e31(22.30)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e5(41.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eCoronary disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.571\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eAbsence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e119(80.41)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e24(80.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e61(87.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e5(71.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e130(93.53)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e11(91.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003ePresence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e29(19.59)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e6(20.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e9(12.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e2(28.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e9(6.47)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e1(8.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eLocation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.839\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.104\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e0.613\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eMCA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e105(70.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e22(73.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e49(70.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e5(71.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e90(64.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e8(66.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eICA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e31(20.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e5(16.67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e20(28.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e1(14.29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e39(28.06)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e4(33.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003eMCA+ICA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e12(8.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13px;\"\u003e\n \u003cp\u003e3(10.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e1(1.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e1(14.29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14px;\"\u003e\n \u003cp\u003e10(7.19)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" style=\"width: 11px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 5px;\"\u003e\n \u003cp\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003eTabe 2\u003c/strong\u003e. Performance of LR models.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"941\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" style=\"width: 68px;\"\u003e\n \u003cp\u003eModels\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"bottom\" style=\"width: 293px;\"\u003e\n \u003cp\u003eIntra-thrombus\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"bottom\" style=\"width: 295px;\"\u003e\n \u003cp\u003ePeri-thrombus\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"bottom\" style=\"width: 285px;\"\u003e\n \u003cp\u003eCombined\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003eTraining \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003eTest \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003eValidation \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003eTraining \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003eTest \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003eValidation \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003eTraining \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003eTest \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003eValidation \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 68px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.876\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.887\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.888\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.831\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.974\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.938\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.753\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.808\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 68px;\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.968\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.741\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.626\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.976\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.908\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.891\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.993\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.927\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.869\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 68px;\"\u003e\n \u003cp\u003e95% CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.9458 - 0.9907\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.5063 - 0.9754\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.4325 - 0.8205\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.9563 - 0.9950\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.7826 - 1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.7615 - 1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.9854 - 1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.8473 - 1.0000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.7776 - 0.9611\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 68px;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.967\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.429\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.333\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.933\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.714\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.667\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.967\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.857\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 68px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.858\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.914\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.935\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.878\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.843\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.932\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.743\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.813\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 68px;\"\u003e\n \u003cp\u003ePPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.333\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.308\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.609\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.312\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.744\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.257\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 68px;\"\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.992\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.941\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.942\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.985\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.967\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.972\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.993\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.981\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.974\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 68px;\"\u003e\n \u003cp\u003eF1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.725\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.375\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.737\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.435\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.841\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.387\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.383\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 68px;\"\u003e\n \u003cp\u003eThreshold\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.196\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.376\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.257\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.182\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.353\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.2364\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.04271\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.09833\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 68px;\"\u003e\n \u003cp\u003eMCC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.415\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.328\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.718\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.485\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.857\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.836\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.456\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0.41\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Radiomics, Malignant cerebral edema, Acute Ischemic Stroke, Retrospective studies, Computed Tomography Angiography","lastPublishedDoi":"10.21203/rs.3.rs-6948356/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6948356/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cb\u003eBackground\u003c/b\u003e\u003c/p\u003e\u003cp\u003eMalignant cerebral edema (MCE) exerts a detrimental impact on clinical outcomes, underscoring the critical need for the development of more precise and objective predictive models. To accurately assess the predictive ability of radiomics and deep learning (DL) features in intrathrombus and perithrombus regions for the risk of MCE after acute ischemic stroke (AIS).\u003c/p\u003e\u003cp\u003e\u003cb\u003eMaterials and Methods\u003c/b\u003e\u003c/p\u003e\u003cp\u003eA retrospective study was conducted, enrolling 406 AIS patients who underwent admission CT before endovascular thrombectomy (EVT). Patients from Center A were randomly divided into training and testing sets at a 7:3 ratio, while those from Center B and Center C formed the external validation cohort. Regions of interest (ROIs) of thrombus and perithrombus were manually delineated and automatically expanded in margin by one pixel. 428 radiomic features were extracted from CT images of intrathrombus and perithrombus regions, and 128 DL features were obtained by inputting these images into a VGG16 architecture. Following features fusion, least absolute shrinkage and selection operator (LASSO) regression was employed for dimensionality reduction. Eleven machine learning classifiers were used for model development. Models\u0026rsquo; performance was evaluated using Matthews correlation coefficient (MCC) and area under the receiver operating characteristic curve (AUC), with AUC differences tested using DeLong\u0026rsquo;s method.\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults\u003c/b\u003e\u003c/p\u003e\u003cp\u003eMCE occurred in 49 patients (12.1%). In the validation cohort, the logistic regression (LR) models demonstrated discriminative performance with perithrombus (LR-peri: MCC\u0026thinsp;=\u0026thinsp;0.857, AUC\u0026thinsp;=\u0026thinsp;0.891), intrathrombus, (LR-intra: MCC\u0026thinsp;=\u0026thinsp;0.328, AUC\u0026thinsp;=\u0026thinsp;0.626), and combined (LR-combined: MCC\u0026thinsp;=\u0026thinsp;0.41, AUC\u0026thinsp;=\u0026thinsp;0.869) models. The LR-combined model exhibited a significantly superior predictive capacity to that of LR-intra (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\u003cp\u003e\u003cb\u003eConclusions\u003c/b\u003e\u003c/p\u003e\u003cp\u003ePerithrombus features can improve the prediction of MCE after AIS, which in turn enables the optimization of medical resource allocation.\u003c/p\u003e\u003cp\u003e\u003cb\u003eClinical relevance statement:\u003c/b\u003e\u003c/p\u003e\u003cp\u003eEmphasis is placed on the critical significance of radiomics extracted from the area in and around the thrombus in predicting MCE after AIS, which has far-reaching significance for improving patient prognosis.\u003c/p\u003e","manuscriptTitle":"Radiomics signature and deep learning signature of intrathrombus and perithrombus for prediction of Malignant Cerebral Edema after Acute Ischemic Stroke: a multicenter CT study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-06 08:04:43","doi":"10.21203/rs.3.rs-6948356/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a42ba533-95ed-480e-b3f6-534f6ed6dee8","owner":[],"postedDate":"August 6th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-08-13T08:24:40+00:00","versionOfRecord":[],"versionCreatedAt":"2025-08-06 08:04:43","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6948356","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6948356","identity":"rs-6948356","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}