Predicting the Biological Behavior of Cervical Squamous Cell Carcinoma: A Machine Learning Approach Using Apparent Transverse Relaxation Rate (R2* maps) Radiomics Nomogram

Predicting the Biological Behavior of Cervical Squamous Cell Carcinoma: A Machine Learning Approach Using Apparent Transverse Relaxation Rate (R2* maps) Radiomics Nomogram | 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 Predicting the Biological Behavior of Cervical Squamous Cell Carcinoma: A Machine Learning Approach Using Apparent Transverse Relaxation Rate (R2* maps) Radiomics Nomogram Siqi Wang, Qingling Song, Changjun Ma, Deshuo Dong, Shifeng Tian, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7082528/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 04 Apr, 2026 Read the published version in BMC Medical Imaging → Version 1 posted 11 You are reading this latest preprint version Abstract Background Accurate prediction of the biological behavior of cervical squamous cell carcinoma (CSCC) is essential for optimizing therapeutic strategies and enhancing patient outcomes. This study aims to develop and validate a radiomics nomogram based on the apparent transverse relaxation rate (R2* maps) to predict deep stromal invasion (DSI), lymph node metastasis (LNM), and lymph-vascular space invasion (LVSI) in CSCC. Material and methods A total of 136 patients with CSCC were included. Patients were divided into two groups for each clinical characteristic: DSI (n = 61) vs. non-DSI (n = 75), LNM (n = 24) vs. non-LNM (n = 112), and LVSI (n = 61) vs. non-LVSI (n = 75). Radiomic features were extracted from axial MRI scans using the ESWAN sequence and post-processed to generate R2* maps, yielding 1,476 features. Clinical factors, including age, tumor diameter, SCC antigen levels, NLR, PLR, WBC count, Hgb level, tumor differentiation grade, history of irregular vaginal bleeding, and menopausal status, were also analyzed. Feature selection was performed using PCC, univariate logistic regression, ANOVA, RFE, and Relief. Clinical risk factors were identified via logistic regression. Machine learning classifiers were used to construct radiomics models, and combined models were developed by integrating the radiomics score with clinical factors. Model performance was assessed using ROC analysis, with AUC, accuracy, specificity, and sensitivity calculated. Radiomics nomograms were constructed for each condition. Results In the training set, radiomics models outperformed clinical models for DSI (AUC 0.824 vs. 0.724), LNM (AUC 0.827 vs. 0.783), and LVSI (AUC 0.897 vs. 0.712). Combined models further improved performance, with AUC values of 0.870 for DSI, 0.837 for LNM, and 0.876 for LVSI. In the testing set, combined models outperformed clinical models for LNM (AUC 0.735 vs. 0.633) and LVSI (AUC 0.775 vs. 0.664), but not for DSI (AUC 0.619 vs. 0.677). Conclusions The R2*-based radiomics nomogram with machine learning outperforms clinical models in predicting CSCC behavior in the training cohort. Combined models boost performance further. Although validation results are mixed, this method shows potential for enhancing personalized treatment and patient management. Cervical squamous cell carcinoma Biological behavior ESWAN Radiomics Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 1 Introduction Cervical cancer remains a significant public health challenge, particularly in low- and middle-income countries, where it is the fourth most common malignancy among women [ 1 ] . Cervical squamous cell carcinoma (CSCC) is the most common pathological type of cervical cancer. Radical hysterectomy combined with lymphadenectomy was recommend as the first choice in National Comprehensive Cancer Network (NCCN) guideline for patients with stage IB1-2/IIA1 cervical cancer [ 2 ] . Deep stromal invasion (DSI), lymph-vascular space invasion (LVSI), and lymph node metastasis (LNM) are recognized pathological indicators that critically impact treatment decisions and prognosis in cervical cancer.DSI and LVSI reflect the tumor's ability to infiltrate surrounding tissues and vascular channels, signaling a more aggressive biological behavior and correlating with a higher risk of locoregional recurrence and distant metastasis [ 3 , 4 ] . According to the Sedlis criteria, patients exhibiting these intermediate-risk factors after radical surgery are recommended to undergo adjuvant therapy to improve survival outcomes [ 5 ] . However, combined modality treatment—surgery followed by adjuvant chemoradiotherapy—has been associated with increased postoperative complications and impaired quality of life, prompting ESGO/ESTRO/ESP guidelines to advocate primary chemoradiotherapy for patients identified with risk factors prior to surgery [ 6 ] . Similarly, LNM represents a high-risk feature with profound prognostic implications: node-positive patients experience a markedly reduced 5-year survival rate compared to those without nodal involvement [ 7 ] . In recognition of its impact, the 2018 FIGO(International Federation of Gynecology and Obstetrics) staging system reclassified pelvic lymph node metastasis as stage IIIC1, emphasizing its importance in staging and treatment planning [ 8 ] . The NCCN guidelines recommend concurrent chemoradiotherapy as the standard of care for patients with LNM [ 2 ] . Therefore, the early and accurate identification of DSI, LVSI, and LNM is crucial for optimizing individualized treatment decisions, minimizing overtreatment, and ultimately improving the prognosis and quality of life of patients with cervical cancer. Enhanced T2-star weighted angiography (ESWAN) is an advanced imaging technique derived from susceptibility weighted imaging (SWI). It enables contrast enhancement based on differences in magnetic susceptibility among various tissues and allows for the quantitative measurement of parameters such as phase value, R2* value, T2* value, and magnetic moment value, among others. The R2* value, representing the apparent transverse relaxation rate, is inversely correlated with tissue oxygen concentration and can serve as a noninvasive biomarker for evaluating tissue oxygenation status. In the context of tumors, increased R2* values are associated with enhanced paramagnetic effects resulting from elevated deoxyhemoglobin levels, which reflect increased oxygen consumption by rapidly proliferating tumor cells, the presence of microhemorrhages within the tumor stroma, and the development of abnormal, inefficient tumor vasculature that leads to localized hypoxia [ 9 ] . ESWAN is gradually being applied to the diagnosis of abdominal and pelvic malignancies, as well as investigations into the biological behavior of tumors [ 10 ] . Radiomics, when combined with machine learning techniques, enables the extraction and analysis of complex tumor characteristics, including heterogeneity and biological behavior, which can aid in the risk stratification and evaluation of gynecologic tumors such as cervical and endometrial cancer [ 11 , 12 ] . Nomograms are valuable predictive tools that integrate multiple clinical, radiological, and pathological variables to offer personalized risk assessments and improve decision-making in oncology [ 13 , 14 ] .The purpose of this study was to investigate the predictive effect of machine learning based on R2* maps of ESWAN sequence on the biological behavior of cervical squamous cell carcinoma (DSI, LVSI and LNM). 2 Materials and Methods 2.1 Patients This retrospective study was approved by our institutional ethics committee (Approval No. YJ-KS-KY-2025-353), with a waiver of informed consent. A total of 136 patients diagnosed with CSCC between January 2012 and November 2019 were finally included in the analysis. Inclusion criteria were:(1) age ranging from 18 to 85 years old; (2) underwent MRI examination included ESWAN sequence performed 2 weeks before surgery; (3) received radical hysterectomy combined with pelvic lymphadenectomy; (4) diagnosed with cervical squamous cell carcinoma according to postoperative pathology, with specific pathologic results including DSI, LNM, LVSI. Exclusion criteria were:(1) received chemotherapy or radiotherapy for cervical lesions before surgery; (2) pathologic results regarding DSI, LVSI and LNM were incomplete; (3) inadequate imaging quality: MRI images with significant artifacts; (4) ambiguous lesions on MRI that were indeterminate or poorly delineated;(5) with a history of other malignant tumors. A total of 183 patients were initially identified with cervical cancer. However, 47 cases were excluded due to insufficient pathological information, inadequate lesion size or image quality, or histological subtypes other than CSCC. The patient selection process is illustrated in Fig. 1. In this study, patients were classified into distinct groups based on different biological behaviors to facilitate independent analyses. Specifically, the cohort was divided according to DSI, LVSI, and LNM. The DSI positive group consisted of 61 cases with invasion depth greater than half of the cervical stroma, while the DSI negative group included 75 cases with invasion depth less than or equal to half. The LVSI positive group included 61 cases, and the LVSI negative group comprised 75 cases. Additionally, patients were categorized into an LNM positive group (24 cases) and an LNM negative group (112 cases). For model development, the patients were randomly assigned to a train dataset (n = 96) and a test dataset (n = 40) in a 7:3 ratio. 2.2 Clinical and radiological data A retrospective analysis was conducted to gather and assess clinical and radiological data from a patient cohort. Clinicopathological parameters included age, tumor diameter, squamous cell carcinoma antigen levels (SCC), tumor differentiation grade, history of irregular vaginal bleeding, menopausal status, Neutrophil-to-Lymphocyte Ratio (NLR), Platelet-to-Lymphocyte Ratio (PLR), White Blood Cell Count (WBC), and Hemoglobin (Hgb). 2.3 MRI acquisition and tumor segmentation MRI scans, including ESWAN sequences, were performed using a 1.5 T scanner (GE Signa HDXT, GE Healthcare, USA) equipped with an 8-channel phased array pelvis coil. All subjects underwent fat-suppressed T2-weighted imaging (FS-T2WI) and ESWAN imaging. The imaging parameters were as follows. 1)Fat-suppressed T2WI: TR/TE = 4000/125 ms, FOV = 40 cm×40 cm, matrix = 320×192, NEX = 4.0, slice thickness = 4 mm, slice gap = 1 mm. 2༉ESWAN: Axial 3D, matrix = 256×192, 5 echoes, TR = 17.4 ms, TE = 3.2 ms, flip angle = 15°, bandwidth = ± 50.0 kHz, FOV = 40 cm×40 cm, slice thickness = 8 mm, reconstruction thickness = 2 mm, NEX = 0.68, scan matrix = 256×160, reconstruction matrix = 512×512, parallel acquisitions acceleration factor = 2, breath hold = 21 s. Flow compensation was applied to acquire continuous axial digital images covering the entire lesion. Both the original magnitude image and the original phase image were obtained. The ESWAN sequences of all patients were post-processed using the Functool software on a GE AW 4.6 workstation to generate R2* maps in DICOM format. These maps were subsequently imported into ITK-SNAP software (v3.6.0, www.itksnap.org) for tumor segmentation. Radiologists manually delineated regions of interest (ROIs) on the R2* maps by tracing tumor margins slice by slice, using T2-weighted images (T2WI) as anatomical references to guide the segmentation process. To ensure segmentation consistency, both interobserver and intraobserver reliability were assessed by two abdominal radiologists (C.M., with 5 years of experience, and Q.S., with 7 years of experience) who were blinded to the pathological results to minimize bias. To further validate the reproducibility of the segmentation process, a randomly selected subset of 20 patient images was independently segmented by the two radiologists (S.W., with 3 years of experience, and J.L., with 3 years of experience). For each patient, two ROIs were delineated, ensuring consistency in the segmentation across both observers. 2.4 Radiomic Feature Extraction, selection and model construction A total of 120 texture features were extracted from the lesions using the uAI Research Portal Software (20240130 version, United Imaging Intelligence Co., Ltd., Shanghai, China). These features encompassed multiple categories, including First-Order Statistics, 3D Shape Features, and higher-order texture matrices such as the Gray-Level Co-occurrence Matrix (GLCM), Gray-Level Dependence Matrix (GLDM), Gray-Level Size Zone Matrix (GLSZM), Gray-Level Run Length Matrix (GLRLM), and the Neighboring Gray-Tone Difference Matrix (NGTDM). To ensure the robustness and reproducibility of the radiomics models, we initially selected radiomics features with high stability, defined as an intraclass correlation coefficient (ICC) of ≥ 0.75, both within and between observers for subsequent analysis. Given the high dimensionality of the dataset, we assessed feature redundancy by calculating Pearson correlation coefficients (PCC). Features with a PCC exceeding 0.990 were merged to enhance feature independence and reduce dimensionality. Prior to model construction, feature selection was performed using methods such as Analysis of Variance (ANOVA), Recursive Feature Elimination (RFE), and the Relief algorithm to identify the most discriminative features. A variety of classifiers were employed to construct radiomics models for predicting biological behaviors in CSCC. These classifiers included Linear Regression, Least Absolute Shrinkage and Selection Operator (LASSO), Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), Decision Tree, Random Forest, AdaBoost, Autoencoder (AE), Logistic Regression via Lasso(LR-Lasso), Gaussian Process, Naïve Bayes, and Multilayer Perceptron. Hyperparameter tuning was conducted using 5-fold cross-validation on the training dataset, with parameter adjustments guided by performance metrics obtained from the validation set. To evaluate the associations between clinical characteristics and DSI, LNM, and LVSI, both univariate and multivariate analyses were performed. Significant clinical risk factors identified through these analyses were subsequently utilized to develop and validate clinical prediction models for DSI, LNM, and LVSI. Moreover, these clinically significant independent risk factors were incorporated into the corresponding radiomics models to generate combined models. Ultimately, a radiomics nomogram integrating both the Rad-score and clinical independent risk factors was constructed. The diagnostic performance of these models was assessed using receiver operating characteristic (ROC) curves. The univariate and multivariate analyses were conducted to evaluate the associations between clinical characteristics and DSI, LNM, and LVSI. Significant clinical risk factors were used to develop and validate clinical prediction models for these outcomes. Additionally, these clinically significant independent risk factors were integrated into the corresponding radiomics models to construct combined models. Finally, a radiomics nomogram incorporating both the rad-score and clinical independent risk factors was developed, and its diagnostic efficiency was evaluated using ROC curves. Three distinct models were developed to predict each biological behavior in both the training and testing datasets. These included: (1) the radiomics model, which incorporated only the rad-score; (2) the clinical model, which integrated multiple clinical risk factors; and (3) the combined model, which encompassed all selected parameters. Each model was independently tested and evaluated on the testing dataset to assess predictive performance. All procedures above were implemented using FeAture Explorer Pro (FAE, v0.5.2) within the Python environment (v3.7.6). The overall workflow of the study—including patient selection, MRI image processing, radiomic feature extraction, model construction, and performance evaluation—is illustrated in Fig. 2. 2.5 Model performance comparison Calibration curves and the Hosmer–Lemeshow (HL) test were used to evaluate the agreement between predicted probabilities and observed outcomes, reflecting the calibration performance of the combined model. Decision curve analysis (DCA) was used to assess clinical utility by quantifying net benefits across different threshold probabilities. Shapley Additive Explanations (SHAP) analysis improved model interpretability by illustrating feature contributions both globally and locally. 2.6 Development of the Radiomics Nomogram Clinical risk factors for each biological behavior were identified and incorporated into the radiomics nomogram through univariate and multivariate logistic regression analyses. The nomogram was constructed by integrating the rad-score with these selected clinical variables using multivariable logistic regression. This nomogram was subsequently utilized to predict the probability of DSI, LNM, and LVSI on an individual basis within the primary cohort. 2.7 Statistical analysis Continuous data were expressed as the mean ± standard deviation (SD) or median with interquartile ranges, depending on the distribution of the data. Categorical data were presented as frequencies or percentages. The normality of the data distribution was assessed using the Kolmogorov-Smirnov test. Data that conformed to a normal distribution were reported as mean ± SD for continuous variables (e.g., age, tumor diameter, SCC level), while categorical variables (e.g., differentiation degree, FIGO stage, irregular vaginal bleeding, menopausal state) were summarized as frequencies or proportions. The reproducibility of radiomics features was systematically evaluated through the application of ICCs, which were utilized to quantify the degree of consistency among feature extractions. The data with bigger ICC values were used for statistical analysis. All statistical analyses were performed using R software (version 4.4.2, https://www.R-project.org/).A two-tailed p value of less than 0.05 was considered to indicate statistical significance. 3 Results 3.1 Clinicopathological characteristics of patients Table 1 presents a comparative analysis of clinicopathological and radiological characteristics between patient subgroups stratified by biological behaviors, as well as between the training and testing cohorts. With the exception of SCC in the DSI and LVSI subsets, as well as FIGO stage in the LNM subset, no significant differences were observed in the clinical and histopathological characteristics between the training and validation cohorts (P > 0.05). Table 1 Clinicopathological characteristics of the dataset. Parameters train group(n = 96) test group(n = 40) p -value DSI subsets DSI = 43 non-DSI = 53 DSI = 18 non-DSI = 22 Age 54.02 ± 12.06 52.28 ± 9.24 54.72 ± 10.61 50.95 ± 9.72 0.205 Tumor Diameter 2.4 ± 0.77 2.3 ± 0.74 5.34 ± 11.19 4.96 ± 12.99 0.871 SCC 8.86 ± 10.12 4.47 ± 5.38 10.77 ± 14.94 5.82 ± 12.81 0.011 NLR 2.55 ± 2.8 1.85 ± 0.82 1.69 ± 0.79 1.65 ± 0.66 0.118 PLR 148.23 ± 72.45 139.4 ± 42.78 122.49 ± 62.04 131.81 ± 37.59 0.734 WBC 6.46 ± 2.54 6.16 ± 1.42 5.8 ± 1.45 5.14 ± 0.94 0.228 Hgb 123.02 ± 15.23 123.59 ± 13.71 124.39 ± 13.21 123.77 ± 10.63 0.928 Differentiation Degree Poorly differentiated 14 25 8 14 0.175 Moderately differentiated 23 23 9 6 Well-differentiated 6 5 1 2 FIGO I 28 43 13 18 0.091 II-III 15 10 5 4 Irregular Vaginal Bleeding negative 22 25 6 9 0.886 positive 23 28 12 13 Menopausal State Premenopausal 18 26 7 10 0.519 Postmenopausal 25 27 11 12 LNM subsets LNM = 16 non-LNM = 80 LNM = 8 non-LNM = 32 Age 49.75 ± 12.58 53.73 ± 10.09 52.12 ± 10.97 52.78 ± 10.15 0.272 Tumor Diameter 2.49 ± 0.77 2.31 ± 0.75 8.59 ± 16.75 4.27 ± 10.76 0.429 SCC 9.56 ± 9.23 5.81 ± 7.79 14.92 ± 18.4 6.33 ± 12.24 0.062 NLR 2.11 ± 0.78 2.18 ± 2.15 1.73 ± 0.72 1.65 ± 0.72 0.853 PLR 143.8 ± 46.6 143.27 ± 60.05 123.78 ± 62.9 128.58 ± 46.88 0.871 WBC 6.27 ± 2 6.3 ± 2.01 5.87 ± 2.22 5.33 ± 0.84 0.810 Hgb 121.81 ± 12.08 123.64 ± 14.79 117.25 ± 11.45 125.75 ± 11.31 0.159 Differentiation Degree Poorly differentiated 5 34 4 18 0.719 Moderately differentiated 9 37 3 12 Well-differentiated 2 9 1 2 FIGO I 8 63 5 26 0.019 II-III 8 17 3 6 Irregular Vaginal Bleeding negative 11 34 2 13 0.386 positive 5 46 6 19 Menopausal State Premenopausal 8 36 4 13 0.739 Postmenopausal 8 44 4 19 LVSI subsets LVSI = 43 non-LVSI = 53 LVSI = 18 non-LVSI = 22 Age 53.84 ± 12.06 52.43 ± 9.27 54.72 ± 10.61 50.95 ± 9.72 0.255 Tumor Diameter 2.38 ± 0.77 2.31 ± 0.74 5.34 ± 11.19 4.96 ± 12.99 0.885 SCC 8.86 ± 10.13 4.47 ± 5.37 10.77 ± 14.94 5.82 ± 12.81 0.011 NLR 2.56 ± 2.8 1.85 ± 0.82 1.69 ± 0.79 1.65 ± 0.66 0.115 PLR 146.15 ± 71.62 141.09 ± 44.17 122.49 ± 62.04 131.81 ± 37.59 0.937 WBC 6.53 ± 2.55 6.11 ± 1.39 5.8 ± 1.45 5.14 ± 0.94 0.148 Hgb 122.95 ± 15.2 123.65 ± 13.72 124.39 ± 13.21 123.77 ± 10.63 0.898 Differentiation Degree Poorly differentiated 14 25 8 14 0.175 Moderately differentiated 23 23 9 6 Well-differentiated 6 5 1 2 FIGO I 28 43 13 18 0.091 II-III 15 10 5 4 Irregular Vaginal Bleeding negative 20 25 6 9 0.886 positive 23 28 12 13 Menopausal State Premenopausal 19 25 7 10 0.766 Postmenopausal 24 28 11 12 3.2 Selection of radiomics features and performance of radiomics models From the 1476 radiomics features extracted from the R2* maps of ESWAN for each patient, we selected 866 features that demonstrated good inter-rater agreement (ICC > 0.75) and applied them for further downscaling. Initially, the data underwent normalization, thereby dimensionality reduction via the Pearson correlation coefficient. Subsequently, distinct label groupings and imaging histology data processing were conducted in accordance with varying biological behaviors. Feature selection was executed using three methodologies: RFE, ANOVA, and the Kruskal-Wallis (KW) test. The final model classification is decided based on different classifiers and finally model construction is performed. As shown in Table 2 , three radiomics models based on different classifiers—AE, SVM, and LRLasso—were developed for predicting DSI, LNM, and LVSI. The optimal model for each subset was selected based on the highest AUC value. For the DSI subset, the AE-based model incorporating 12 radiomic features achieved the best performance, with an AUC of 0.824 and an accuracy of 0.781 in the training set, and an AUC of 0.712 with an accuracy of 0.725 in the test set (Fig. 3 A). For the LNM subset, the LRLasso-based model utilizing 27 features demonstrated the highest efficacy, achieving an AUC of 0.827 and an accuracy of 0.842 in the training set, and an AUC of 0.740 with an accuracy of 0.740 in the test set (Fig. 3 B). For the LVSI subset, the SVM-based model with 19 features showed the best performance, with an AUC of 0.897 and an accuracy of 0.865 in the training set, and an AUC of 0.750 with an accuracy of 0.750 in the test set (Fig. 3 C). Table 2 Discriminative performance of different radiomics classifiers for predicting the three biological characteristics. Different models Training group Testing group AUC SEN SPE ACC AUC SEN SPE ACC DSI AE 0.824 0.814 0.755 0.781 0.712 0.833 0.636 0.725 SVM 0.757 0.512 0.906 0.729 0.605 0.722 0.591 0.650 LRLasso 0.756 0.512 0.906 0.729 0.605 0.722 0.591 0.600 LNM AE 0.821 0.882 0.731 0.758 0.630 0.857 0.412 0.488 SVM 0.813 0.647 0.897 0.853 0.727 0.571 0.882 0.829 LRLasso 0.827 0.647 0.885 0.842 0.740 0.572 0.912 0.740 LVSI AE 0.811 0.814 0.755 0.781 0.722 0.778 0.682 0.725 SVM 0.897 0.861 0.868 0.865 0.750 0.611 0.864 0.750 LRLasso 0.862 0.767 0.849 0.813 0.737 0.889 0.500 0.675 3.4 Construction and Evaluation of Clinical, Radiomics Feature After conducting univariate and multivariate logistic analyses, significant predictors were identified for each clinical endpoint: SCC for DSI status, FIGO stage for LNM status, and SCC for LVSI status (Table 3 ). These clinical characteristics were subsequently integrated into the optimal radiomics model to develop the combined models. Table 3 Univariate and multivariate analyses of risk factors by binary logistic regression models. Univariate analysis Multivariate analysis Variable OR (95%CI) p -value OR (95%CI) p -value DSI subset age 1.02(0.99–1.06) 0.19 - Tumor diameter 1.00(0.95–1.07) 0.87 - SCC 1.05(1.01–1.11) 0.02 1.05(1.01–1.11) 0.02 Differentiation degree 1.96(0.95–4.08) 0.07 - FIGO 2.13(0.97–4.76) 0.06 - Irregular vaginal bleeding 1.12(0.56–2.22) 0.75 - Menopausal state 1.33(0.67–2.64) 0.41 - NLR 1.25(0.98–1.8) 0.14 - PLR 1.00(0.99–1.01) 0.72 - WBC 1.13(0.94–1.4) 0.22 - Hgb 1.00(0.97–1.02) 0.93 - LNM subset age 0.97(0.93–1.02) 0.22 - tumor_diameter 1.02(0.97–1.09) 0.31 - SCC 1.04(1-1.08) 0.03 - Differentiation_Degree 1.41(0.55–3.75) 0.47 - FIGO 3.27(1.29–8.3) 0.01 3.27(1.29–8.3) 0.01 irregular_vaginal_bleeding 0.61(0.25–1.48) 0.28 - menopausal_state 0.77(0.32–1.9) 0.58 - NLR 0.98(0.68–1.24) 0.91 - PLR 0.99(0.99–1.01) 0.88 - WBC 1.03(0.79–1.28) 0.79 - Hgb 0.97(0.95–1.01) 0.20 - LVSI subset age 1.02(0.99–1.05) 0.24 - tumor_diameter 1.00(0.95–1.06) 0.89 - SCC 1.05(1.01–1.11) 0.02 1.05(1.01–1.11) 0.02 Differentiation_Degree 1.96(0.95–4.08) 0.07 - FIGO 2.13(0.97–4.76) 0.06 - irregular_vaginal_bleeding 1.12(0.56–2.22) 0.75 - menopausal_state 1.18(0.6–2.34) 0.64 - NLR 1.26(0.99–1.81) 0.14 - PLR 1.00(0.99–1.01) 0.93 - WBC 1.17(0.96–1.45) 0.14 - Hgb 1.00(0.97–1.02) 0.90 - When DSI was utilized as the basis for grouping, the training cohort demonstrated an AUC value of 0.724 (95% CI: 0.609–0.815), with an accuracy rate of 0.6456, sensitivity of 0.791, and specificity of 0.528. In the validation cohort, the AUC was 0.677 (95% CI: 0.505–0.849), accompanied by an accuracy of 0.675, sensitivity of 0.833, and specificity of 0.546.When LNM was employed as the grouping criterion, the training cohort exhibited an AUC of 0.783 (95% CI: 0.639–0.927), accuracy of 0.771, sensitivity of 0.750, and specificity of 0.775. For the test cohort, the AUC was 0.633 (95% CI: 0.401–0.865), with an accuracy of 0.750, sensitivity of 0.500, and specificity of 0.813.In the model based on LVSI for grouping, the training cohort had an AUC of 0.712 (95% CI: 0.609–0.816), accuracy of 0.688, sensitivity of 0.744, and specificity of 0.642. The test cohort showed an AUC of 0.664 (95% CI: 0.493–0.835), accuracy of 0.650, sensitivity of 0.944, and specificity of 0.409. 3.5 Model Performance Comparison and Nomogram Development The calibration curve of the combined model for the four subsets demonstrated that the predicted values were in good agreement with the observed values (Fig. 4). The Hosmer–Lemeshow test showed good calibration for all models in the training cohort (DSI: P = 0.222; LNM: P = 0.851; LVSI: P = 0.277), with all P-values exceeding the 0.05 threshold, indicating strong agreement between predicted and observed outcomes. In the training cohort, the AUCs of the combined model were significantly higher than the clinical model for predicting DSI, LNM, and LVSI (DSI, AUC combined = 0.870 vs. AUC clinical = 0.724; LNM, AUC combined = 0.837 vs. AUC clinical = 0.783; LVSI, AUC combined = 0.876 vs. AUC clinical = 0.712). In the testing cohort, significant differences in the AUCs of the two models were observed for the prediction of DSI and LVSI (DSI, AUC combined = 0.619 vs. AUC clinical = 0.677; LVSI, AUC combined = 0.775 vs. AUC clinical = 0.664) (Table 4 ). The DCA for the combined models is presented in Fig. 5 . Across a range of threshold probabilities (20–40%), the combined model’s superior performance in both training and testing sets indicates enhanced clinical utility for risk prediction and decision-making. A nomogram was developed by integrating clinical, radiological, and histological features through logistic regression to visualize the combined model and enable individualized prediction of each biological characteristic. (Fig. 6 ). Table 4 Performances of models in predicting response to three biological behaviors. Different models Training group Testing group AUC(95%CI) SEN SPE ACC AUC(95%CI) SEN SPE ACC DSI subset Radiomics model 0.824[0.738–0.910] 0.814 0.755 0.781 0.712[0.546–0.878] 0.833 0.636 0.725 Clinical model 0.724[0.609–0.815] 0.791 0.528 0.646 0.677[0.505–0.849] 0.833 0.546 0.675 Combined model 0.870[0.797–0.943] 0.861 0.755 0.802 0.619[0.437–0.801] 0.778 0.546 0.650 LNM subset Radiomics model 0.827[0.717–0.938] 0.647 0.885 0.842 0.739[0.485–0.994] 0.572 0.912 0.740 Clinical model 0.783[0.639–0.927] 0.750 0.775 0.771 0.633[0.401–0.865] 0.500 0.813 0.750 Combined model 0.837[0.732–0.941] 0.706 0.846 0.853 0.735[0.4820–0.9801] 0.571 0.941 0.878 LVSI Radiomics model 0.897[0.828–0.966] 0.861 0.868 0.865 0.750[0.591–0.909] 0.611 0.864 0.750 Clinical model 0.712[0.609–0.816] 0.744 0.642 0.688 0.664[0.493–0.835] 0.944 0.409 0.650 Combined model 0.876[0.803–0.949] 0.907 0.736 0.813 0.775[0.619–0.932] 0.611 0.955 0.800 3.5 SHAP Analysis for the Combined Logistic Regression Model The SHAP method is used to interpret model outputs by evaluating the contribution of each feature to the prediction. This approach offers both global explanations, which examine feature-level impacts, and local explanations, which focus on individual predictions. SHAP dependence plots (Figs. 7, 8 , 9 A-B) reveal the influence of the top two features, with the x-axis showing feature values and the y-axis displaying SHAP values that indicate each feature's contribution. The most influential features for predicting the DSI subset are log_glcm_log.sigma.3D.2.0.mm.3D.DifferenceEntropy and wavelet_glszm_wavelet.LHH.GrayLevelNonUniformity. For the LNM subset, the key features with the highest contribution are log_gldm_log.sigma.4.0.mm.3D. DependenceNonUniformity and wavelet_glrlm_wavelet.HHH.RunPercentage. In the LVSI subset, the most significant features are wavelet_glcm_wavelet.LLH.Autocorrelation and wavelet_gldm_wavelet.LLH.LargeDependenceHighGrayLevelEmphasis. These features, identified through SHAP analysis, highlight the critical radiomic markers associated with each subset, providing valuable insights for the prediction of DSI, LNM, and LVSI. A bar plot (Figs. 7, 8 , 9 C) ranks features by their mean absolute SHAP values, highlighting those most influential on the model's predictions. SHAP waterfall plots (Figs. 7, 8 , 9 D-E) demonstrate how individual features contribute to the predicted probability of DSI for both non-DSI/LNM/LVSI (label = 0) and DSI/LNM/LVSI (label = 1) cases, showing the cumulative effect of features on predictions. The SHAP beeswarm plot (Figs. 7, 8 , 9 F) visualizes SHAP value distributions across samples, with dot color indicating feature values and spread showing the impact's magnitude and direction. Lastly, the feature interaction strength plot (Figs. 7, 8 , 9 G) evaluates how feature interactions contribute to DSI/LNM/LVSI prediction, with higher values indicating more significant interactions. In three groups, the final prediction probability f(x) integrates the baseline value and all feature contributions, combining global and local insights from the SHAP method. 4 Discussion In this study, we utilized 11 advanced machine learning classifiers to assess the preoperative prediction of three distinct biological behaviors—DSI, LNM, and LVSI—in patients with squamous cervical cancer. The radiomics features were extracted from R2* maps generated by ESWAN sequences in MRI scans. Our findings revealed that the three biological behaviors could be effectively predicted using radiomics features derived from AE, LRLasso, and SVM respectively. Moreover, the integration of radiomics features with clinical features into a combined model significantly improved the predictive performance for DSI, LNM, and LVSI. Additionally, nomograms were established as practical and precise tools for outcome prediction and high-risk patient stratification, thereby supporting the development of personalized therapeutic strategies. Beyond radiomics features, our univariate/multivariate analyses identified clinical biomarkers critical for risk stratification. The outcomes of univariate and multivariate logistic regression analyses demonstrate that SCC levels and the FIGO stage are significant predictors of DSI, LNM, and LVSI. SCC levels serve as an independent prognostic factor for DSI and LVSI. Moreover, the FIGO stage emerges as a critical predictor of LNM, while the FIGO stage is a critical predictor of LNM. These findings are consistent with the study conducted by Zhu et al. [ 15 ] , which established that elevated preoperative SCC concentrations are significantly correlated with FIGO stage, tumor size exceeding 4 cm, stromal infiltration, LNM, and LVSI. An SCC threshold of 2.75 ng/mL or higher and the presence of LVSI were identified as independent determinants influencing LNM. Additionally, the study indicates that SCC possesses a moderate predictive capacity for LNM, with an AUC of 0.703. These results highlight the potential utility of SCC as a biomarker for predicting lymphatic spread and vascular invasion in early-stage cervical carcinoma. While Zhu et al. confirmed SCC’s prognostic value, our study further demonstrates that integrating SCC with R2* radiomics (AUC = 0.845) surpasses standalone biomarker models (AUC = 0.703 ) [ 15 ] . Radiomics analysis derived from MRI has demonstrated significant potential in enhancing preoperative diagnostic accuracy for DSI, LNM, and LVSI in patients with early-stage cervical cancer. For instance, Ren et al. demonstrated that MRI-based radiomics analysis significantly improves the diagnostic accuracy of DSI compared with conventional radiologist interpretation [ 16 ] . Similarly, research by Wu et al. and Shi et al. indicates that multiparametric MRI radiomics can reliably assess LNM and LVSI, achieving high diagnostic performance metrics, including elevated AUC values [ 17 , 18 ] . The superior contrast resolution of ESWAN likely underpins the high specificity of our R2*-based nomogram (0.955 for LVSI), despite its lower sensitivity compared to multiparametric MRI [ 29 ] . The ESWAN sequence, characterized by its increased sensitivity to microhemorrhages and hemosiderin deposits, provides superior signal-to-noise ratio and tissue contrast in uterine MRI imaging relative to conventional SWI protocols [ 19 , 20 ] . This enhancement facilitates the detection of subtle lesions and enables precise evaluation of uterine pathologies such as endometrial carcinoma and adenomyosis [ 21 ] . Although it entails a marginally extended acquisition duration, the sequence's improved image resolution and diagnostic yield substantiate its utility in clinical radiological assessments [ 19 ] . CT radiomics offers valuable insights, although it may exhibit reduced sensitivity to specific tumor phenotypes compared to MRI. Ren et al. employed treatment planning CT radiomics to forecast therapeutic efficacy and hematological toxicities in patients with locally advanced cervical carcinoma [ 22 ] .Lucia et al. utilized pretreatment MRI radiomic features to predict clinical outcomes in locally advanced cervical cancer undergoing chemoradiotherapy, demonstrating that radiomic biomarkers can enhance prognostic accuracy beyond conventional clinical staging parameters [ 23 ] . This evidence indicates that MRI-based radiomics holds promise for predicting treatment response and overall survival in cervical malignancies. Furthermore, Zhang et al. reported that MRI radiomic signatures could stratify survival probabilities in patients with locally advanced cervical squamous cell carcinoma receiving concurrent chemoradiotherapy [ 24 ] . This underscores the capacity of MRI radiomic analysis to quantify tumor phenotypic features that are predictive of clinical prognosis. The findings mentioned above underscore the progression of radiomic analysis in cervical carcinoma, employing CT and MRI to advance from fundamental efficacy prediction to prognostic modeling of therapeutic response and survival outcomes. Our study has further explored several biological behaviors of cervical squamous cell carcinoma, which can potentially inform treatment planning and patient prognosis. Our results demonstrate that for DSI and LNM in cervical squamous cell carcinoma, the combined model incorporating R2* exhibits robust predictive performance. Conversely, for LVSI, the diagnostic performance remains comparable, with the AUC not surpassing that of the standalone radiomic model. Previous studies have developed various prediction models utilizing MRI data to assess deep stromal invasion in patients with cervical cancer. Ren et al. investigated the diagnostic efficacy of a radiomics model based on T2WI for preoperative assessment of DSI in cervical cancer, demonstrating superior performance compared to radiologists, particularly in terms of sensitivity and specificity. In their validation cohort, the radiomics model achieved an area under the AUC of 0.879 [ 16 ] . Similarly, Yan et al. employed a combined approach utilizing T2WI and contrast-enhanced T1-weighted imaging (CE-T1WI) within a machine learning approach for preoperative DSI prediction. Their radiomics signature achieved an AUC of 0.951 in the training dataset and 0.882 in the testing dataset [ 25 ] . These findings highlight the potential of multimodal MRI radiomics for accurate preoperative staging of deep stromal invasion in cervical cancer. Although standalone R2* radiomics underperformed multimodal MRI for DSI prediction (AUC = 0.879), our combined model achieved comparable accuracy (AUC = 0.870), suggesting clinical factors compensate for single-sequence limitations [ 25 ] . However, our findings indicate that these models outperform the ESWAN-based R2* parameter radiomics model in predicting DSI. Considering that our research focused on R2*, a less frequently employed sequence in this domain, we hypothesize that a multimodal imaging approach combining various sequences could potentially improve diagnostic accuracy. The study by Deng et al. demonstrated the potential of non-invasive differentiation between LNM and vascular endothelial growth factor (VEGF) expression in cervical cancer [ 26 ] . Their finding highlighted non-invasive LNM prediction via radiomics, while our results refine this by identifying SCC and FIGO stage as key predictors in specific subsets (e.g., SCC for LVSI, FIGO for LNM), enabling targeted risk stratification. Specifically, in the subsets with DSI and LVSI, SCC was identified as an independent prognostic factor through multivariate logistic regression analysis (P < 0.05). In the subset with LNM, FIGO staging emerged as an independent prognostic factor (P < 0.05). These findings highlight the differential predictive roles of SCC and FIGO staging in distinct pathological subsets of CSCC. Given that the sample sizes in these studies exceed the small-sample threshold (n > 30), the observed discrepancies may be attributed to the specific patient subsets analyzed, rather than random variation. Collectively, these studies highlight the necessity of incorporating patient-specific factors and subgroup stratification when assessing prognostic biomarkers in CSCC. Meta-analyses have reported an AUC of 0.83 [95%CI: 0.76–0.89] for radiomics models in detecting LNM. The diagnostic performance can be slightly enhanced by combining multiple imaging sequences, imaging histology, and clinical factors. In our single-modality study, the radiomics model achieved an AUC of 0.807 in the training cohort, which aligns with the aforementioned findings. Li et al. also highlighted that studies incorporating clinical factors tend to have higher diagnostic odds ratios (DOR) compared to those using only radiomics features. Consistent with this observation, the combined model in our study demonstrated significantly better performance than the radiomics model alone [ 27 ] . The study by Shi et al. demonstrated the potential of intratumoral and peritumoral LNM-specific analyses, as well as the development of radiomics models incorporating multi-regional features, MR-reported lymph node status, and tumor diameter. This approach achieved a high AUC in the training cohort [ 20 ] . Consistent with these findings, the combined model in the present study (AUC = 0.845) outperformed the single-radiomics model (AUC = 0.807), highlighting the benefits of integrating multiple data sources to enhance predictive efficacy. For LVSI prediction, existing T2WI-based nomograms show limited AUC values (C-index = 0.78) [ 28 ] , while our R2*-integrated model achieved superior performance (AUC = 0.897, specificity = 0.955). Although the sensitivity was 0.611, lower than that of Xiao et al.’s multiparametric approach (sensitivity = 0.90) [ 29 ] , the significantly higher specificity underscores R2*’s utility in reliably identifying high-risk cases with minimal false positives. This aligns with prior evidence that advanced sequences like ESWAN enhance microscopic tumor feature detection [ 19 , 20 ] . The combined nomogram demonstrated consistent efficacy across biological behaviors: for DSI, AUC = 0.870 (sensitivity = 0.861, specificity = 0.755); for LVSI, AUC = 0.876 (sensitivity = 0.907, specificity = 0.736). Unlike invasive biopsies requiring histologic assessment, this non-invasive tool provides comprehensive risk stratification, supporting personalized therapeutic planning [ 28 , 30 ] . In contrast to conventional histological techniques, which necessitate invasive biopsies and may be compromised by sampling variability, our radiomics-based nomogram provides a non-invasive alternative with distinct advantages. By leveraging advanced imaging features, it circumvents the procedural risks associated with invasive biopsies, such as infection and hemorrhage. Moreover, this method offers a more holistic and objective evaluation of tumor characteristics, thereby enabling clinicians to formulate well-informed decisions without exposing patients to unnecessary invasive interventions [ 30 ] . Accurate preoperative prediction of biological behavior is essential for optimizing personalized therapeutic protocols in clinical practice. For example, patients classified as high risk for LVSI or DSI via our nomogram may derive significant benefit from intensified chemotherapy or radiation therapy regimens. Conversely, those identified as low risk could potentially be spared from unnecessary aggressive interventions. This individualized stratification not only improves clinical outcomes but also reduces the morbidity associated with overtreatment [ 31 ] . The calibration curves demonstrated good agreement between predicted and observed outcomes in both training and validation cohorts, confirming the reliability of radiomic and combined models in identifying distinct biological behaviors, including DSI, LNM, and LVSI. These findings are consistent with previous studies that reported high calibration performance in radiomics-based predictive models for LVSI in cervical cancer patients [ 32 ] . DCA further validated the clinical utility of the combined model, particularly in predicting DSI and LVSI, where it provided greater net benefit across a range of threshold probabilities, in line with prior evidence suggesting superior clinical benefit of integrated models over radiomics or clinical features alone [ 33 ] . For LNM, the model demonstrated substantial net benefit primarily among low-risk cervical cancer patients, suggesting its value in supporting early intervention strategies. A similar trend of limited benefit in high-risk populations has also been reported in bladder cancer radiomics research, where model performance was constrained by the underrepresentation of high-risk cases [ 34 ] . Nonetheless, the diminished net benefit across broader or high-risk populations may stem from the limited sample size of high-risk cases within the current dataset. This underscores the imperative to increase sample sizes and enhance the representation of high-risk groups in future investigations to optimize model accuracy and external validity. Radiomics features can capture the intrinsic heterogeneity of tumors that is often imperceptible to the naked eye. In the present study, the utilization of radiomics-based nomograms and unimodal imaging (R2* maps) represents a significant innovation. When combined with machine learning classifiers, these features can more accurately reflect the overall characteristics of the tumor, thereby enhancing the model's discriminative ability. Radiomics has been extensively applied in the differential diagnosis and survival prognosis of cervical cancer. In our current investigation, the combined model incorporating logistic regression (LR)-based clinical and radiological features demonstrated superior predictive validity compared to clinical and radiological models alone in terms of DSI and LNM. Although the combined model's validity for LVSI was slightly lower than ideal, its efficacy was significantly higher than that of similar studies utilizing conventional MRI sequences, thus underscoring the significance of our study. Moreover, the combined model exhibits good clinical applicability, and the visual nomogram based on this model may serve as a valuable tool for clinical use. While the current investigation yielded encouraging findings, several methodological limitations merit consideration. Firstly, to focus on elucidating the diagnostic potential of novel imaging sequence, this study exclusively analyzed R2* parametric maps within the ESWAN sequence for single-parameter assessment. This targeted approach does not diminish the validity and reproducibility of multi-parametric imaging studies, which are recommended for future research. Secondly, the evaluation of the three biological markers—DSI, LNM, and LVSI—was conducted independently. This approach may have overlooked the potential prognostic significance of their interrelationships, warranting further investigation. Lastly, the relatively limited sample size and the retrospective study design may inherently constrain the generalizability of the findings. 5 Conclusion In summary, the combined model demonstrated robust predictive performance with respect to the biological behavior of CSCC. Moreover, visual nomograms derived from the combined model serve as valuable tools for risk stratification, individualized treatment planning, and enhanced surveillance of patients with CSCC. Abbreviations CSCC：Cervical Squamous Cell Carcinoma NCCN：National Comprehensive Cancer Network R2* maps：the apparent transverse relaxation rate DSI：Deep Stromal Invasion LNM：Lymph Node Metastasis LVSI：Lymph-Vascular Space Invasion FIGO: International Federation of Gynecology and Obstetrics ESWAN：Enhanced Susceptibility-Weighted Angiography SWI：Susceptibility Weighted Imaging SCC：Squamous Cell Carcinoma Antigen NLR：Neutrophil-to-Lymphocyte Ratio PLR：Platelet-to-Lymphocyte Ratio WBC：White Blood Cell Count Hgb：Hemoglobin FS-T2WI：Fat-Suppressed T2-Weighted Imaging CE-T1WI：contrast-enhanced T1-weighted imaging DWI：Diffusion-Weighted Imaging ROI：Region of Interest GLCM：Gray-Level Co-occurrence Matrix GLDM：Gray-Level Dependence Matrix GLSZM：Gray-Level Size Zone Matrix GLRLM：Gray-Level Run Length Matrix NGTDM：Neighboring Gray-Tone Difference Matrix ICC：Intra-class Correlation Coefficient PCC：Pearson Correlation Coefficients ANOVA：Analysis of Variance RFE：Recursive Feature Elimination LASSO：Least Absolute Shrinkage and Selection Operator SVM：Support Vector Machine LDA：Linear Discriminant Analysis AE：Autoencoder LR-Lasso：Logistic Regression via Lasso ROC：Receiver Operating Characteristic AUC：Area Under the Curve HL test：Hosmer–Lemeshow Test DCA：Decision Curve Analysis SHAP：Shapley Additive Explanations SD：Standard Deviation KW test：Kruskal-Wallis Test VEGF：vascular endothelial growth factor DOR: diagnostic odds ratios Declarations Acknowledgements The authors thank all healthcare professionals who contributed to this study. Author contributions S.W., Q.S., and C.M. were responsible for case collection. Image evaluation was performed by C.M., while image processing and ROI delineation were conducted by S.W. and Q.S. The manuscript was drafted by S.W., and critically revised by D.D., S.T., and A.L. for important intellectual content. Funding The study is supported by the Hospital Fund of the First Affiliated Hospital of Dalian Medical University (2021HZ015) and the Dalian Municipal Medical Science Research Program (2023DF038). Data availability Data are available from the corresponding authors upon reasonable request. Ethics approval and consent to participate The study was conducted in accordance with the Declaration of Helsinki. Ethical approval was obtained from the Ethics Committee of the First Affiliated Hospital of Dalian Medical University (Approval No. YJ-KS-KY-2025-353). Written informed consent was obtained from all participants prior to inclusion in the study. Consent for publication Not applicable. Competing interests The authors declare no competing interests. References Duncan J, Harris M, Skyers N, Bailey A, Figueroa JP. A Call for Low- and Middle-Income Countries to Commit to the Elimination of Cervical Cancer. Lancet Reg Health Am. 2021;2:100036. doi:10.1016/j.lana.2021.100036 Abu-Rustum NR, Yashar CM, Arend R, et al. NCCN Guidelines® Insights: Cervical Cancer, Version 1.2024. J Natl Compr Canc Netw. 2023;21(12):1224-1233. doi:10.6004/jnccn.2023.0062. Cao L, Wen H, Feng Z, Han X, Zhu J, Wu X:Role of adjuvant therapy after radical hysterectomy in intermediate-risk, early-stage cervical cancer. International journal of gynecological cancer :official journal of the International Gynecological Cancer Society 2021, 31(1):52-58. Zhu J, Cao L, Wen H, Bi R, Wu X, Ke G:The clinical and prognostic implication of deep stromal invasion in cervical cancer patients undergoing radical hysterectomy. Journal of Cancer 2020, 11(24):7368-7377. Cao L, Wen H, Feng Z, Han X, Zhu J, Wu X. Role of adjuvant therapy after radical hysterectomy in intermediate-risk, early-stage cervical cancer. Int J Gynecol Cancer. 2021;31(1):52-58. doi:10.1136/ijgc-2020-001974. Cibula D, Rosaria Raspollini M, Planchamp F, Centeno C, Chargari C, Felix A, Fischerová D, Jahnn-Kuch D, Joly F, Kohler C et al:ESGO/ESTRO/ESP Guidelines for the management of patients with cervical cancer - Update 2023. Radiotherapy and oncology :journal of the European Society for Therapeutic Radiology and Oncology 2023, 184:109682. Xu Y, Zhang M, Zhang J, et al. Neoadjuvant chemotherapy increases the 5-year overall survival of patients with resectable cervical cancer: A systematic review and meta-analysis. Taiwan J Obstet Gynecol. 2021;60(3):433-441. doi:10.1016/j.tjog.2021.03.008. Kim J, Cho Y, Kim N, et al. Magnetic resonance imaging-based validation of the 2018 FIGO staging system in patients treated with definitive radiotherapy for locally advanced cervix cancer. Gynecol Oncol. 2021;160(3):735-741. doi:10.1016/j.ygyno.2020.12.012 . Xin JY, Gao SS, Liu JG, Sun CF, Han Y, Sun XH, Wang XZ, Wang B:The value of ESWAN in diagnosis and differential diagnosis of prostate cancer:Preliminary study. Magnetic resonance imaging 2017, 44:26-31. Han X, Sun M, Wang M, Fan R, Chen D, Xie L, Liu A:The enhanced T(2) star weighted angiography (ESWAN) value for differentiating borderline from malignant epithelial ovarian tumors. European journal of radiology 2019, 118:187-193. Günakan E, Atan S, Haberal AN, Küçükyıldız İ A, Gökçe E, Ayhan A:A novel prediction method for lymph node involvement in endometrial cancer:machine learning. International journal of gynecological cancer :official journal of the International Gynecological Cancer Society 2019, 29(2):320-324. Chu R, Zhang Y, Qiao X, Xie L, Chen W, Zhao Y, Xu Y, Yuan Z, Liu X, Yin A et al:Risk Stratification of Early-Stage Cervical Cancer with Intermediate-Risk Factors:Model Development and Validation Based on Machine Learning Algorithm. The oncologist 2021, 26(12):e2217-e2226. Sun ZD, Zhang Y, Yang YS, et al. Construction and validation of a novel nomogram for prediction of lymph node metastasis in HER2-positive breast cancer: based on the optimal number of examined lymph nodes for accurate nodal staging. BMC Womens Health. 2025;25(1):132. Published 2025 Mar 22. doi:10.1186/s12905-025-03663-w Guo Q, Gao Y, Lin Y, et al. A nomogram of preoperative indicators predicting lymph vascular space invasion in cervical cancer. Arch Gynecol Obstet. 2024;309(5):2079-2087. doi:10.1007/s00404-024-07385-6 Zhu C, Zhang W, Wang X, Jiao L, Chen L, Jiang J. Predictive value of preoperative serum squamous cell carcinoma antigen level for lymph node metastasis in early-stage cervical squamous cell carcinoma. Medicine (Baltimore). 2021;100(33):e26960. doi:10.1097/MD.0000000000026960 Ren J, Li Y, Yang JJ, et al. MRI-based radiomics analysis improves preoperative diagnostic performance for the depth of stromal invasion in patients with early stage cervical cancer. Insights Imaging. 2022;13(1):17. Published 2022 Jan 29. doi:10.1186/s13244-022-01156-0 Wu Q, Shi D, Dou S, et al. Radiomics Analysis of Multiparametric MRI evaluates the pathological features of cervical squamous cell carcinoma. J Magn Reson Imaging. 2019;49(4):1141–8. Shi J, Dong Y, Jiang W, et al. MRI-based peritumoral radiomics analysis for preoperative prediction of lymph node metastasis in early-stage cervical cancer: a multi-center study. Magn Reson Imaging. 2022;88:1–8. Guo LF, Wang G, Zhu XY, Liu C, Cui L. Comparison of ESWAN, SWI-SPGR, and 2D T2*-weighted GRE sequence for depicting cerebral microbleeds. Clin Neuroradiol. 2013;23(2):121-127. doi:10.1007/s00062-012-0185-7 Vanaerde O, Budzik JF, Mackowiak A, et al. Comparison between enhanced susceptibility-weighted angiography and time of flight sequences in the detection of arterial occlusion in acute ischemic stroke. J Neuroradiol. 2017;44(3):210-216.doi:10.1016/j.neurad.2017.01.12510.1016/j.neurad.2017.01.125 Ghosh A, Singh T, Bagga R, Srinivasan R, Singla V, Khandelwal N. T2 relaxometry mapping in demonstrating layered uterine architecture: parameter optimization and utility in endometrial carcinoma and adenomyosis: a feasibility study. Br J Radiol. 2018;91(1081):20170377. doi:10.1259/bjr.20170377 Ren K, Shen L, Qiu J, et al. Treatment planning computed tomography radiomics for predicting treatment outcomes and haematological toxicities in locally advanced cervical cancer treated with radiotherapy: A retrospective cohort study. BJOG. 2023;130(2):222-230. doi:10.1111/1471-0528.17285 Lucia F, Bourbonne V, Pleyers C, et al. Multicentric development and evaluation of 18F-FDG PET/CT and MRI radiomics models to predict para-aortic lymph node involvement in locally advanced cervical cancer. Eur J Nucl Med Mol Imaging. 2023;50(8):2514-2528. doi:10.1007/s00259-023-06180-w Zhang X, Zhao J, Zhang Q, et al. MRI-based radiomics value for predicting the survival of patients with locally advanced cervical squamous cell cancer treated with concurrent chemoradiotherapy. Cancer Imaging. 2022;22(1):35. Published 2022 Jul 16. doi:10.1186/s40644-022-00474-2 Yan H, Huang G, Yang Z, Chen Y, Xiang Z. Machine Learning-Based Multiparametric Magnetic Resonance Imaging Radiomics Model for Preoperative Predicting the Deep Stromal Invasion in Patients with Early Cervical Cancer. J Imaging Inform Med. 2024;37(1):230-246. doi:10.1007/s10278-023-00906-w Deng X, Liu M, Sun J, et al. Feasibility of MRI-based radiomics features for predicting lymph node metastases and VEGF expression in cervical cancer. Eur J Radiol. 2021;134:109429. Li L, Zhang J, Zhe X, et al. A meta-analysis of MRI-based radiomic features for predicting lymph node metastasis in patients with cervical cancer. Eur J Radiol. 2022;151:110243. Guo L, Zhang R, Xu Y, et al. Predicting the status of lymphovascular space invasion using quantitative parameters from synthetic MRI in cervical squamous cell carcinoma without lymphatic metastasis. Front Oncol. 2024;14:1304793. Published 2024 Feb 6. doi:10.3389/fonc.2024.1304793 Xiao M, Li Y, Ma F, Zhang G, Qiang J. Multiparametric MRI radiomics nomogram for predicting lymph-vascular space invasion in early-stage cervical cancer. Br J Radiol. 2022;95(1134):20211076. doi:10.1259/bjr.20211076 Wu L, Li S, Li S, Lin Y, Wei D. Preoperative magnetic resonance imaging-radiomics in cervical cancer: a systematic review and meta-analysis. Front Oncol. 2024;14:1416378. Published 2024 Jul 4. doi:10.3389/fonc.2024.1416378 Tang S, Yen A, Wang K, Albuquerque K, Wang J. Progression-Free Survival Prediction for Locally Advanced Cervical Cancer After Chemoradiotherapy With MRI-based Radiomics. Clin Oncol (R Coll Radiol). 2025;38:103702. doi:10.1016/j.clon.2024.103702 Ma NN, Wang T, Lv YN, Li SD. An MRI radiomics-based model for the prediction of invasion of the lymphovascular space in patients with cervical cancer. Front Oncol. 2024;14:1394427. Published 2024 Jul 5. doi:10.3389/fonc.2024.1394427 Huang G, Cui Y, Wang P, et al. Multi-Parametric Magnetic Resonance Imaging-Based Radiomics Analysis of Cervical Cancer for Preoperative Prediction of Lymphovascular Space Invasion. Front Oncol. 2022;11:663370. Published 2022 Jan 12. doi:10.3389/fonc.2021.663370 Wu S, Zheng J, Li Y, et al. Development and Validation of an MRI-Based Radiomics Signature for the Preoperative Prediction of Lymph Node Metastasis in Bladder Cancer. EBioMedicine. 2018;34:76-84. doi:10.1016/j.ebiom.2018.07.029 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 04 Apr, 2026 Read the published version in BMC Medical Imaging → Version 1 posted Editorial decision: Revision requested 09 Mar, 2026 Reviews received at journal 07 Mar, 2026 Reviewers agreed at journal 23 Feb, 2026 Reviewers agreed at journal 14 Jan, 2026 Reviews received at journal 05 Sep, 2025 Reviewers agreed at journal 26 Aug, 2025 Reviewers invited by journal 24 Jul, 2025 Editor invited by journal 15 Jul, 2025 Editor assigned by journal 11 Jul, 2025 Submission checks completed at journal 11 Jul, 2025 First submitted to journal 09 Jul, 2025 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-7082528","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":491286159,"identity":"67371c8c-482d-4618-8879-bd080e6d14ae","order_by":0,"name":"Siqi Wang","email":"","orcid":"","institution":"the First Affiliated Hospital of Dalian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Siqi","middleName":"","lastName":"Wang","suffix":""},{"id":491286160,"identity":"d1293532-9b90-4ad0-8200-bea98314bc6c","order_by":1,"name":"Qingling Song","email":"","orcid":"","institution":"the First Affiliated Hospital of Dalian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Qingling","middleName":"","lastName":"Song","suffix":""},{"id":491286161,"identity":"11576146-6164-4ad1-a94d-69571a1c567a","order_by":2,"name":"Changjun Ma","email":"","orcid":"","institution":"the First Affiliated Hospital of Dalian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Changjun","middleName":"","lastName":"Ma","suffix":""},{"id":491286162,"identity":"ffc3b069-5af6-475e-938f-a46751550bdf","order_by":3,"name":"Deshuo Dong","email":"","orcid":"","institution":"the First Affiliated Hospital of Dalian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Deshuo","middleName":"","lastName":"Dong","suffix":""},{"id":491286163,"identity":"9477b394-0a47-4340-bfb1-464620535364","order_by":4,"name":"Shifeng Tian","email":"","orcid":"","institution":"the First Affiliated Hospital of Dalian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shifeng","middleName":"","lastName":"Tian","suffix":""},{"id":491286164,"identity":"aa44cde3-ad5d-450e-a051-1ea19fe6844b","order_by":5,"name":"Ailian Liu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA00lEQVRIiWNgGAWjYBACPmYgkVAgwcDA3mAAEmBsIKSFDazFAKiF5wCxWsAkSLFEArFa2HkMPzwwsJA3l3y8dTMPg43shgPMzx7gdxiPMdB8CcOds9PKbvMwpBlvOMBmbkBAiwFIC+OG2zlmQC2HEzcc4GGTIGTLD6AW+w03z4C0/CdKixnIlsQNN3hAWg4Qo4WtzAKoJXnDmbSym3MMko1nHmYzw6uFn//w5ps/KupsNxw/vO3Gmwo72b7jzc/wakEDoKBiJkH9KBgFo2AUjALsAAC0BUDVQ+4pmgAAAABJRU5ErkJggg==","orcid":"","institution":"the First Affiliated Hospital of Dalian Medical University","correspondingAuthor":true,"prefix":"","firstName":"Ailian","middleName":"","lastName":"Liu","suffix":""}],"badges":[],"createdAt":"2025-07-09 09:53:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7082528/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7082528/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12880-026-02333-z","type":"published","date":"2026-04-04T15:59:33+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":87827066,"identity":"4756594d-32ef-4db2-a82c-cee41f087057","added_by":"auto","created_at":"2025-07-29 11:53:55","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":118273,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart detailing the inclusion and exclusion criteria for the study.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7082528/v1/00ea6bce1925fcb38715a2fa.jpg"},{"id":87827215,"identity":"52a8b70a-638f-47b6-a463-e63212dc6495","added_by":"auto","created_at":"2025-07-29 11:54:04","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":100899,"visible":true,"origin":"","legend":"\u003cp\u003eFlow chart of this study.\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7082528/v1/28a1a29e753ed1892abc3bba.jpg"},{"id":87827065,"identity":"cef46526-00f2-4a8a-9be3-3d5637fc963d","added_by":"auto","created_at":"2025-07-29 11:53:55","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":115109,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves of radiomics models and combined models.\u003c/p\u003e\n\u003cp\u003ePanels A–C correspond to the radiomics models for DSI, LNM, and LVSI, respectively, while panels D–F correspond to the combined models of radiomics and clinical features for DSI, LNM, and LVSI, respectively.\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7082528/v1/f8c2c32121bc2d09b3727a2f.jpg"},{"id":87827150,"identity":"18551c93-289b-4746-bcb7-6633572b6950","added_by":"auto","created_at":"2025-07-29 11:54:00","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":87805,"visible":true,"origin":"","legend":"\u003cp\u003eCalibration curves for the combined model.\u003c/p\u003e\n\u003cp\u003e(A) DSI; (B) LNM; (C) LVSI. For the training group (1) and the test group (2), the values of each predictor are converted into risk scores based on the corresponding “dots” at the top of the nomogram. The individual risk scores of these predictors are then summed to obtain the “total score.” The corresponding predicted probabilities at the bottom of the column plot represent the DSI, LNM, and LVSI, respectively.\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7082528/v1/9ef0eff1be76e4fb51e52ca0.jpg"},{"id":87827151,"identity":"e3320e69-8bff-496d-97e5-3a6e42eb08a7","added_by":"auto","created_at":"2025-07-29 11:54:00","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":117117,"visible":true,"origin":"","legend":"\u003cp\u003eDecision curves for the combined model.\u003c/p\u003e\n\u003cp\u003e(A) DSI; (B) LNM; (C) LVSI. The training group is indicated by (1), and the test group by (2). The y-axis represents the net benefit, while the x-axis indicates the threshold probability.\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7082528/v1/97ffe9d6667388ada7be19c8.jpg"},{"id":87827160,"identity":"8068ea0b-1a05-4a45-aeff-a6022cc634d2","added_by":"auto","created_at":"2025-07-29 11:54:01","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":86644,"visible":true,"origin":"","legend":"\u003cp\u003eNomograms based on the combined models.(A) DSI; (B) LNM; (C) LVSI.\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7082528/v1/f15f6fd40848c28c8a6c25c2.jpg"},{"id":87827213,"identity":"da878fd0-efc6-4ca5-9efd-af847700137c","added_by":"auto","created_at":"2025-07-29 11:54:04","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":107633,"visible":true,"origin":"","legend":"\u003cp\u003eGlobal model explanation by the SHAP method for DSI combined model.\u003c/p\u003e\n\u003cp\u003e(A-B)SHAP dependence plots for the two most influential features in DSI prediction. X-axis shows the feature value, and Y-axis shows its SHAP value (i.e., contribution to the model output).(C)Bar plot showing the global feature importance for DSI prediction based on the mean absolute SHAP values. Features ranked higher have stronger overall influence on the model's output.(D-E)SHAP waterfall plots for representative cases of both non-DSI (label = 0) and DSI (label = 1) are presented. The bars in each plot represent the contributions of individual features to the predicted probability of DSI. The plot visualizes how each feature pushes the prediction toward or away from DSI. (F)SHAP beeswarm plot for DSI prediction. Each dot represents one sample, color indicates the feature value (red = high, blue = low), and horizontal spread shows the magnitude and direction of feature impact. (G)Feature interaction strength plot for the DSI model. Higher values suggest stronger feature interactions in contributing to DSI prediction. SHAP: Shapley Additive explanation; SCC: squamous cell carcinoma antigen.\u003c/p\u003e","description":"","filename":"7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7082528/v1/bd71abb4a80ed8714c0b9741.jpg"},{"id":87827163,"identity":"bdb0aaca-1c3c-4977-95aa-1c8fae92d144","added_by":"auto","created_at":"2025-07-29 11:54:01","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":127071,"visible":true,"origin":"","legend":"\u003cp\u003eGlobal model explanation by the SHAP method for LNM combined model.\u003c/p\u003e\n\u003cp\u003e(A-B) SHAP dependence plots of two most influential features in the LNM model. It depicts how varying this feature affects its SHAP value and thus the prediction. (C)Bar plot showing the global feature importance for LNM prediction based on SHAP values. Features with higher bars have greater overall influence on predicting lymph node metastasis. (D-E) SHAP waterfall plots for representative non-LNM and LNM patients. The plots show how individual features contribute to the model’s prediction, with bars indicating the direction and magnitude of each feature’s impact on the likelihood of lymph node metastasis. (F) SHAP beeswarm plot for LNM prediction. Each point represents a sample, colored by the feature value. The SHAP value (x-axis) indicates the extent to which the feature contributes to predicting LNM. (G) Feature interaction strength plot for LNM prediction. Larger values imply stronger interactions between features when predicting LNM status. SHAP: Shapley Additive explanation; FIGO: International Federation of Gynecology and Obstetrics Staging System.\u003c/p\u003e","description":"","filename":"8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7082528/v1/771272cda58fcdd16e1909ea.jpg"},{"id":87827062,"identity":"9f1812f5-4a43-4904-a63b-98cc5b7fd239","added_by":"auto","created_at":"2025-07-29 11:53:53","extension":"jpg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":119297,"visible":true,"origin":"","legend":"\u003cp\u003eGlobal model explanation by the SHAP method for LVSI combined model.\u003c/p\u003e\n\u003cp\u003e(A-B)SHAP dependence plots for the two most influential features. Each point represents a sample. The x-axis shows the actual feature value, and the y-axis shows its SHAP value, indicating the direction and strength of influence on the LVSI prediction. (C)Bar plot showing the global feature importance based on the mean absolute SHAP values. Features at the top contribute the most to the model's predictions for LVSI classification. (D-E)SHAP waterfall plots for representative LVSI and non-LVSI samples. These plots decompose the model output into individual feature contributions, showing how positive SHAP values push the prediction toward LVSI, while negative values push it toward non-LVSI. (F)SHAP beeswarm plot illustrating the distribution of SHAP values for each feature across all samples. Color represents the feature value (red: high, blue: low), and horizontal spread indicates the effect magnitude. (G)Feature interaction strength plot. Higher scores indicate stronger interactions between each feature and all other features in predicting LVSI status. SHAP: Shapley Additive explanation; SCC: squamous cell carcinoma antigen;\u003c/p\u003e","description":"","filename":"9.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7082528/v1/e5bc91dcf96e2ebeb7041bdd.jpg"},{"id":106343507,"identity":"d531985c-d884-4ce1-85ab-7962bf32fc08","added_by":"auto","created_at":"2026-04-07 16:06:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2537243,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7082528/v1/602c3a45-1f7e-4c25-93e3-8d89ca8b1fcc.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Predicting the Biological Behavior of Cervical Squamous Cell Carcinoma: A Machine Learning Approach Using Apparent Transverse Relaxation Rate (R2* maps) Radiomics Nomogram","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eCervical cancer remains a significant public health challenge, particularly in low- and middle-income countries, where it is the fourth most common malignancy among women \u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. Cervical squamous cell carcinoma (CSCC) is the most common pathological type of cervical cancer. Radical hysterectomy combined with lymphadenectomy was recommend as the first choice in National Comprehensive Cancer Network (NCCN) guideline for patients with stage IB1-2/IIA1 cervical cancer \u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. Deep stromal invasion (DSI), lymph-vascular space invasion (LVSI), and lymph node metastasis (LNM) are recognized pathological indicators that critically impact treatment decisions and prognosis in cervical cancer.DSI and LVSI reflect the tumor's ability to infiltrate surrounding tissues and vascular channels, signaling a more aggressive biological behavior and correlating with a higher risk of locoregional recurrence and distant metastasis \u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e. According to the Sedlis criteria, patients exhibiting these intermediate-risk factors after radical surgery are recommended to undergo adjuvant therapy to improve survival outcomes \u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e. However, combined modality treatment\u0026mdash;surgery followed by adjuvant chemoradiotherapy\u0026mdash;has been associated with increased postoperative complications and impaired quality of life, prompting ESGO/ESTRO/ESP guidelines to advocate primary chemoradiotherapy for patients identified with risk factors prior to surgery \u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e. Similarly, LNM represents a high-risk feature with profound prognostic implications: node-positive patients experience a markedly reduced 5-year survival rate compared to those without nodal involvement \u003csup\u003e[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e. In recognition of its impact, the 2018 FIGO(International Federation of Gynecology and Obstetrics) staging system reclassified pelvic lymph node metastasis as stage IIIC1, emphasizing its importance in staging and treatment planning \u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e. The NCCN guidelines recommend concurrent chemoradiotherapy as the standard of care for patients with LNM \u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. Therefore, the early and accurate identification of DSI, LVSI, and LNM is crucial for optimizing individualized treatment decisions, minimizing overtreatment, and ultimately improving the prognosis and quality of life of patients with cervical cancer.\u003c/p\u003e\u003cp\u003eEnhanced T2-star weighted angiography (ESWAN) is an advanced imaging technique derived from susceptibility weighted imaging (SWI). It enables contrast enhancement based on differences in magnetic susceptibility among various tissues and allows for the quantitative measurement of parameters such as phase value, R2* value, T2* value, and magnetic moment value, among others. The R2* value, representing the apparent transverse relaxation rate, is inversely correlated with tissue oxygen concentration and can serve as a noninvasive biomarker for evaluating tissue oxygenation status. In the context of tumors, increased R2* values are associated with enhanced paramagnetic effects resulting from elevated deoxyhemoglobin levels, which reflect increased oxygen consumption by rapidly proliferating tumor cells, the presence of microhemorrhages within the tumor stroma, and the development of abnormal, inefficient tumor vasculature that leads to localized hypoxia \u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e. ESWAN is gradually being applied to the diagnosis of abdominal and pelvic malignancies, as well as investigations into the biological behavior of tumors \u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eRadiomics, when combined with machine learning techniques, enables the extraction and analysis of complex tumor characteristics, including heterogeneity and biological behavior, which can aid in the risk stratification and evaluation of gynecologic tumors such as cervical and endometrial cancer\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e. Nomograms are valuable predictive tools that integrate multiple clinical, radiological, and pathological variables to offer personalized risk assessments and improve decision-making in oncology\u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e.The purpose of this study was to investigate the predictive effect of machine learning based on R2* maps of ESWAN sequence on the biological behavior of cervical squamous cell carcinoma (DSI, LVSI and LNM).\u003c/p\u003e"},{"header":"2 Materials and Methods","content":"\u003cdiv id=\"Sec3\"\u003e\n \u003ch2\u003e2.1 Patients\u003c/h2\u003e\n \u003cp\u003eThis retrospective study was approved by our institutional ethics committee (Approval No. YJ-KS-KY-2025-353), with a waiver of informed consent. A total of 136 patients diagnosed with CSCC between January 2012 and November 2019 were finally included in the analysis. Inclusion criteria were:(1) age ranging from 18 to 85 years old; (2) underwent MRI examination included ESWAN sequence performed 2 weeks before surgery; (3) received radical hysterectomy combined with pelvic lymphadenectomy; (4) diagnosed with cervical squamous cell carcinoma according to postoperative pathology, with specific pathologic results including DSI, LNM, LVSI. Exclusion criteria were:(1) received chemotherapy or radiotherapy for cervical lesions before surgery; (2) pathologic results regarding DSI, LVSI and LNM were incomplete; (3) inadequate imaging quality: MRI images with significant artifacts; (4) ambiguous lesions on MRI that were indeterminate or poorly delineated;(5) with a history of other malignant tumors. A total of 183 patients were initially identified with cervical cancer. However, 47 cases were excluded due to insufficient pathological information, inadequate lesion size or image quality, or histological subtypes other than CSCC. The patient selection process is illustrated in Fig.\u0026nbsp;1.\u003c/p\u003e\n \u003cp\u003eIn this study, patients were classified into distinct groups based on different biological behaviors to facilitate independent analyses. Specifically, the cohort was divided according to DSI, LVSI, and LNM. The DSI positive group consisted of 61 cases with invasion depth greater than half of the cervical stroma, while the DSI negative group included 75 cases with invasion depth less than or equal to half. The LVSI positive group included 61 cases, and the LVSI negative group comprised 75 cases. Additionally, patients were categorized into an LNM positive group (24 cases) and an LNM negative group (112 cases). For model development, the patients were randomly assigned to a train dataset (n\u0026thinsp;=\u0026thinsp;96) and a test dataset (n\u0026thinsp;=\u0026thinsp;40) in a 7:3 ratio.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\"\u003e\n \u003ch2\u003e2.2 Clinical and radiological data\u003c/h2\u003e\n \u003cp\u003eA retrospective analysis was conducted to gather and assess clinical and radiological data from a patient cohort. Clinicopathological parameters included age, tumor diameter, squamous cell carcinoma antigen levels (SCC), tumor differentiation grade, history of irregular vaginal bleeding, menopausal status, Neutrophil-to-Lymphocyte Ratio (NLR), Platelet-to-Lymphocyte Ratio (PLR), White Blood Cell Count (WBC), and Hemoglobin (Hgb).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec5\"\u003e\n \u003ch2\u003e2.3 MRI acquisition and tumor segmentation\u003c/h2\u003e\n \u003cp\u003eMRI scans, including ESWAN sequences, were performed using a 1.5 T scanner (GE Signa HDXT, GE Healthcare, USA) equipped with an 8-channel phased array pelvis coil. All subjects underwent fat-suppressed T2-weighted imaging (FS-T2WI) and ESWAN imaging. The imaging parameters were as follows. 1)Fat-suppressed T2WI: TR/TE\u0026thinsp;=\u0026thinsp;4000/125 ms, FOV\u0026thinsp;=\u0026thinsp;40 cm\u0026times;40 cm, matrix\u0026thinsp;=\u0026thinsp;320\u0026times;192, NEX\u0026thinsp;=\u0026thinsp;4.0, slice thickness\u0026thinsp;=\u0026thinsp;4 mm, slice gap\u0026thinsp;=\u0026thinsp;1 mm. 2༉ESWAN: Axial 3D, matrix\u0026thinsp;=\u0026thinsp;256\u0026times;192, 5 echoes, TR\u0026thinsp;=\u0026thinsp;17.4 ms, TE\u0026thinsp;=\u0026thinsp;3.2 ms, flip angle\u0026thinsp;=\u0026thinsp;15\u0026deg;, bandwidth\u0026thinsp;=\u0026thinsp;\u0026plusmn;\u0026thinsp;50.0 kHz, FOV\u0026thinsp;=\u0026thinsp;40 cm\u0026times;40 cm, slice thickness\u0026thinsp;=\u0026thinsp;8 mm, reconstruction thickness\u0026thinsp;=\u0026thinsp;2 mm, NEX\u0026thinsp;=\u0026thinsp;0.68, scan matrix\u0026thinsp;=\u0026thinsp;256\u0026times;160, reconstruction matrix\u0026thinsp;=\u0026thinsp;512\u0026times;512, parallel acquisitions acceleration factor\u0026thinsp;=\u0026thinsp;2, breath hold\u0026thinsp;=\u0026thinsp;21 s. Flow compensation was applied to acquire continuous axial digital images covering the entire lesion. Both the original magnitude image and the original phase image were obtained.\u003c/p\u003e\n \u003cp\u003eThe ESWAN sequences of all patients were post-processed using the Functool software on a GE AW 4.6 workstation to generate R2* maps in DICOM format. These maps were subsequently imported into ITK-SNAP software (v3.6.0, www.itksnap.org) for tumor segmentation. Radiologists manually delineated regions of interest (ROIs) on the R2* maps by tracing tumor margins slice by slice, using T2-weighted images (T2WI) as anatomical references to guide the segmentation process. To ensure segmentation consistency, both interobserver and intraobserver reliability were assessed by two abdominal radiologists (C.M., with 5 years of experience, and Q.S., with 7 years of experience) who were blinded to the pathological results to minimize bias. To further validate the reproducibility of the segmentation process, a randomly selected subset of 20 patient images was independently segmented by the two radiologists (S.W., with 3 years of experience, and J.L., with 3 years of experience). For each patient, two ROIs were delineated, ensuring consistency in the segmentation across both observers.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec6\"\u003e\n \u003ch2\u003e2.4 Radiomic Feature Extraction, selection and model construction\u003c/h2\u003e\n \u003cp\u003eA total of 120 texture features were extracted from the lesions using the uAI Research Portal Software (20240130 version, United Imaging Intelligence Co., Ltd., Shanghai, China). These features encompassed multiple categories, including First-Order Statistics, 3D Shape Features, and higher-order texture matrices such as the Gray-Level Co-occurrence Matrix (GLCM), Gray-Level Dependence Matrix (GLDM), Gray-Level Size Zone Matrix (GLSZM), Gray-Level Run Length Matrix (GLRLM), and the Neighboring Gray-Tone Difference Matrix (NGTDM).\u003c/p\u003e\n \u003cp\u003eTo ensure the robustness and reproducibility of the radiomics models, we initially selected radiomics features with high stability, defined as an intraclass correlation coefficient (ICC) of \u0026ge;\u0026thinsp;0.75, both within and between observers for subsequent analysis. Given the high dimensionality of the dataset, we assessed feature redundancy by calculating Pearson correlation coefficients (PCC). Features with a PCC exceeding 0.990 were merged to enhance feature independence and reduce dimensionality. Prior to model construction, feature selection was performed using methods such as Analysis of Variance (ANOVA), Recursive Feature Elimination (RFE), and the Relief algorithm to identify the most discriminative features. A variety of classifiers were employed to construct radiomics models for predicting biological behaviors in CSCC. These classifiers included Linear Regression, Least Absolute Shrinkage and Selection Operator (LASSO), Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), Decision Tree, Random Forest, AdaBoost, Autoencoder (AE), Logistic Regression via Lasso(LR-Lasso), Gaussian Process, Na\u0026iuml;ve Bayes, and Multilayer Perceptron. Hyperparameter tuning was conducted using 5-fold cross-validation on the training dataset, with parameter adjustments guided by performance metrics obtained from the validation set.\u003c/p\u003e\n \u003cp\u003eTo evaluate the associations between clinical characteristics and DSI, LNM, and LVSI, both univariate and multivariate analyses were performed. Significant clinical risk factors identified through these analyses were subsequently utilized to develop and validate clinical prediction models for DSI, LNM, and LVSI. Moreover, these clinically significant independent risk factors were incorporated into the corresponding radiomics models to generate combined models. Ultimately, a radiomics nomogram integrating both the Rad-score and clinical independent risk factors was constructed. The diagnostic performance of these models was assessed using receiver operating characteristic (ROC) curves.\u003c/p\u003e\n \u003cp\u003eThe univariate and multivariate analyses were conducted to evaluate the associations between clinical characteristics and DSI, LNM, and LVSI. Significant clinical risk factors were used to develop and validate clinical prediction models for these outcomes. Additionally, these clinically significant independent risk factors were integrated into the corresponding radiomics models to construct combined models. Finally, a radiomics nomogram incorporating both the rad-score and clinical independent risk factors was developed, and its diagnostic efficiency was evaluated using ROC curves.\u003c/p\u003e\n \u003cp\u003eThree distinct models were developed to predict each biological behavior in both the training and testing datasets. These included: (1) the radiomics model, which incorporated only the rad-score; (2) the clinical model, which integrated multiple clinical risk factors; and (3) the combined model, which encompassed all selected parameters. Each model was independently tested and evaluated on the testing dataset to assess predictive performance. All procedures above were implemented using FeAture Explorer Pro (FAE, v0.5.2) within the Python environment (v3.7.6). The overall workflow of the study\u0026mdash;including patient selection, MRI image processing, radiomic feature extraction, model construction, and performance evaluation\u0026mdash;is illustrated in Fig.\u0026nbsp;2.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec7\"\u003e\n \u003ch2\u003e2.5 Model performance comparison\u003c/h2\u003e\n \u003cp\u003eCalibration curves and the Hosmer\u0026ndash;Lemeshow (HL) test were used to evaluate the agreement between predicted probabilities and observed outcomes, reflecting the calibration performance of the combined model. Decision curve analysis (DCA) was used to assess clinical utility by quantifying net benefits across different threshold probabilities. Shapley Additive Explanations (SHAP) analysis improved model interpretability by illustrating feature contributions both globally and locally.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec8\"\u003e\n \u003ch2\u003e2.6 Development of the Radiomics Nomogram\u003c/h2\u003e\n \u003cp\u003eClinical risk factors for each biological behavior were identified and incorporated into the radiomics nomogram through univariate and multivariate logistic regression analyses. The nomogram was constructed by integrating the rad-score with these selected clinical variables using multivariable logistic regression. This nomogram was subsequently utilized to predict the probability of DSI, LNM, and LVSI on an individual basis within the primary cohort.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec9\"\u003e\n \u003ch2\u003e2.7 Statistical analysis\u003c/h2\u003e\n \u003cp\u003eContinuous data were expressed as the mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (SD) or median with interquartile ranges, depending on the distribution of the data. Categorical data were presented as frequencies or percentages. The normality of the data distribution was assessed using the Kolmogorov-Smirnov test. Data that conformed to a normal distribution were reported as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD for continuous variables (e.g., age, tumor diameter, SCC level), while categorical variables (e.g., differentiation degree, FIGO stage, irregular vaginal bleeding, menopausal state) were summarized as frequencies or proportions. The reproducibility of radiomics features was systematically evaluated through the application of ICCs, which were utilized to quantify the degree of consistency among feature extractions. The data with bigger ICC values were used for statistical analysis. All statistical analyses were performed using R software (version 4.4.2, https://www.R-project.org/).A two-tailed \u003cem\u003ep\u003c/em\u003e value of less than 0.05 was considered to indicate statistical significance.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"3 Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003e3.1 Clinicopathological characteristics of patients\u003c/h2\u003e\n \u003cp\u003eTable \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e presents a comparative analysis of clinicopathological and radiological characteristics between patient subgroups stratified by biological behaviors, as well as between the training and testing cohorts. With the exception of SCC in the DSI and LVSI subsets, as well as FIGO stage in the LNM subset, no significant differences were observed in the clinical and histopathological characteristics between the training and validation cohorts (P\u0026thinsp;\u0026gt;\u0026thinsp;0.05).\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eClinicopathological characteristics of the dataset.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eParameters\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003etrain group(n\u0026thinsp;=\u0026thinsp;96)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003etest group(n\u0026thinsp;=\u0026thinsp;40)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDSI subsets\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDSI\u0026thinsp;=\u0026thinsp;43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enon-DSI\u0026thinsp;=\u0026thinsp;53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDSI\u0026thinsp;=\u0026thinsp;18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enon-DSI\u0026thinsp;=\u0026thinsp;22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e54.02\u0026thinsp;\u0026plusmn;\u0026thinsp;12.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e52.28\u0026thinsp;\u0026plusmn;\u0026thinsp;9.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e54.72\u0026thinsp;\u0026plusmn;\u0026thinsp;10.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e50.95\u0026thinsp;\u0026plusmn;\u0026thinsp;9.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.205\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTumor Diameter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.4\u0026thinsp;\u0026plusmn;\u0026thinsp;0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.3\u0026thinsp;\u0026plusmn;\u0026thinsp;0.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.34\u0026thinsp;\u0026plusmn;\u0026thinsp;11.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.96\u0026thinsp;\u0026plusmn;\u0026thinsp;12.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.871\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSCC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8.86\u0026thinsp;\u0026plusmn;\u0026thinsp;10.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.47\u0026thinsp;\u0026plusmn;\u0026thinsp;5.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10.77\u0026thinsp;\u0026plusmn;\u0026thinsp;14.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.82\u0026thinsp;\u0026plusmn;\u0026thinsp;12.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.011\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.55\u0026thinsp;\u0026plusmn;\u0026thinsp;2.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.85\u0026thinsp;\u0026plusmn;\u0026thinsp;0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.69\u0026thinsp;\u0026plusmn;\u0026thinsp;0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.65\u0026thinsp;\u0026plusmn;\u0026thinsp;0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.118\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e148.23\u0026thinsp;\u0026plusmn;\u0026thinsp;72.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e139.4\u0026thinsp;\u0026plusmn;\u0026thinsp;42.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e122.49\u0026thinsp;\u0026plusmn;\u0026thinsp;62.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e131.81\u0026thinsp;\u0026plusmn;\u0026thinsp;37.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.734\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.46\u0026thinsp;\u0026plusmn;\u0026thinsp;2.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.16\u0026thinsp;\u0026plusmn;\u0026thinsp;1.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.8\u0026thinsp;\u0026plusmn;\u0026thinsp;1.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.14\u0026thinsp;\u0026plusmn;\u0026thinsp;0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.228\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e123.02\u0026thinsp;\u0026plusmn;\u0026thinsp;15.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e123.59\u0026thinsp;\u0026plusmn;\u0026thinsp;13.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e124.39\u0026thinsp;\u0026plusmn;\u0026thinsp;13.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e123.77\u0026thinsp;\u0026plusmn;\u0026thinsp;10.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.928\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDifferentiation Degree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePoorly differentiated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.175\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eModerately differentiated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWell-differentiated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFIGO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.091\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eII-III\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIrregular Vaginal Bleeding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.886\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003epositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMenopausal State\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePremenopausal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.519\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePostmenopausal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLNM subsets\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLNM\u0026thinsp;=\u0026thinsp;16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enon-LNM\u0026thinsp;=\u0026thinsp;80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLNM\u0026thinsp;=\u0026thinsp;8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enon-LNM\u0026thinsp;=\u0026thinsp;32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e49.75\u0026thinsp;\u0026plusmn;\u0026thinsp;12.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e53.73\u0026thinsp;\u0026plusmn;\u0026thinsp;10.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e52.12\u0026thinsp;\u0026plusmn;\u0026thinsp;10.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e52.78\u0026thinsp;\u0026plusmn;\u0026thinsp;10.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.272\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTumor Diameter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.49\u0026thinsp;\u0026plusmn;\u0026thinsp;0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.31\u0026thinsp;\u0026plusmn;\u0026thinsp;0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8.59\u0026thinsp;\u0026plusmn;\u0026thinsp;16.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.27\u0026thinsp;\u0026plusmn;\u0026thinsp;10.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.429\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSCC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9.56\u0026thinsp;\u0026plusmn;\u0026thinsp;9.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.81\u0026thinsp;\u0026plusmn;\u0026thinsp;7.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14.92\u0026thinsp;\u0026plusmn;\u0026thinsp;18.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.33\u0026thinsp;\u0026plusmn;\u0026thinsp;12.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.062\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.11\u0026thinsp;\u0026plusmn;\u0026thinsp;0.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.18\u0026thinsp;\u0026plusmn;\u0026thinsp;2.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.73\u0026thinsp;\u0026plusmn;\u0026thinsp;0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.65\u0026thinsp;\u0026plusmn;\u0026thinsp;0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.853\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e143.8\u0026thinsp;\u0026plusmn;\u0026thinsp;46.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e143.27\u0026thinsp;\u0026plusmn;\u0026thinsp;60.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e123.78\u0026thinsp;\u0026plusmn;\u0026thinsp;62.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e128.58\u0026thinsp;\u0026plusmn;\u0026thinsp;46.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.871\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.27\u0026thinsp;\u0026plusmn;\u0026thinsp;2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.3\u0026thinsp;\u0026plusmn;\u0026thinsp;2.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.87\u0026thinsp;\u0026plusmn;\u0026thinsp;2.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.33\u0026thinsp;\u0026plusmn;\u0026thinsp;0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.810\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e121.81\u0026thinsp;\u0026plusmn;\u0026thinsp;12.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e123.64\u0026thinsp;\u0026plusmn;\u0026thinsp;14.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e117.25\u0026thinsp;\u0026plusmn;\u0026thinsp;11.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e125.75\u0026thinsp;\u0026plusmn;\u0026thinsp;11.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.159\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDifferentiation Degree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePoorly differentiated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.719\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eModerately differentiated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWell-differentiated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFIGO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.019\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eII-III\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIrregular Vaginal Bleeding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.386\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003epositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMenopausal State\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePremenopausal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.739\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePostmenopausal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLVSI subsets\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLVSI\u0026thinsp;=\u0026thinsp;43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enon-LVSI\u0026thinsp;=\u0026thinsp;53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLVSI\u0026thinsp;=\u0026thinsp;18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enon-LVSI\u0026thinsp;=\u0026thinsp;22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e53.84\u0026thinsp;\u0026plusmn;\u0026thinsp;12.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e52.43\u0026thinsp;\u0026plusmn;\u0026thinsp;9.27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e54.72\u0026thinsp;\u0026plusmn;\u0026thinsp;10.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e50.95\u0026thinsp;\u0026plusmn;\u0026thinsp;9.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.255\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTumor Diameter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.38\u0026thinsp;\u0026plusmn;\u0026thinsp;0.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.31\u0026thinsp;\u0026plusmn;\u0026thinsp;0.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.34\u0026thinsp;\u0026plusmn;\u0026thinsp;11.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.96\u0026thinsp;\u0026plusmn;\u0026thinsp;12.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.885\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSCC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8.86\u0026thinsp;\u0026plusmn;\u0026thinsp;10.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.47\u0026thinsp;\u0026plusmn;\u0026thinsp;5.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10.77\u0026thinsp;\u0026plusmn;\u0026thinsp;14.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.82\u0026thinsp;\u0026plusmn;\u0026thinsp;12.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.011\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.56\u0026thinsp;\u0026plusmn;\u0026thinsp;2.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.85\u0026thinsp;\u0026plusmn;\u0026thinsp;0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.69\u0026thinsp;\u0026plusmn;\u0026thinsp;0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.65\u0026thinsp;\u0026plusmn;\u0026thinsp;0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.115\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e146.15\u0026thinsp;\u0026plusmn;\u0026thinsp;71.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e141.09\u0026thinsp;\u0026plusmn;\u0026thinsp;44.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e122.49\u0026thinsp;\u0026plusmn;\u0026thinsp;62.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e131.81\u0026thinsp;\u0026plusmn;\u0026thinsp;37.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.937\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.53\u0026thinsp;\u0026plusmn;\u0026thinsp;2.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6.11\u0026thinsp;\u0026plusmn;\u0026thinsp;1.39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.8\u0026thinsp;\u0026plusmn;\u0026thinsp;1.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.14\u0026thinsp;\u0026plusmn;\u0026thinsp;0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.148\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e122.95\u0026thinsp;\u0026plusmn;\u0026thinsp;15.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e123.65\u0026thinsp;\u0026plusmn;\u0026thinsp;13.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e124.39\u0026thinsp;\u0026plusmn;\u0026thinsp;13.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e123.77\u0026thinsp;\u0026plusmn;\u0026thinsp;10.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.898\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDifferentiation Degree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePoorly differentiated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.175\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eModerately differentiated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWell-differentiated\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFIGO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.091\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eII-III\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIrregular Vaginal Bleeding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enegative\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.886\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003epositive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMenopausal State\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"4\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePremenopausal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.766\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePostmenopausal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003e3.2 Selection of radiomics features and performance of radiomics models\u003c/h2\u003e\n \u003cp\u003eFrom the 1476 radiomics features extracted from the R2* maps of ESWAN for each patient, we selected 866 features that demonstrated good inter-rater agreement (ICC\u0026thinsp;\u0026gt;\u0026thinsp;0.75) and applied them for further downscaling. Initially, the data underwent normalization, thereby dimensionality reduction via the Pearson correlation coefficient. Subsequently, distinct label groupings and imaging histology data processing were conducted in accordance with varying biological behaviors. Feature selection was executed using three methodologies: RFE, ANOVA, and the Kruskal-Wallis (KW) test. The final model classification is decided based on different classifiers and finally model construction is performed.\u003c/p\u003e\n \u003cp\u003eAs shown in Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e, three radiomics models based on different classifiers\u0026mdash;AE, SVM, and LRLasso\u0026mdash;were developed for predicting DSI, LNM, and LVSI. The optimal model for each subset was selected based on the highest AUC value. For the DSI subset, the AE-based model incorporating 12 radiomic features achieved the best performance, with an AUC of 0.824 and an accuracy of 0.781 in the training set, and an AUC of 0.712 with an accuracy of 0.725 in the test set (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eA). For the LNM subset, the LRLasso-based model utilizing 27 features demonstrated the highest efficacy, achieving an AUC of 0.827 and an accuracy of 0.842 in the training set, and an AUC of 0.740 with an accuracy of 0.740 in the test set (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eB). For the LVSI subset, the SVM-based model with 19 features showed the best performance, with an AUC of 0.897 and an accuracy of 0.865 in the training set, and an AUC of 0.750 with an accuracy of 0.750 in the test set (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003eC).\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eDiscriminative performance of different radiomics classifiers for predicting the three biological characteristics.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eDifferent models\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"4\"\u003e\n \u003cp\u003eTraining group\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"4\"\u003e\n \u003cp\u003eTesting group\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSEN\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSPE\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eACC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSEN\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSPE\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eACC\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"9\"\u003e\n \u003cp\u003eDSI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.824\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.814\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.755\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.781\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.712\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.636\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.725\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.757\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.512\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.906\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.729\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.605\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.722\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.591\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.650\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLRLasso\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.756\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.512\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.906\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.729\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.605\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.722\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.591\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.600\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"9\"\u003e\n \u003cp\u003eLNM\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.821\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.882\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.731\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.758\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.630\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.857\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.412\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.488\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.813\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.647\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.897\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.853\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.727\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.571\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.882\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.829\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLRLasso\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.827\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.647\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.885\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.842\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.740\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.572\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.912\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.740\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"9\"\u003e\n \u003cp\u003eLVSI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.811\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.814\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.755\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.781\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.722\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.778\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.682\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.725\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.897\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.861\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.868\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.865\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.611\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.864\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLRLasso\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.862\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.767\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.849\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.813\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.737\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.889\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.675\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003e3.4 Construction and Evaluation of Clinical, Radiomics Feature\u003c/h2\u003e\n \u003cp\u003eAfter conducting univariate and multivariate logistic analyses, significant predictors were identified for each clinical endpoint: SCC for DSI status, FIGO stage for LNM status, and SCC for LVSI status (Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e). These clinical characteristics were subsequently integrated into the optimal radiomics model to develop the combined models.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eUnivariate and multivariate analyses of risk factors by binary logistic regression models.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\" style=\"width: 31.7269%;\"\u003e\n \u003cp\u003eUnivariate analysis\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\" style=\"width: 32.5301%;\"\u003e\n \u003cp\u003eMultivariate analysis\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eVariable\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eOR (95%CI)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003eOR (95%CI)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003ep\u003c/em\u003e-value\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDSI subset\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.02(0.99\u0026ndash;1.06)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTumor diameter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00(0.95\u0026ndash;1.07)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSCC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.05(1.01\u0026ndash;1.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e1.05(1.01\u0026ndash;1.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"left\" class=\"colspec\"\u003e\u003cbr\u003e0.02\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDifferentiation degree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.96(0.95\u0026ndash;4.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFIGO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.13(0.97\u0026ndash;4.76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\" style=\"width: 32.5302%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIrregular vaginal bleeding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.12(0.56\u0026ndash;2.22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMenopausal state\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.33(0.67\u0026ndash;2.64)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.25(0.98\u0026ndash;1.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00(0.99\u0026ndash;1.01)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.13(0.94\u0026ndash;1.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00(0.97\u0026ndash;1.02)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLNM subset\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.97(0.93\u0026ndash;1.02)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003etumor_diameter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.02(0.97\u0026ndash;1.09)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSCC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.04(1-1.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDifferentiation_Degree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.41(0.55\u0026ndash;3.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFIGO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.27(1.29\u0026ndash;8.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e3.27(1.29\u0026ndash;8.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.01\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eirregular_vaginal_bleeding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.61(0.25\u0026ndash;1.48)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emenopausal_state\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.77(0.32\u0026ndash;1.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.98(0.68\u0026ndash;1.24)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.99(0.99\u0026ndash;1.01)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.03(0.79\u0026ndash;1.28)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.97(0.95\u0026ndash;1.01)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLVSI subset\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.02(0.99\u0026ndash;1.05)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003etumor_diameter\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00(0.95\u0026ndash;1.06)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSCC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.05(1.01\u0026ndash;1.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e1.05(1.01\u0026ndash;1.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.02\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDifferentiation_Degree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.96(0.95\u0026ndash;4.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFIGO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.13(0.97\u0026ndash;4.76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eirregular_vaginal_bleeding\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.12(0.56\u0026ndash;2.22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emenopausal_state\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.18(0.6\u0026ndash;2.34)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.26(0.99\u0026ndash;1.81)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00(0.99\u0026ndash;1.01)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.17(0.96\u0026ndash;1.45)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHgb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.00(0.97\u0026ndash;1.02)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 10.6425%;\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 21.8876%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eWhen DSI was utilized as the basis for grouping, the training cohort demonstrated an AUC value of 0.724 (95% CI: 0.609\u0026ndash;0.815), with an accuracy rate of 0.6456, sensitivity of 0.791, and specificity of 0.528. In the validation cohort, the AUC was 0.677 (95% CI: 0.505\u0026ndash;0.849), accompanied by an accuracy of 0.675, sensitivity of 0.833, and specificity of 0.546.When LNM was employed as the grouping criterion, the training cohort exhibited an AUC of 0.783 (95% CI: 0.639\u0026ndash;0.927), accuracy of 0.771, sensitivity of 0.750, and specificity of 0.775. For the test cohort, the AUC was 0.633 (95% CI: 0.401\u0026ndash;0.865), with an accuracy of 0.750, sensitivity of 0.500, and specificity of 0.813.In the model based on LVSI for grouping, the training cohort had an AUC of 0.712 (95% CI: 0.609\u0026ndash;0.816), accuracy of 0.688, sensitivity of 0.744, and specificity of 0.642. The test cohort showed an AUC of 0.664 (95% CI: 0.493\u0026ndash;0.835), accuracy of 0.650, sensitivity of 0.944, and specificity of 0.409.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003e3.5 Model Performance Comparison and Nomogram Development\u003c/h2\u003e\n \u003cp\u003eThe calibration curve of the combined model for the four subsets demonstrated that the predicted values were in good agreement with the observed values (Fig.\u0026nbsp;4). The Hosmer\u0026ndash;Lemeshow test showed good calibration for all models in the training cohort (DSI: P\u0026thinsp;=\u0026thinsp;0.222; LNM: P\u0026thinsp;=\u0026thinsp;0.851; LVSI: P\u0026thinsp;=\u0026thinsp;0.277), with all P-values exceeding the 0.05 threshold, indicating strong agreement between predicted and observed outcomes. In the training cohort, the AUCs of the combined model were significantly higher than the clinical model for predicting DSI, LNM, and LVSI (DSI, AUC\u003csub\u003ecombined\u003c/sub\u003e = 0.870 vs. AUC\u003csub\u003eclinical\u003c/sub\u003e = 0.724; LNM, AUC\u003csub\u003ecombined\u003c/sub\u003e = 0.837 vs. AUC\u003csub\u003eclinical\u003c/sub\u003e = 0.783; LVSI, AUC\u003csub\u003ecombined\u003c/sub\u003e = 0.876 vs. AUC\u003csub\u003eclinical\u003c/sub\u003e = 0.712). In the testing cohort, significant differences in the AUCs of the two models were observed for the prediction of DSI and LVSI (DSI, AUC\u003csub\u003ecombined\u003c/sub\u003e = 0.619 vs. AUC\u003csub\u003eclinical\u003c/sub\u003e = 0.677; LVSI, AUC\u003csub\u003ecombined\u003c/sub\u003e = 0.775 vs. AUC\u003csub\u003eclinical\u003c/sub\u003e = 0.664) (Table \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e). The DCA for the combined models is presented in Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e. Across a range of threshold probabilities (20\u0026ndash;40%), the combined model\u0026rsquo;s superior performance in both training and testing sets indicates enhanced clinical utility for risk prediction and decision-making. A nomogram was developed by integrating clinical, radiological, and histological features through logistic regression to visualize the combined model and enable individualized prediction of each biological characteristic. (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e).\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab4\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003ePerformances of models in predicting response to three biological behaviors.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003eDifferent models\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"4\"\u003e\n \u003cp\u003eTraining group\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"4\"\u003e\n \u003cp\u003eTesting group\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUC(95%CI)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSEN\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSPE\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eACC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAUC(95%CI)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSEN\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSPE\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eACC\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"9\"\u003e\n \u003cp\u003eDSI subset\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRadiomics model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.824[0.738\u0026ndash;0.910]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.814\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.755\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.781\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.712[0.546\u0026ndash;0.878]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.636\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.725\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eClinical\u003c/p\u003e\n \u003cp\u003emodel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.724[0.609\u0026ndash;0.815]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.791\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.528\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.646\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.677[0.505\u0026ndash;0.849]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.546\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.675\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCombined\u003c/p\u003e\n \u003cp\u003emodel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.870[0.797\u0026ndash;0.943]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.861\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.755\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.802\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.619[0.437\u0026ndash;0.801]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.778\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.546\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.650\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"9\"\u003e\n \u003cp\u003eLNM subset\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRadiomics model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.827[0.717\u0026ndash;0.938]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.647\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.885\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.842\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.739[0.485\u0026ndash;0.994]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.572\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.912\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.740\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eClinical\u003c/p\u003e\n \u003cp\u003emodel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.783[0.639\u0026ndash;0.927]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.775\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.771\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.633[0.401\u0026ndash;0.865]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.813\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCombined\u003c/p\u003e\n \u003cp\u003emodel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.837[0.732\u0026ndash;0.941]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.706\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.846\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.853\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.735[0.4820\u0026ndash;0.9801]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.571\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.941\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.878\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"9\"\u003e\n \u003cp\u003eLVSI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRadiomics model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.897[0.828\u0026ndash;0.966]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.861\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.868\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.865\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.750[0.591\u0026ndash;0.909]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.611\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.864\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eClinical\u003c/p\u003e\n \u003cp\u003emodel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.712[0.609\u0026ndash;0.816]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.744\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.642\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.688\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.664[0.493\u0026ndash;0.835]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.944\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.409\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.650\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCombined\u003c/p\u003e\n \u003cp\u003emodel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.876[0.803\u0026ndash;0.949]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.907\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.736\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.813\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.775[0.619\u0026ndash;0.932]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.611\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.955\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.800\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n \u003ch2\u003e3.5 SHAP Analysis for the Combined Logistic Regression Model\u003c/h2\u003e\n \u003cp\u003eThe SHAP method is used to interpret model outputs by evaluating the contribution of each feature to the prediction. This approach offers both global explanations, which examine feature-level impacts, and local explanations, which focus on individual predictions. SHAP dependence plots (Figs.\u0026nbsp;7,\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e,\u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003eA-B) reveal the influence of the top two features, with the x-axis showing feature values and the y-axis displaying SHAP values that indicate each feature\u0026apos;s contribution. The most influential features for predicting the DSI subset are log_glcm_log.sigma.3D.2.0.mm.3D.DifferenceEntropy and wavelet_glszm_wavelet.LHH.GrayLevelNonUniformity. For the LNM subset, the key features with the highest contribution are log_gldm_log.sigma.4.0.mm.3D. DependenceNonUniformity and wavelet_glrlm_wavelet.HHH.RunPercentage. In the LVSI subset, the most significant features are wavelet_glcm_wavelet.LLH.Autocorrelation and wavelet_gldm_wavelet.LLH.LargeDependenceHighGrayLevelEmphasis. These features, identified through SHAP analysis, highlight the critical radiomic markers associated with each subset, providing valuable insights for the prediction of DSI, LNM, and LVSI. A bar plot (Figs.\u0026nbsp;7,\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e,\u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003eC) ranks features by their mean absolute SHAP values, highlighting those most influential on the model\u0026apos;s predictions. SHAP waterfall plots (Figs.\u0026nbsp;7,\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e,\u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003eD-E) demonstrate how individual features contribute to the predicted probability of DSI for both non-DSI/LNM/LVSI (label\u0026thinsp;=\u0026thinsp;0) and DSI/LNM/LVSI (label\u0026thinsp;=\u0026thinsp;1) cases, showing the cumulative effect of features on predictions. The SHAP beeswarm plot (Figs.\u0026nbsp;7,\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e,\u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003eF) visualizes SHAP value distributions across samples, with dot color indicating feature values and spread showing the impact\u0026apos;s magnitude and direction. Lastly, the feature interaction strength plot (Figs.\u0026nbsp;7,\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e,\u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003eG) evaluates how feature interactions contribute to DSI/LNM/LVSI prediction, with higher values indicating more significant interactions. In three groups, the final prediction probability f(x) integrates the baseline value and all feature contributions, combining global and local insights from the SHAP method.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eIn this study, we utilized 11 advanced machine learning classifiers to assess the preoperative prediction of three distinct biological behaviors\u0026mdash;DSI, LNM, and LVSI\u0026mdash;in patients with squamous cervical cancer. The radiomics features were extracted from R2* maps generated by ESWAN sequences in MRI scans. Our findings revealed that the three biological behaviors could be effectively predicted using radiomics features derived from AE, LRLasso, and SVM respectively. Moreover, the integration of radiomics features with clinical features into a combined model significantly improved the predictive performance for DSI, LNM, and LVSI. Additionally, nomograms were established as practical and precise tools for outcome prediction and high-risk patient stratification, thereby supporting the development of personalized therapeutic strategies.\u003c/p\u003e\u003cp\u003eBeyond radiomics features, our univariate/multivariate analyses identified clinical biomarkers critical for risk stratification. The outcomes of univariate and multivariate logistic regression analyses demonstrate that SCC levels and the FIGO stage are significant predictors of DSI, LNM, and LVSI. SCC levels serve as an independent prognostic factor for DSI and LVSI. Moreover, the FIGO stage emerges as a critical predictor of LNM, while the FIGO stage is a critical predictor of LNM.\u003c/p\u003e\u003cp\u003eThese findings are consistent with the study conducted by Zhu et al.\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e, which established that elevated preoperative SCC concentrations are significantly correlated with FIGO stage, tumor size exceeding 4 cm, stromal infiltration, LNM, and LVSI. An SCC threshold of 2.75 ng/mL or higher and the presence of LVSI were identified as independent determinants influencing LNM. Additionally, the study indicates that SCC possesses a moderate predictive capacity for LNM, with an AUC of 0.703. These results highlight the potential utility of SCC as a biomarker for predicting lymphatic spread and vascular invasion in early-stage cervical carcinoma. While Zhu et al. confirmed SCC\u0026rsquo;s prognostic value, our study further demonstrates that integrating SCC with R2* radiomics (AUC\u0026thinsp;=\u0026thinsp;0.845) surpasses standalone biomarker models (AUC\u0026thinsp;=\u0026thinsp;0.703 ) \u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eRadiomics analysis derived from MRI has demonstrated significant potential in enhancing preoperative diagnostic accuracy for DSI, LNM, and LVSI in patients with early-stage cervical cancer. For instance, Ren et al. demonstrated that MRI-based radiomics analysis significantly improves the diagnostic accuracy of DSI compared with conventional radiologist interpretation\u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e. Similarly, research by Wu et al. and Shi et al. indicates that multiparametric MRI radiomics can reliably assess LNM and LVSI, achieving high diagnostic performance metrics, including elevated AUC values\u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e. The superior contrast resolution of ESWAN likely underpins the high specificity of our R2*-based nomogram (0.955 for LVSI), despite its lower sensitivity compared to multiparametric MRI \u003csup\u003e[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe ESWAN sequence, characterized by its increased sensitivity to microhemorrhages and hemosiderin deposits, provides superior signal-to-noise ratio and tissue contrast in uterine MRI imaging relative to conventional SWI protocols \u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e. This enhancement facilitates the detection of subtle lesions and enables precise evaluation of uterine pathologies such as endometrial carcinoma and adenomyosis \u003csup\u003e[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e. Although it entails a marginally extended acquisition duration, the sequence's improved image resolution and diagnostic yield substantiate its utility in clinical radiological assessments \u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eCT radiomics offers valuable insights, although it may exhibit reduced sensitivity to specific tumor phenotypes compared to MRI. Ren et al. employed treatment planning CT radiomics to forecast therapeutic efficacy and hematological toxicities in patients with locally advanced cervical carcinoma \u003csup\u003e[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e.Lucia et al. utilized pretreatment MRI radiomic features to predict clinical outcomes in locally advanced cervical cancer undergoing chemoradiotherapy, demonstrating that radiomic biomarkers can enhance prognostic accuracy beyond conventional clinical staging parameters \u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/sup\u003e. This evidence indicates that MRI-based radiomics holds promise for predicting treatment response and overall survival in cervical malignancies. Furthermore, Zhang et al. reported that MRI radiomic signatures could stratify survival probabilities in patients with locally advanced cervical squamous cell carcinoma receiving concurrent chemoradiotherapy \u003csup\u003e[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/sup\u003e. This underscores the capacity of MRI radiomic analysis to quantify tumor phenotypic features that are predictive of clinical prognosis.\u003c/p\u003e\u003cp\u003eThe findings mentioned above underscore the progression of radiomic analysis in cervical carcinoma, employing CT and MRI to advance from fundamental efficacy prediction to prognostic modeling of therapeutic response and survival outcomes. Our study has further explored several biological behaviors of cervical squamous cell carcinoma, which can potentially inform treatment planning and patient prognosis. Our results demonstrate that for DSI and LNM in cervical squamous cell carcinoma, the combined model incorporating R2* exhibits robust predictive performance. Conversely, for LVSI, the diagnostic performance remains comparable, with the AUC not surpassing that of the standalone radiomic model.\u003c/p\u003e\u003cp\u003ePrevious studies have developed various prediction models utilizing MRI data to assess deep stromal invasion in patients with cervical cancer. Ren et al. investigated the diagnostic efficacy of a radiomics model based on T2WI for preoperative assessment of DSI in cervical cancer, demonstrating superior performance compared to radiologists, particularly in terms of sensitivity and specificity. In their validation cohort, the radiomics model achieved an area under the AUC of 0.879 \u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e. Similarly, Yan et al. employed a combined approach utilizing T2WI and contrast-enhanced T1-weighted imaging (CE-T1WI) within a machine learning approach for preoperative DSI prediction. Their radiomics signature achieved an AUC of 0.951 in the training dataset and 0.882 in the testing dataset \u003csup\u003e[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/sup\u003e. These findings highlight the potential of multimodal MRI radiomics for accurate preoperative staging of deep stromal invasion in cervical cancer. Although standalone R2* radiomics underperformed multimodal MRI for DSI prediction (AUC\u0026thinsp;=\u0026thinsp;0.879), our combined model achieved comparable accuracy (AUC\u0026thinsp;=\u0026thinsp;0.870), suggesting clinical factors compensate for single-sequence limitations \u003csup\u003e[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eHowever, our findings indicate that these models outperform the ESWAN-based R2* parameter radiomics model in predicting DSI. Considering that our research focused on R2*, a less frequently employed sequence in this domain, we hypothesize that a multimodal imaging approach combining various sequences could potentially improve diagnostic accuracy.\u003c/p\u003e\u003cp\u003eThe study by Deng et al. demonstrated the potential of non-invasive differentiation between LNM and vascular endothelial growth factor (VEGF) expression in cervical cancer \u003csup\u003e[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/sup\u003e. Their finding highlighted non-invasive LNM prediction via radiomics, while our results refine this by identifying SCC and FIGO stage as key predictors in specific subsets (e.g., SCC for LVSI, FIGO for LNM), enabling targeted risk stratification. Specifically, in the subsets with DSI and LVSI, SCC was identified as an independent prognostic factor through multivariate logistic regression analysis (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05). In the subset with LNM, FIGO staging emerged as an independent prognostic factor (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05). These findings highlight the differential predictive roles of SCC and FIGO staging in distinct pathological subsets of CSCC. Given that the sample sizes in these studies exceed the small-sample threshold (n\u0026thinsp;\u0026gt;\u0026thinsp;30), the observed discrepancies may be attributed to the specific patient subsets analyzed, rather than random variation. Collectively, these studies highlight the necessity of incorporating patient-specific factors and subgroup stratification when assessing prognostic biomarkers in CSCC.\u003c/p\u003e\u003cp\u003eMeta-analyses have reported an AUC of 0.83 [95%CI: 0.76\u0026ndash;0.89] for radiomics models in detecting LNM. The diagnostic performance can be slightly enhanced by combining multiple imaging sequences, imaging histology, and clinical factors. In our single-modality study, the radiomics model achieved an AUC of 0.807 in the training cohort, which aligns with the aforementioned findings. Li et al. also highlighted that studies incorporating clinical factors tend to have higher diagnostic odds ratios (DOR) compared to those using only radiomics features. Consistent with this observation, the combined model in our study demonstrated significantly better performance than the radiomics model alone \u003csup\u003e[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe study by Shi et al. demonstrated the potential of intratumoral and peritumoral LNM-specific analyses, as well as the development of radiomics models incorporating multi-regional features, MR-reported lymph node status, and tumor diameter. This approach achieved a high AUC in the training cohort \u003csup\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e. Consistent with these findings, the combined model in the present study (AUC\u0026thinsp;=\u0026thinsp;0.845) outperformed the single-radiomics model (AUC\u0026thinsp;=\u0026thinsp;0.807), highlighting the benefits of integrating multiple data sources to enhance predictive efficacy.\u003c/p\u003e\u003cp\u003eFor LVSI prediction, existing T2WI-based nomograms show limited AUC values (C-index\u0026thinsp;=\u0026thinsp;0.78) \u003csup\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/sup\u003e, while our R2*-integrated model achieved superior performance (AUC\u0026thinsp;=\u0026thinsp;0.897, specificity\u0026thinsp;=\u0026thinsp;0.955). Although the sensitivity was 0.611, lower than that of Xiao et al.\u0026rsquo;s multiparametric approach (sensitivity\u0026thinsp;=\u0026thinsp;0.90) \u003csup\u003e[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e, the significantly higher specificity underscores R2*\u0026rsquo;s utility in reliably identifying high-risk cases with minimal false positives. This aligns with prior evidence that advanced sequences like ESWAN enhance microscopic tumor feature detection \u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe combined nomogram demonstrated consistent efficacy across biological behaviors: for DSI, AUC\u0026thinsp;=\u0026thinsp;0.870 (sensitivity\u0026thinsp;=\u0026thinsp;0.861, specificity\u0026thinsp;=\u0026thinsp;0.755); for LVSI, AUC\u0026thinsp;=\u0026thinsp;0.876 (sensitivity\u0026thinsp;=\u0026thinsp;0.907, specificity\u0026thinsp;=\u0026thinsp;0.736). Unlike invasive biopsies requiring histologic assessment, this non-invasive tool provides comprehensive risk stratification, supporting personalized therapeutic planning \u003csup\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eIn contrast to conventional histological techniques, which necessitate invasive biopsies and may be compromised by sampling variability, our radiomics-based nomogram provides a non-invasive alternative with distinct advantages. By leveraging advanced imaging features, it circumvents the procedural risks associated with invasive biopsies, such as infection and hemorrhage. Moreover, this method offers a more holistic and objective evaluation of tumor characteristics, thereby enabling clinicians to formulate well-informed decisions without exposing patients to unnecessary invasive interventions \u003csup\u003e[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eAccurate preoperative prediction of biological behavior is essential for optimizing personalized therapeutic protocols in clinical practice. For example, patients classified as high risk for LVSI or DSI via our nomogram may derive significant benefit from intensified chemotherapy or radiation therapy regimens. Conversely, those identified as low risk could potentially be spared from unnecessary aggressive interventions. This individualized stratification not only improves clinical outcomes but also reduces the morbidity associated with overtreatment \u003csup\u003e[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe calibration curves demonstrated good agreement between predicted and observed outcomes in both training and validation cohorts, confirming the reliability of radiomic and combined models in identifying distinct biological behaviors, including DSI, LNM, and LVSI. These findings are consistent with previous studies that reported high calibration performance in radiomics-based predictive models for LVSI in cervical cancer patients \u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e. DCA further validated the clinical utility of the combined model, particularly in predicting DSI and LVSI, where it provided greater net benefit across a range of threshold probabilities, in line with prior evidence suggesting superior clinical benefit of integrated models over radiomics or clinical features alone \u003csup\u003e[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e. For LNM, the model demonstrated substantial net benefit primarily among low-risk cervical cancer patients, suggesting its value in supporting early intervention strategies. A similar trend of limited benefit in high-risk populations has also been reported in bladder cancer radiomics research, where model performance was constrained by the underrepresentation of high-risk cases \u003csup\u003e[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/sup\u003e. Nonetheless, the diminished net benefit across broader or high-risk populations may stem from the limited sample size of high-risk cases within the current dataset. This underscores the imperative to increase sample sizes and enhance the representation of high-risk groups in future investigations to optimize model accuracy and external validity.\u003c/p\u003e\u003cp\u003eRadiomics features can capture the intrinsic heterogeneity of tumors that is often imperceptible to the naked eye. In the present study, the utilization of radiomics-based nomograms and unimodal imaging (R2* maps) represents a significant innovation. When combined with machine learning classifiers, these features can more accurately reflect the overall characteristics of the tumor, thereby enhancing the model's discriminative ability. Radiomics has been extensively applied in the differential diagnosis and survival prognosis of cervical cancer. In our current investigation, the combined model incorporating logistic regression (LR)-based clinical and radiological features demonstrated superior predictive validity compared to clinical and radiological models alone in terms of DSI and LNM. Although the combined model's validity for LVSI was slightly lower than ideal, its efficacy was significantly higher than that of similar studies utilizing conventional MRI sequences, thus underscoring the significance of our study. Moreover, the combined model exhibits good clinical applicability, and the visual nomogram based on this model may serve as a valuable tool for clinical use.\u003c/p\u003e\u003cp\u003eWhile the current investigation yielded encouraging findings, several methodological limitations merit consideration. Firstly, to focus on elucidating the diagnostic potential of novel imaging sequence, this study exclusively analyzed R2* parametric maps within the ESWAN sequence for single-parameter assessment. This targeted approach does not diminish the validity and reproducibility of multi-parametric imaging studies, which are recommended for future research. Secondly, the evaluation of the three biological markers\u0026mdash;DSI, LNM, and LVSI\u0026mdash;was conducted independently. This approach may have overlooked the potential prognostic significance of their interrelationships, warranting further investigation. Lastly, the relatively limited sample size and the retrospective study design may inherently constrain the generalizability of the findings.\u003c/p\u003e"},{"header":"5 Conclusion","content":"\u003cp\u003eIn summary, the combined model demonstrated robust predictive performance with respect to the biological behavior of CSCC. Moreover, visual nomograms derived from the combined model serve as valuable tools for risk stratification, individualized treatment planning, and enhanced surveillance of patients with CSCC.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eCSCC：Cervical Squamous Cell Carcinoma\u003c/p\u003e\n\u003cp\u003eNCCN：National Comprehensive Cancer Network\u003c/p\u003e\n\u003cp\u003eR2* maps：the apparent transverse relaxation rate\u003c/p\u003e\n\u003cp\u003eDSI：Deep Stromal Invasion\u003c/p\u003e\n\u003cp\u003eLNM：Lymph Node Metastasis\u003c/p\u003e\n\u003cp\u003eLVSI：Lymph-Vascular Space Invasion\u003c/p\u003e\n\u003cp\u003eFIGO: International Federation of Gynecology and Obstetrics\u003c/p\u003e\n\u003cp\u003eESWAN：Enhanced Susceptibility-Weighted Angiography\u003c/p\u003e\n\u003cp\u003eSWI：Susceptibility Weighted Imaging\u003c/p\u003e\n\u003cp\u003eSCC：Squamous Cell Carcinoma Antigen\u003c/p\u003e\n\u003cp\u003eNLR：Neutrophil-to-Lymphocyte Ratio\u003c/p\u003e\n\u003cp\u003ePLR：Platelet-to-Lymphocyte Ratio\u003c/p\u003e\n\u003cp\u003eWBC：White Blood Cell Count\u003c/p\u003e\n\u003cp\u003eHgb：Hemoglobin\u003c/p\u003e\n\u003cp\u003eFS-T2WI：Fat-Suppressed T2-Weighted Imaging\u003c/p\u003e\n\u003cp\u003eCE-T1WI：contrast-enhanced T1-weighted imaging\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDWI：Diffusion-Weighted Imaging\u003c/p\u003e\n\u003cp\u003eROI：Region of Interest\u003c/p\u003e\n\u003cp\u003eGLCM：Gray-Level Co-occurrence Matrix\u003c/p\u003e\n\u003cp\u003eGLDM：Gray-Level Dependence Matrix\u003c/p\u003e\n\u003cp\u003eGLSZM：Gray-Level Size Zone Matrix\u003c/p\u003e\n\u003cp\u003eGLRLM：Gray-Level Run Length Matrix\u003c/p\u003e\n\u003cp\u003eNGTDM：Neighboring Gray-Tone Difference Matrix\u003c/p\u003e\n\u003cp\u003eICC：Intra-class Correlation Coefficient\u003c/p\u003e\n\u003cp\u003ePCC：Pearson Correlation Coefficients\u003c/p\u003e\n\u003cp\u003eANOVA：Analysis of Variance\u003c/p\u003e\n\u003cp\u003eRFE：Recursive Feature Elimination\u003c/p\u003e\n\u003cp\u003eLASSO：Least Absolute Shrinkage and Selection Operator\u003c/p\u003e\n\u003cp\u003eSVM：Support Vector Machine\u003c/p\u003e\n\u003cp\u003eLDA：Linear Discriminant Analysis\u003c/p\u003e\n\u003cp\u003eAE：Autoencoder\u003c/p\u003e\n\u003cp\u003eLR-Lasso：Logistic Regression via Lasso\u003c/p\u003e\n\u003cp\u003eROC：Receiver Operating Characteristic\u003c/p\u003e\n\u003cp\u003eAUC：Area Under the Curve\u003c/p\u003e\n\u003cp\u003eHL test：Hosmer\u0026ndash;Lemeshow Test\u003c/p\u003e\n\u003cp\u003eDCA：Decision Curve Analysis\u003c/p\u003e\n\u003cp\u003eSHAP：Shapley Additive Explanations\u003c/p\u003e\n\u003cp\u003eSD：Standard Deviation\u003c/p\u003e\n\u003cp\u003eKW test：Kruskal-Wallis Test\u003c/p\u003e\n\u003cp\u003eVEGF：vascular endothelial growth factor\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDOR: diagnostic odds ratios\u0026nbsp;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors thank all healthcare professionals who contributed to this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eS.W., Q.S., and C.M. were responsible for case collection. Image evaluation was performed by C.M., while image processing and ROI delineation were conducted by S.W. and Q.S. The manuscript was drafted by S.W., and critically revised by D.D., S.T., and A.L. for important intellectual content.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study is supported by the Hospital Fund of the First Affiliated Hospital of Dalian Medical University (2021HZ015) and the Dalian Municipal Medical Science Research Program (2023DF038).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData are available from the corresponding authors upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was conducted in accordance with the Declaration of Helsinki. Ethical approval was obtained from the Ethics Committee of the First Affiliated Hospital of Dalian Medical University (Approval No. YJ-KS-KY-2025-353). Written informed consent was obtained from all participants prior to inclusion in the study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDuncan J, Harris M, Skyers N, Bailey A, Figueroa JP. A Call for Low- and Middle-Income Countries to Commit to the Elimination of Cervical Cancer. Lancet Reg Health Am. 2021;2:100036. doi:10.1016/j.lana.2021.100036 \u003c/li\u003e\n\u003cli\u003eAbu-Rustum NR, Yashar CM, Arend R, et al. NCCN Guidelines\u0026reg; Insights: Cervical Cancer, Version 1.2024. J Natl Compr Canc Netw. 2023;21(12):1224-1233. doi:10.6004/jnccn.2023.0062. \u003c/li\u003e\n\u003cli\u003eCao L, Wen H, Feng Z, Han X, Zhu J, Wu X:Role of adjuvant therapy after radical hysterectomy in intermediate-risk, early-stage cervical cancer. International journal of gynecological cancer :official journal of the International Gynecological Cancer Society 2021, 31(1):52-58.\u003c/li\u003e\n\u003cli\u003eZhu J, Cao L, Wen H, Bi R, Wu X, Ke G:The clinical and prognostic implication of deep stromal invasion in cervical cancer patients undergoing radical hysterectomy. Journal of Cancer 2020, 11(24):7368-7377.\u003c/li\u003e\n\u003cli\u003eCao L, Wen H, Feng Z, Han X, Zhu J, Wu X. Role of adjuvant therapy after radical hysterectomy in intermediate-risk, early-stage cervical cancer. Int J Gynecol Cancer. 2021;31(1):52-58. doi:10.1136/ijgc-2020-001974.\u003c/li\u003e\n\u003cli\u003eCibula D, Rosaria Raspollini M, Planchamp F, Centeno C, Chargari C, Felix A, Fischerov\u0026aacute; D, Jahnn-Kuch D, Joly F, Kohler C et al:ESGO/ESTRO/ESP Guidelines for the management of patients with cervical cancer - Update 2023. Radiotherapy and oncology :journal of the European Society for Therapeutic Radiology and Oncology 2023, 184:109682.\u003c/li\u003e\n\u003cli\u003eXu Y, Zhang M, Zhang J, et al. Neoadjuvant chemotherapy increases the 5-year overall survival of patients with resectable cervical cancer: A systematic review and meta-analysis. Taiwan J Obstet Gynecol. 2021;60(3):433-441. doi:10.1016/j.tjog.2021.03.008.\u003c/li\u003e\n\u003cli\u003eKim J, Cho Y, Kim N, et al. Magnetic resonance imaging-based validation of the 2018 FIGO staging system in patients treated with definitive radiotherapy for locally advanced cervix cancer. Gynecol Oncol. 2021;160(3):735-741. doi:10.1016/j.ygyno.2020.12.012 .\u003c/li\u003e\n\u003cli\u003eXin JY, Gao SS, Liu JG, Sun CF, Han Y, Sun XH, Wang XZ, Wang B:The value of ESWAN in diagnosis and differential diagnosis of prostate cancer:Preliminary study. Magnetic resonance imaging 2017, 44:26-31.\u003c/li\u003e\n\u003cli\u003eHan X, Sun M, Wang M, Fan R, Chen D, Xie L, Liu A:The enhanced T(2) star weighted angiography (ESWAN) value for differentiating borderline from malignant epithelial ovarian tumors. European journal of radiology 2019, 118:187-193.\u003c/li\u003e\n\u003cli\u003eG\u0026uuml;nakan E, Atan S, Haberal AN, K\u0026uuml;\u0026ccedil;\u0026uuml;kyıldız İ A, G\u0026ouml;k\u0026ccedil;e E, Ayhan A:A novel prediction method for lymph node involvement in endometrial cancer:machine learning. International journal of gynecological cancer :official journal of the International Gynecological Cancer Society 2019, 29(2):320-324.\u003c/li\u003e\n\u003cli\u003eChu R, Zhang Y, Qiao X, Xie L, Chen W, Zhao Y, Xu Y, Yuan Z, Liu X, Yin A et al:Risk Stratification of Early-Stage Cervical Cancer with Intermediate-Risk Factors:Model Development and Validation Based on Machine Learning Algorithm. The oncologist 2021, 26(12):e2217-e2226.\u003c/li\u003e\n\u003cli\u003eSun ZD, Zhang Y, Yang YS, et al. Construction and validation of a novel nomogram for prediction of lymph node metastasis in HER2-positive breast cancer: based on the optimal number of examined lymph nodes for accurate nodal staging. BMC Womens Health. 2025;25(1):132. Published 2025 Mar 22. doi:10.1186/s12905-025-03663-w\u003c/li\u003e\n\u003cli\u003eGuo Q, Gao Y, Lin Y, et al. A nomogram of preoperative indicators predicting lymph vascular space invasion in cervical cancer. Arch Gynecol Obstet. 2024;309(5):2079-2087. doi:10.1007/s00404-024-07385-6\u003c/li\u003e\n\u003cli\u003eZhu C, Zhang W, Wang X, Jiao L, Chen L, Jiang J. Predictive value of preoperative serum squamous cell carcinoma antigen level for lymph node metastasis in early-stage cervical squamous cell carcinoma. Medicine (Baltimore). 2021;100(33):e26960. doi:10.1097/MD.0000000000026960\u003c/li\u003e\n\u003cli\u003eRen J, Li Y, Yang JJ, et al. MRI-based radiomics analysis improves preoperative diagnostic performance for the depth of stromal invasion in patients with early stage cervical cancer. Insights Imaging. 2022;13(1):17. Published 2022 Jan 29. doi:10.1186/s13244-022-01156-0\u003c/li\u003e\n\u003cli\u003eWu Q, Shi D, Dou S, et al. Radiomics Analysis of Multiparametric MRI evaluates the pathological features of cervical squamous cell carcinoma. J Magn Reson Imaging. 2019;49(4):1141\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003eShi J, Dong Y, Jiang W, et al. MRI-based peritumoral radiomics analysis for preoperative prediction of lymph node metastasis in early-stage cervical cancer: a multi-center study. Magn Reson Imaging. 2022;88:1\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003eGuo LF, Wang G, Zhu XY, Liu C, Cui L. Comparison of ESWAN, SWI-SPGR, and 2D T2*-weighted GRE sequence for depicting cerebral microbleeds. Clin Neuroradiol. 2013;23(2):121-127. doi:10.1007/s00062-012-0185-7 \u003c/li\u003e\n\u003cli\u003eVanaerde O, Budzik JF, Mackowiak A, et al. Comparison between enhanced susceptibility-weighted angiography and time of flight sequences in the detection of arterial occlusion in acute ischemic stroke. J Neuroradiol. 2017;44(3):210-216.doi:10.1016/j.neurad.2017.01.12510.1016/j.neurad.2017.01.125\u003c/li\u003e\n\u003cli\u003eGhosh A, Singh T, Bagga R, Srinivasan R, Singla V, Khandelwal N. T2 relaxometry mapping in demonstrating layered uterine architecture: parameter optimization and utility in endometrial carcinoma and adenomyosis: a feasibility study. Br J Radiol. 2018;91(1081):20170377. doi:10.1259/bjr.20170377 \u003c/li\u003e\n\u003cli\u003eRen K, Shen L, Qiu J, et al. Treatment planning computed tomography radiomics for predicting treatment outcomes and haematological toxicities in locally advanced cervical cancer treated with radiotherapy: A retrospective cohort study. BJOG. 2023;130(2):222-230. doi:10.1111/1471-0528.17285\u003c/li\u003e\n\u003cli\u003eLucia F, Bourbonne V, Pleyers C, et al. Multicentric development and evaluation of 18F-FDG PET/CT and MRI radiomics models to predict para-aortic lymph node involvement in locally advanced cervical cancer. Eur J Nucl Med Mol Imaging. 2023;50(8):2514-2528. doi:10.1007/s00259-023-06180-w \u003c/li\u003e\n\u003cli\u003eZhang X, Zhao J, Zhang Q, et al. MRI-based radiomics value for predicting the survival of patients with locally advanced cervical squamous cell cancer treated with concurrent chemoradiotherapy. Cancer Imaging. 2022;22(1):35. Published 2022 Jul 16. doi:10.1186/s40644-022-00474-2\u003c/li\u003e\n\u003cli\u003eYan H, Huang G, Yang Z, Chen Y, Xiang Z. Machine Learning-Based Multiparametric Magnetic Resonance Imaging Radiomics Model for Preoperative Predicting the Deep Stromal Invasion in Patients with Early Cervical Cancer. J Imaging Inform Med. 2024;37(1):230-246. doi:10.1007/s10278-023-00906-w \u003c/li\u003e\n\u003cli\u003eDeng X, Liu M, Sun J, et al. Feasibility of MRI-based radiomics features for predicting lymph node metastases and VEGF expression in cervical cancer. Eur J Radiol. 2021;134:109429.\u003c/li\u003e\n\u003cli\u003eLi L, Zhang J, Zhe X, et al. A meta-analysis of MRI-based radiomic features for predicting lymph node metastasis in patients with cervical cancer. Eur J Radiol. 2022;151:110243.\u003c/li\u003e\n\u003cli\u003eGuo L, Zhang R, Xu Y, et al. Predicting the status of lymphovascular space invasion using quantitative parameters from synthetic MRI in cervical squamous cell carcinoma without lymphatic metastasis. Front Oncol. 2024;14:1304793. Published 2024 Feb 6. doi:10.3389/fonc.2024.1304793\u003c/li\u003e\n\u003cli\u003eXiao M, Li Y, Ma F, Zhang G, Qiang J. Multiparametric MRI radiomics nomogram for predicting lymph-vascular space invasion in early-stage cervical cancer. Br J Radiol. 2022;95(1134):20211076. doi:10.1259/bjr.20211076 \u003c/li\u003e\n\u003cli\u003eWu L, Li S, Li S, Lin Y, Wei D. Preoperative magnetic resonance imaging-radiomics in cervical cancer: a systematic review and meta-analysis. Front Oncol. 2024;14:1416378. Published 2024 Jul 4. doi:10.3389/fonc.2024.1416378\u003c/li\u003e\n\u003cli\u003eTang S, Yen A, Wang K, Albuquerque K, Wang J. Progression-Free Survival Prediction for Locally Advanced Cervical Cancer After Chemoradiotherapy With MRI-based Radiomics. Clin Oncol (R Coll Radiol). 2025;38:103702. doi:10.1016/j.clon.2024.103702 \u003c/li\u003e\n\u003cli\u003eMa NN, Wang T, Lv YN, Li SD. An MRI radiomics-based model for the prediction of invasion of the lymphovascular space in patients with cervical cancer. Front Oncol. 2024;14:1394427. Published 2024 Jul 5. doi:10.3389/fonc.2024.1394427\u003c/li\u003e\n\u003cli\u003eHuang G, Cui Y, Wang P, et al. Multi-Parametric Magnetic Resonance Imaging-Based Radiomics Analysis of Cervical Cancer for Preoperative Prediction of Lymphovascular Space Invasion. Front Oncol. 2022;11:663370. Published 2022 Jan 12. doi:10.3389/fonc.2021.663370\u003c/li\u003e\n\u003cli\u003eWu S, Zheng J, Li Y, et al. Development and Validation of an MRI-Based Radiomics Signature for the Preoperative Prediction of Lymph Node Metastasis in Bladder Cancer. EBioMedicine. 2018;34:76-84. doi:10.1016/j.ebiom.2018.07.029\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Cervical squamous cell carcinoma, Biological behavior, ESWAN, Radiomics","lastPublishedDoi":"10.21203/rs.3.rs-7082528/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7082528/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003eAccurate prediction of the biological behavior of cervical squamous cell carcinoma (CSCC) is essential for optimizing therapeutic strategies and enhancing patient outcomes. This study aims to develop and validate a radiomics nomogram based on the apparent transverse relaxation rate (R2* maps) to predict deep stromal invasion (DSI), lymph node metastasis (LNM), and lymph-vascular space invasion (LVSI) in CSCC.\u003c/p\u003e\u003ch2\u003eMaterial and methods\u003c/h2\u003e\u003cp\u003eA total of 136 patients with CSCC were included. Patients were divided into two groups for each clinical characteristic: DSI (n\u0026thinsp;=\u0026thinsp;61) vs. non-DSI (n\u0026thinsp;=\u0026thinsp;75), LNM (n\u0026thinsp;=\u0026thinsp;24) vs. non-LNM (n\u0026thinsp;=\u0026thinsp;112), and LVSI (n\u0026thinsp;=\u0026thinsp;61) vs. non-LVSI (n\u0026thinsp;=\u0026thinsp;75). Radiomic features were extracted from axial MRI scans using the ESWAN sequence and post-processed to generate R2* maps, yielding 1,476 features. Clinical factors, including age, tumor diameter, SCC antigen levels, NLR, PLR, WBC count, Hgb level, tumor differentiation grade, history of irregular vaginal bleeding, and menopausal status, were also analyzed. Feature selection was performed using PCC, univariate logistic regression, ANOVA, RFE, and Relief. Clinical risk factors were identified via logistic regression. Machine learning classifiers were used to construct radiomics models, and combined models were developed by integrating the radiomics score with clinical factors. Model performance was assessed using ROC analysis, with AUC, accuracy, specificity, and sensitivity calculated. Radiomics nomograms were constructed for each condition.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eIn the training set, radiomics models outperformed clinical models for DSI (AUC 0.824 vs. 0.724), LNM (AUC 0.827 vs. 0.783), and LVSI (AUC 0.897 vs. 0.712). Combined models further improved performance, with AUC values of 0.870 for DSI, 0.837 for LNM, and 0.876 for LVSI. In the testing set, combined models outperformed clinical models for LNM (AUC 0.735 vs. 0.633) and LVSI (AUC 0.775 vs. 0.664), but not for DSI (AUC 0.619 vs. 0.677).\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e\u003cp\u003eThe R2*-based radiomics nomogram with machine learning outperforms clinical models in predicting CSCC behavior in the training cohort. Combined models boost performance further. Although validation results are mixed, this method shows potential for enhancing personalized treatment and patient management.\u003c/p\u003e","manuscriptTitle":"Predicting the Biological Behavior of Cervical Squamous Cell Carcinoma: A Machine Learning Approach Using Apparent Transverse Relaxation Rate (R2* maps) Radiomics Nomogram","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-29 11:53:25","doi":"10.21203/rs.3.rs-7082528/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-09T08:58:14+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-08T02:57:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"66671920146506369176752584736077367531","date":"2026-02-23T08:57:01+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"57369612249947007695556553630620778590","date":"2026-01-14T13:07:55+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-09-05T09:15:18+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"6354464008681476862263492376308928967","date":"2025-08-27T02:12:48+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-07-24T14:38:37+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-07-15T21:35:09+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-07-11T11:27:58+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-07-11T11:26:26+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Imaging","date":"2025-07-09T09:38:28+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1b39762f-1f3a-4cd2-a3bc-2439208e479b","owner":[],"postedDate":"July 29th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-04-07T16:03:06+00:00","versionOfRecord":{"articleIdentity":"rs-7082528","link":"https://doi.org/10.1186/s12880-026-02333-z","journal":{"identity":"bmc-medical-imaging","isVorOnly":false,"title":"BMC Medical Imaging"},"publishedOn":"2026-04-04 15:59:33","publishedOnDateReadable":"April 4th, 2026"},"versionCreatedAt":"2025-07-29 11:53:25","video":"","vorDoi":"10.1186/s12880-026-02333-z","vorDoiUrl":"https://doi.org/10.1186/s12880-026-02333-z","workflowStages":[]},"version":"v1","identity":"rs-7082528","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7082528","identity":"rs-7082528","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}