Deep Learning-Based MRI Segmentation for Non-Invasive Prediction of Microsatellite Instability in Endometrial Cancer: A Multicenter Study

Deep Learning-Based MRI Segmentation for Non-Invasive Prediction of Microsatellite Instability in Endometrial Cancer: A Multicenter Study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Deep Learning-Based MRI Segmentation for Non-Invasive Prediction of Microsatellite Instability in Endometrial Cancer: A Multicenter Study Ke Wang, Xiaoli Song, Xiaoyi Liu, Xuqing Lin, Hongjian Luo, Xinyi Gou, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8106086/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract Objective Preoperative determination of microsatellite instability (MSI) status in endometrial cancer (EC) is crucial for guiding immunotherapy, but currently relies on invasive tissue sampling. This study aimed to develop and validate a fully automated, non-invasive framework using deep learning-based MRI segmentation and radiomic analysis for preoperative prediction of MSI status in EC. Methods In this retrospective multicenter study, 867 EC patients from three institutions were enrolled (593 for model development, 274 for external testing). MSI status was determined by immunohistochemistry. A cascaded V-Net model was developed for automated tumor segmentation on T2-weighted, diffusion-weighted, and contrast-enhanced T1-weighted images. Radiomic features were extracted and selected based on interobserver reliability, redundancy removal, and least absolute shrinkage and selection operator regression. Three classifiers—support vector machine (SVM), random forest (RF), and logistic regression (LR)—were trained and externally validated. Segmentation performance was assessed using the Dice similarity coefficient (DSC), sensitivity, and specificity. Classification performance was evaluated by the area under the curve (AUC), accuracy, sensitivity, and specificity, with group comparisons using DeLong tests. Results In the external testing cohort, automated segmentation achieved mean DSCs of 85.8%±13.3% for DWI, 76.2%±17.3% for T2WI, and 80.5%±14.6% for CE-T1WI. The radiomic models based on automated segmentation yielded AUCs of 0.876 (SVM), 0.867 (RF), and 0.800 (LR) for predicting MSI status, which were not significantly different from the performance of models based on manual segmentation (AUCs: 0.915, 0.890, 0.821; all P > 0.05). Conclusion Deep learning-based automated MRI segmentation can reliably delineate EC tumors, and when combined with radiomic analysis, provides an effective, non-invasive method for preoperative prediction of MSI status. This automated framework demonstrates performance comparable to manual segmentation, supporting its potential integration into clinical workflows to aid preoperative decision-making. Endometrial cancer Microsatellite instability Deep Learning Radiomics Magnetic Resonance Imaging Segmentation Prediction Model Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Among gynecologic cancers, endometrial cancer (EC) ranks highest in prevalence, with approximately 3% of women facing a lifetime risk of its development. The increasing prevalence of risk factors such as obesity has contributed to increases in both incidence and mortality rates for EC.( 1 ) Historically, risk stratification was based on histomorphological features like grade and histological type, yet even seasoned gynecological pathologists often find these characteristics to be inconsistently reproducible.( 2 ) This clinical issue highlights the pressing need for dependable, noninvasive biomarkers to enhance preoperative risk evaluation and support personalized treatment strategies. The DNA mismatch repair (MMR) system corrects errors that occur during DNA replication, recombination, or even after therapy-induced damage. A deficiency in MMR genes leads to the accumulation of mutations in microsatellite regions, which in turn results in microsatellite instability (MSI).( 3 ) Among all tumor types, EC exhibits the highest proportion of MMR deficiency (MMR-d), with up to 25% to 31% of EC cases classified as MSI.( 4 ) In 2013, The Cancer Genome Atlas (TCGA) categorized MSI as a distinct molecular subtype of EC.( 5 ) MSI status serves as an important prognostic biomarker and is crucial for treatment planning, as MSI tend to show enhanced responsiveness to programmed death ligand 1 (PD-L1) immunotherapies compared with microsatellite-stable (MSS) cases.( 6 ) Currently, immunohistochemistry (IHC) analysis of tumor specimens is the standard method for evaluating MSI mutation status, with cases of EC classified as either MSI or MSS.( 7 ) While postoperative IHC remains the diagnostic gold standard, current MSI assessment protocols necessitate a 2–4 week waiting period for pathological specimen processing. This delay is particularly critical for patients with advanced EC, as early initiation of immunotherapy in MSI cases can significantly improve progression-free survival. A preoperative noninvasive approach could enable timely immunotherapy planning, reduce unnecessary surgeries in low-risk cases, and mitigate risks associated with invasive biopsies in frail patients. Radiomic analysis has recently emerged as a promising technique. This method involves extracting high-dimensional and quantitative features from medical images, thereby providing detailed insights into lesion morphology and heterogeneity. Such quantitative metrics have shown potential as noninvasive biomarkers for staging diseases, evaluating treatment response, and assessing prognosis.( 8 , 9 ) A recent systematic review and meta-analysis found that MRI-based radiomic analysis performed preoperatively in EC patients can reliably predict tumor grade, myometrial invasion (MI), lymph node metastasis (LNM), and lymphovascular space invasion (LVSI).( 10 ) Studies have shown that radiomic features may correlate with tumor molecular biomarkers and that these features can be used to identify the molecular subtype of the disease preoperatively.( 11 , 12 ) However, traditional manual segmentation is not only labor-intensive but also time-consuming, particularly when large datasets are involved. Consequently, deep learning–based automated segmentation techniques have garnered significant attention as a means to overcome these limitations.( 13 ) In particular, developing an automated segmentation algorithm using multicenter data is critical for addressing inter-institutional variability and improving model robustness. In this study, we aimed to develop and validate a multiparametric MRI-based deep learning automated segmentation model for patients with EC. Furthermore, by comparing automated segmentation's performance with that of manual segmentation, we explored the feasibility of integrating deep learning techniques and radiomic analysis into clinical practice for the prediction of MSI molecular subtypes in EC. Materials and Methods Patient Selection This retrospective study was approved by the Institutional Review Boards of Peking University People's Hospital (Approval No. 2025PHB057-001; Date: 2025-01-25). Given the retrospective nature of this study, the requirement for informed consent was waived. The study included 593 patients with EC who underwent preoperative pelvic MRI scans at Peking University People's Hospital (Center 1) between May 2007 and September 2023, 181 patients from Second Hospital of Shanxi Medical University (Center 2) between May 2017 and September 2022, and 93 patients from The Third Affiliated Hospital of Zunyi Medical University (Center 3) between May 2019 and November 2022. Patients from Center 1 served as the internal cross-validation (internal CV) set for training machine learning models to identify MSI status in patients with EC. Patients from Centers II and III comprised the external testing cohort. Detailed inclusion and exclusion criteria, along with a comprehensive study flowchart, are presented in Fig. 1 . Determination of MSI Status Using IHC Analysis All patients underwent a total hysterectomy, bilateral salpingo-hysterectomy, and pelvic lymph node sampling. The expression of four MMR gene proteins (MLH1, MSH2, MSH6, and PMS2) were determined by IHC, which served as the reference standard to establish MSI status. Two pathologists (each with > 10 years of experience) reviewed and confirmed the results. Patients were subsequently classified into two groups: those with all four MMR proteins expressed (designated as MSS) and those missing one or more MMR proteins (designated as MSI). MRI Acquisition Each patient received a pelvic MRI scan encompassing the entire tumor region. The imaging protocol included several sequences: Axial T2-weighted imaging (T2WI); Axial diffusion-weighted imaging (DWI) with b values of 0 and 1000 s/mm²; and Axial contrast-enhanced T1-weighted imaging (CE-T1WI) performed with gadodiamide at a dosage of 0.1 mmol/kg. All patients underwent CE-T1WI scans, and to minimize motion artifacts, 20 mg of hyoscine butylbromide was administered intramuscularly prior to imaging in all patients. The detailed MRI acquisition parameters are shown in Table 1 . Table 1 MRI acquisition parameters Parameter Scanner Sequence TR/TE (ms) FOV (mm) Slice Thickness (mm) Matrix Center1 Philips ingenia 3.0 T T2WI 3739/90.8 360 7 480 × 348 DWI 4900/58.9 380 7 128×204 CE-T1WI 3.9/1.8 400 4 320×208 GE discovery 750 3.0 T T2WI 3825/89.8 380 6 384 × 224 DWI 4875/58.9 380 6 160 × 160 CE-T1WI 4.0/1.7 400 3 288 × 224 Center2 GE discovery 750 W 3.0 T T2WI 5526/71.6 380 5 288 × 288 DWI 6500/68.8 380 5 128 × 130 CE-T1WI 6.2/2.1 400 4 256 × 256 Siemes uMR588 3.0 T T2WI 4360/88.0 340 4 320×256 Center3 DWI 5600/60.0 340 1 128×78 CE-T1WI 4.8/1.3 400 3 320 × 203 Preprocessing and Segmentation Figure 2 shows the workflow of the DL and radiomic study. To reduce bias and eliminate intensity differences from various imaging protocols, all raw MR images were preprocessed before extracting features. Specifically, volumes of interest (VOIs) were normalized using z-score gray-level standardization and resampled to a uniform resolution (1 mm × 1 mm × 1 mm) via B-spline interpolation. The intensity range was limited to µ ± 3σ to lessen the impact of contrast and brightness variations. ( 14 ) Radiologists with at least six years of pelvic MRI experience from each institution manually outlined the three-dimensional tumor VOIs on axial oblique slices from T2WI, CE-T1WI, and DWI sequences. These contours were then reviewed by a senior radiologist with over 20 years of experience. To evaluate the consistency of manual segmentation, two radiologists from different institutions jointly and randomly selected T1WI, T2WI, and DWI images from 150 patients for repeated segmentation. Although the reviewer knew the patients had endometrial carcinoma, they were blinded to the detailed clinical information and histopathological findings. Deep Learning–Based Automated Segmentation The automated segmentation framework was built on the V-Net architecture, which is effective for volumetric medical image segmentation and uses skip connections to improve information flow between layers. Given the high resolution, large field of view (FOV) of pelvic MRI, and challenges from multi-center data differences, a cascaded lightweight V-Net structure was used. The first V-Net processed down-sampled, large FOV image blocks to roughly localize the tumor, while the second V-Net refined the segmentation using high-resolution data. Figure 3 shows the original V-Net alongside its cascaded framework for clarity. Training and testing of the V-Net were conducted in two steps: internal CV and external testing. For internal CV, data from 593 patients at Center 1 were used with a 5-fold cross-validation scheme to train, test, and evaluate the segmentation method. External testing was carried out on data from 274patients at Center 2 and Center 3 using the best model from internal CV. The network underwent training for 1,200 iterations. A learning rate of 10⁻⁵ and a batch size of 16 were used, with cross-entropy employed as the loss function. Optimization was achieved using the Adam optimizer ( 15 ). All experiments were conducted on the uAI Research Portal (version 20240730) platform, and the V-Net network was implemented using the Pytorch framework.( 16 ) Feature Extraction and Selection Radiomic features from VOIs across different MRI sequences were automatically extracted using the PyRadiomics software package (version 3.0.1). The extracted features were categorized as first-order (n = 18), shape-based (n = 14), and higher-order (n = 75), all standardized using z scores to remove scale differences. A three-step feature selection process was used to identify the most predictive features. First, data from 100 randomly selected patients were reviewed by two radiologists from different centers (each with at least six years of experience), and the intraclass correlation coefficient (ICC) analysis was applied to retain features with ICC > 0.85. Next, Spearman’s rank correlation test was used to remove redundant features (correlation coefficient ≥ 0.9).Finally,the least absolute shrinkage and selection operator (LASSO) method—with 5-fold CV and an optimized penalty parameter—was applied to keep features with nonzero coefficients that were relevant to MSI status in EC patients. Classification Model Construction Dataset 1 from Center 1, containing 593 patients (497 with MSS and 96 with MSI), was used to test the feasibility of radiomic analysis for predicting MSI status in EC. Model training and feature selection were performed using 5-fold CV, with the adaptive synthetic (ADASYN) sampling technique applied in each training fold to address class imbalance.( 17 ) Based on the selected radiomic features, three classifiers—support vector machine (SVM), random forest (RF), and logistic regression (LR)—were developed using the scikit-learn library in Python (version 3.6). These machine learning algorithms were optimized via grid search and 5-fold CV. External testing, conducted on data from Centers 2 and 3 (totaling 230 MSS and 44 MSI cases), compared the classification performance of models based on manual versus automated segmentation, using the feature sets derived from internal CV. Statistical Analysis For the automated segmentation model, performance metrics including the Dice similarity coefficient (DSC), sensitivity, and specificity were calculated for the entire tumor, tumor core, and enhanced tumor regions. The DSC measured the overlap between the automated segmentation and the ground truth relative to the total area. Recall was defined as the ratio of correctly detected tumor voxels to total tumor voxels, and precision as the ratio of correctly detected tumor voxels to all voxels identified by the automated segmentation. For radiomics classification models, receiver operating characteristic (ROC) curves were generated to assess performance, with optimal cutoff points determined using the Youden index on the training cohort and then applied to two independent validation cohorts. Model performance was evaluated using metrics such as AUC, accuracy, sensitivity, and specificity, and the DeLong test was used to compare AUCs among the three models. All statistical tests were two-tailed, with a P value < 0.05 considered statistically significant. Results Clinical Characteristics Table 2 summarizes the clinical characteristics of patients. In the internal CV cohort, the prevalence of MSI was 16.19% (96 out of 593 patients), while in the external testing cohort it was 16.06% (44 out of 274 patients), which suggests a comparable distribution (P = 0.942). Notably, in the internal CV cohort, lymphovascular space invasion (LVSI) differed markedly, while tumor grading was significantly distinct in the external testing group. Table 2 Clinicopathologic Characteristics of Patients With MSI and MSS Tumors in the Internal CV and External Validation Cohorts Clinicopathologic Characteristic Subgroup Internal CV Cohort External Validation Cohort MSI ( n = 96) MSS ( n = 497) P Value MSI ( n = 44) MSS ( n = 230) P Value Mean age ± standard deviation (y) 58.4 ± 6.9 57.2 ± 9.5 0.372 56.9 ± 8.3 58.9 ± 7.7 0.266 Histologic Grade, n (%) High 33 (34.8) 202 (41.0) 0.451 12 (27.3) 103 (44.8) 0.043 Medium 39 (40.6) 187 (37.9) 23 (52.3) 77 (33.5) Low 24 (25.0) 104 (21.1) 9 (20.5) 50 (21.7) FIGO stage, n (%) I 76(79.2) 418(84.1) 0.227 34(77.3) 187(81.3) 0.334 II 8(7.3) 34(6.8) 4(9.1) 25(10.9) III 11(11.5) 36(7.2) 6(13.6) 17(7.4) IV 1(1.0) 5(1.0) 0(0.0) 1(0.4) Deep Myometrial invasion, n (%) Present 20 (20.8) 107 (21.5) 0.986 14 (31.8) 58 (25.2) 0.469 Absent 74 (77.1) 390 (78.5) 30 (68.2) 172 (74.8) Lymph vascular space invasion, n (%) Present 19 (19.8) 54 (10.9) 0.023 11 (25.0) 39 (17.0) 0.293 Absent 77 (80.2) 443 (89.1) 33 (75.0) 191 (83.0) Lymph node metastasis, n (%) Present 10 (10.4) 35 (7.0) 0.351 7 (15.9) 21 (9.6) 0.316 Absent 86 (89.6) 462 (93.0) 37 (84.1) 201 (90.4) Maximum tumor diameter, n (%) < 5.0 cm 59 (61.5) 335 (67.2) 0.312 25 (56.8) 159 (76.3) 0.156 ≥ 5.0 cm 37 (38.5) 162 (32.8) 19 (43.2) 71 (30.9) CV = cross-validation; MSI = microsatellite instability; MSS = microsatellite stability. Inter-observer Agreement in Manual Segmentation High inter-observer consistency was demonstrated by DSC of 89.4% ± 16.6% (T2WI), 91.2% ± 5.4% (DWI), and 92.2% ± 7.9% (CE-T1WI) between the two radiologists. Deep Learning–Based Automated Segmentation In the external testing set, the mean DSC ± standard deviation values were 76.2% ± 17.3% (95% confidence interval [CI], 72.8%-78.5%) for T2WI, 85.8% ± 13.3% (95% CI, 83.8%-87.3%) for DWI, and 80.5% ± 14.6% (95% CI, 76.2%-81.7%) for CE-TIWI (Table 3 ). Figure 4 illustrates typical cases of lesion segmentation for T2WI, DWI, and CE-TIWI images. Table 3 Performance of Automated Segmentation in the External Validation Cohort Sequence Dice Similarity Coefficient Recall Precision Mean ± Standard Deviation MEDIAN Mean ± Standard Deviation Mean ± Standard Deviation T2WI 76.2% ± 17.3% 77.9% 85.4% ± 17.2% 71.9% ± 6.3% DWI 85.8% ± 13.3% 87.4% 83.3% ± 14.2% 78.6% ± 8.1% CE-T1WI 80.5% ± 14.6% 84.6% 82.93% ± 16.4% 73.4% ± 15.0% CE-T1WI = contrast-enhanced T1-weighted imaging; DWI = diffusion-weighted imaging; T2WI = T2-weighted imaging. Radiomics Analysis Among the 5,652 radiomic features extracted from T2WI, DWI, and CE-TIWI, features with ICC values 0.9), followed by univariate logistic regression selection. Ultimately, 29 features were selected through LASSO regression, including 7 from T2WI, 8 from DWI, and 14 from CE-TIWI, to establish the radiomic model. Analysis of the selected radiomic features revealed no significant correlations among them, as illustrated by Fig. 5 . In the internal CV cohort, the mean AUCs obtained through 5-fold CV using the automated segmentation model were 0.876 (95% CI, 0.842–0.906) with SVM, 0.937 (95% CI, 0.911–0.960) with RF, and 0.812 (95% CI, 0.765–0.853) with LR. For the external testing cohort, the AUCs achieved by the automated segmentation model were 0.876 (95% CI, 0.814–0.927) with SVM, 0.867 (95% CI, 0.780–0.942) with RF, and 0.800 (95% CI, 0.701–0.887) with LR, and the AUCs achieved by the manual segmentation model were 0.915 (95% CI, 0.853–0.962) with SVM, 0.890 (95% CI, 0.829–0.946) with RF, and 0.821 (95% CI, 0.751–0.883) with LR (Table 4 ). Table 4 Classification Performance of Different Classifiers Based on Different Segmentation Methods for Predicting the MSI Status of the External Validation Set Cohort Classifier Segmentation Method Accuracy Sensitivity Specificity AUC (95% CI) P S Value a P C Value b Internal CV RF Manual 0.897 0.75 0.926 0.937 (0.911–0.960) 3.34E-11 LR Manual 0.694 0.865 0.541 0.812 (0.765–0.853) 3.04E-25 SVM Manual 0.804 0.792 0.807 0.876 (0.842–0.906) 0.183 External validation RF Automated 0.874 0.794 0.882 0.867 (0.780–0.942) 0.167 8.10E-29 Manual 0.928 0.568 0.989 0.915 (0.853–0.962) 2.68E-26 LR Automated 0.767 0.833 0.64 0.800 (0.701–0.887) 0.294 6.55E-17 Manual 0.764 0.886 0.627 0.821 (0.751–0.883) 1.27E-16 SVM Automated 0.831 0.778 0.84 0.876 (0.814–0.927) 0.077 0.844 Manual 0.798 0.864 0.787 0.890 (0.829–0.946) 0.561 AUC = area under the curve; CI = confidence interval; CV = cross-validation; LR = logistic regression; RF = random forest; SVM = support vector machine. a Significance level for comparisons between different segmentation methods (automated vs manual) with the same classifiers. b Significance level for comparisons between RF and LR, LR and SVM, and RF and SVM in each cohort. In the internal CV cohort, RF was significantly superior to SVM and LR. However, in the external testing set, there was no significant difference between RF and SVM with manual segmentation ( P = 0.561) or with automated segmentation ( P = 0.844), and both were significantly better than LR. For models based on the same classifiers (SVM, RF, and LR) but different segmentation methods, the differences in AUCs were not significant ( P = 0.077–0.294). Figure 6 presents the ROC curves and DCA results for the classification models. Discussion In this investigation, we employed an automated segmentation approach based on a V-Net cascading model to delineate lesions on multimodal MR images of EC, and we then developed a radiomic classification framework to evaluate MSI/MSS status. The segmentation demonstrated strong performance, achieving DSCs in the external testing cohort of 85.8% ± 13.3% with DWI, 76.2% ± 17.3% with T2WI, and 80.5% ± 14.6% with CE-T1WI. Radiomics analysis using automated segmentation exhibited performance comparable to manual segmentation, with consistent accuracy across multi-center datasets. These results underscore the efficacy and generalizability of our automated approach despite inherent heterogeneity in multi-Center 1maging data. The inherent complexity and pronounced heterogeneity characteristic of EC imaging data pose significant challenges to accurate tumor segmentation, particularly in multicenter settings.( 18 ) Current research in this field demonstrates notable limitations: Hodneland et al.'s pioneering study( 19 ) employed contrast-enhanced T1-weighted imaging (CE-T1WI) as the sole input for machine learning segmentation, achieving median DSC of 0.84 and 0.77 across two readers. However, this single-center investigation (n = 34) restricted its analysis to CE-T1WI, thereby constraining the generalizability of its findings. Kurata et al.( 20 ) advanced the methodology through multisequence integration (T2WI, DWI, and ADC maps) combined with ensemble learning, attaining a mean DSC of 0.806 with sensitivity and positive predictive value (PPV) of 0.816 and 0.834, respectively. Nevertheless, the single-institution origin of these data raises concerns regarding model robustness against protocol variability. Our cascaded V-Net addresses multicenter variability through coarse-to-fine localization, mimicking radiologists’ workflow. The proposed cascaded V-Net model demonstrated high segmentation consistency across multi-center MRI data, external validation across three institutions with heterogeneous MRI protocols demonstrates unprecedented generalizability. Yet segmentation failures still occurred in certain cases (Supplementary Material Fig. S1 ). First, the low contrast between tumors and adjacent tissues on T2WI sequences may hinder precise boundary delineation, particularly in cases with infiltrative tumors or coexisting adenomyosis. Second, variations in image quality across different scanners (e.g., resolution, signal-to-noise ratio, or artifact levels) could destabilize the model. For instance, low-field MRI images in the external test set exhibited local detail loss due to noise interference, thereby compromising segmentation accuracy. Furthermore, segmentation errors for small lesions disproportionately reduced DSC scores, as the limited voxels in minute tumors amplify the impact of minor segmentation inaccuracies. Finally, interference from adjacent pathologies (e.g., concurrent uterine fibroids) might lead to misidentification of tumor regions, especially when malignant and benign lesions exhibit overlapping imaging features. Despite these limitations, the observed failure cases remained sporadic, and the model's overall performance in external validation proved comparable to manual segmentation, thereby confirming its clinical utility. Comparative analysis with existing literature underscores our model's translational superiority. Lin et al.( 21 ) d developed a model to predict MSI status in endometrial cancer using T2WI and CE-T1WI from 296 patients, the classification models achieved AUC values of 0.752 in the training cohort and 0.723 in the validation cohort. Jia et al.( 22 ) implemented a multiparametric MRI radiomics framework (T2WI, CE-T1WI, ADC) using SVM, attaining exceptional AUCs of 0.905 (training), 0.875 (internal test), and 0.862 (external test). In our investigation, both automated and manual segmentation-based classifiers demonstrated robust MSI classification performance. While RF exhibited superior training performance (AUC = 0.85 vs. SVM = 0.80), their external test accuracies showed no statistically significant difference (p > 0.05). This observation aligns with established machine learning principles - RF's ensemble methodology reduces feature/sample dependency to prevent overfitting, whereas SVM optimizes hyperplane determination for minimal misclassification. Both algorithms have been extensively validated in oncological applications including genomic profiling and survival prediction.( 23 , 24 ) The comparatively lower LR performance (AUC = 0.72) likely reflects its limited capacity to model complex radiomic feature interactions, suggesting LR’s linear decision boundaries likely failed to capture nonlinear interactions among textural features, whereas ensemble methods like RF inherently model complex hierarchies.( 25 ) Class imbalance in machine learning model development can significantly compromise classifier performance. To address this challenge, our study employed the Adaptive Synthetic Sampling (ADASYN) technique, which dynamically generates synthetic samples of the minority class within each cross-validation fold of the training set to enhance model learning. Crucially, the application of ADASYN strictly adheres to a core principle in machine learning: oversampling is exclusively applied during the training phase, while validation and external test sets retain their original data distributions. Specifically, in the 5-fold cross-validation framework, ADASYN was independently implemented on the training subset of each fold, whereas validation subsets remained untouched by synthetic sample generation. This design effectively prevents data leakage and ensures unbiased model evaluation. Furthermore, we mitigated potential noise sensitivity induced by oversampling through LASSO regression for discriminative feature selection, coupled with regularization strategies such as kernel function constraints in SVM and out-of-bag error evaluation in RF. Notably, the model demonstrated consistent classification performance between internal validation (AUC = 0.876) and external testing cohorts, with no significant difference compared to manual segmentation (P > 0.05). This alignment underscores the method’s robust generalizability in real-world scenarios. Collectively, the synergistic integration of rigorous data isolation protocols and regularization techniques successfully balances the challenges of class imbalance and overfitting risks, establishing a reliable technical framework for radiomics-based molecular subtyping research. Notably, more than half of the radiomic features used in our approach were textural (20/29), with GLSZM being the most prevalent (9/29), suggesting that GLSZM may be more closely associated with MMR genes. Radiomic feature analysis has demonstrated significant differences in GLCM entropy values between Lynch syndrome patients (a hereditary tumor syndrome caused by germline mutations in MMR genes) and the control cohort.( 26 ).As an extension of the GLCM methodology, the GLSZM analyzes the distribution characteristics of spatially connected pixel clusters, with its regional entropy parameter effectively quantifying the uncertainty in both zone size and grayscale value distribution.( 27 ). The GLSZM features were higher in the MSI group compared to the MSS group overall (Supplementary Material Fig.S2). This imaging-based analytical approach enables precise characterization of tumor microstructural heterogeneity, where elevated entropy values suggest increased randomness in textural patterns. The observed feature elevation in MSI tumors aligns with their inherently disordered growth patterns. Notably, previous clinical studies have confirmed that endometrial carcinomas with MSI positivity are typically associated with higher tumor grades, deeper myometrial invasion, and increased lymph node metastasis rates, consistent with the more heterogeneous microenvironment reflected by these quantitative imaging features.( 28 , 29 ) These pathological characteristics correlate with histopathological evidence of enhanced intratumoral architectural complexity observed in mismatch repair-deficient lesions. Importantly, radiomics-based spatial heterogeneity analysis provides a non-invasive means to characterize these microstructural spatial variations, demonstrating remarkable consistency with histomorphological findings.( 30 ) The determination of MSI status in EC carries substantial clinical relevance, as it facilitates Lynch syndrome screening and informs prognosis and therapeutic strategies, particularly regarding immunotherapy response.( 31 ) Conventional tissue-based methods, such as PCR and IHC, are constrained by their invasiveness, cost, time requirements, and dependence on sufficient tissue samples, which may delay clinical decisions. Our radiomics framework offers a rapid, non-invasive preoperative alternative that enables molecular subtyping and enhances risk stratification accuracy, thereby addressing a critical unmet clinical need. Furthermore, there were no significant differences between the segmentation strategies (manual vs. automated) across the three classifiers. These findings suggest the automated segmentation method performs robustly in radiomic analysis. Although the underlying pathological mechanisms remain unclear, our results imply that MSI tumors may exhibit greater histological heterogeneity than MSS tumors—a difference that radiomic classifiers appear capable of capturing. This study had several limitations. First, the number of positive samples was relatively small. Although oversampling was used to mitigate the data distribution imbalance to some extent, a larger and more uniformly distributed dataset, especially from different centers, would facilitate the training of more robust segmentation and radiomic models. Nevertheless, the outcomes were statistically reliable given the current sample size. Although IHC is a reliable and cost-effective technique for assessing MMR proteins and determining MSI status, PCR is still considered the gold standard for MSI evaluation. A previous study found that the consistency between PCR and IHC in determining MSI status is nearly 92%.( 32 ) Finally, this study primarily focused on evaluating MSI status as a single indicator of EC. Tumor occurrence, development, and treatment response are influenced by various biological characteristics; therefore, further optimization of the experimental design is needed to conduct comprehensive analyses of multiple biological characteristics. Conclusion In this multicenter study, we demonstrated that a deep learning-based automated segmentation model achieved performance comparable to manual segmentation, and when combined with radiomic analysis, provided a robust, non-invasive method for the preoperative prediction of MSI status in endometrial cancer By integrating our framework into preoperative MRI workflows, clinicians could stratify EC patients into MSI/MSS subgroups before surgery. Preoperative identification of MSI status enables early enrollment in neoadjuvant immunotherapy trials, particularly for patients with high surgical risk or those seeking fertility preservation. Conversely, MSS patients may avoid costly immunotherapy with limited benefit. Future efforts should prioritize multi-center validation, integration of multi-modal biomarkers, and prospective trials to evaluate therapeutic impact. Enhancing model interpretability and addressing ethical challenges will be critical for real-world translation. These advances could streamline precision oncology workflows in EC management. Abbreviations ADASYN Adaptive Synthetic Sampling approach for imbalanced learning AUC Area Under the Curve CE-T1WI Contrast-Enhanced T1-weighted Imaging CV Cross - Validation DSC Dice Similarity Coefficient DWI Diffusion-Weighted Imaging EC Endometrial Cancer ICC Intraclass Correlation Coefficient IHC Immunohistochemistry LASSO Least Absolute Shrinkage and Selection Operator LR Logistic Regression LNM Lymph Node Metastasis LVSI Lymphovascular Space Invasion MI Myometrial Invasion MMR DNA Mismatch Repair MSS Microsatellite Stability MSI Microsatellite Instability MRI Magnetic Resonance Imaging PPV Positive Predictive Value RF Random Forest ROC Receiver Operating Characteristic SVM Support Vector Machine T2WI T2-weighted Imaging VOI Volume of Interest Declarations Ethics approval and consent to participate This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Institutional Review Boards of Peking University People's Hospital (Approval No. 2025PHB057-001; Date: 2025-01-25). Due to the retrospective nature of this study involving the analysis of existing data, the requirement for informed consent was waived by the Institutional Review Boards of Peking University People's Hospital. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Funding This work was supported by the National Key Research and Development Program of China (Grant Number: 2022YFC2704405). Author Contribution Ke Wang and Xiaoli Song were responsible for data collection and analysis and were the main contributors to writing the manuscript. Xinyi Gou and Xuqing Lin analyzed the images. Hongjian Luo examined the patients and provided images. Yi Wang and Nan Hong analyzed the data. Rong Zhou and Jin Cheng were responsible for the study design, and the final approval of the version to be published. All authors read and approved the final manuscript. Acknowledgements Not applicable.. Data Availability The datasets generated and/or analysed during the current study are not publicly available due to privacy or ethical restrictions, but are available from the corresponding author on reasonable request. References Crosbie EJ, Kitson SJ, McAlpine JN, Mukhopadhyay A, Powell ME, Singh N. Endometrial cancer. Lancet. 2022;399(10333):1412–28. Moribata Y, Kurata Y, Nishio M, Kido A, Otani S, Himoto Y, et al. Automatic segmentation of bladder cancer on MRI using a convolutional neural network and reproducibility of radiomics features: a two-center study. Sci Rep. 2023;13:628. Meyer LA, Broaddus RR, Lu KH. Endometrial cancer and Lynch syndrome: clinical and pathologic considerations. Cancer Control J Moffitt Cancer Cent. 2009;16(1):14–22. Ryan NaJ, Glaire MA, Blake D, Cabrera-Dandy M, Evans DG, Crosbie EJ. The proportion of endometrial cancers associated with Lynch syndrome: a systematic review of the literature and meta-analysis. Genet Med. 2019;21(10):2167. Cancer Genome Atlas Research Network, Kandoth C, Schultz N, Cherniack AD, Akbani R, Liu Y, et al. Integrated genomic characterization of endometrial carcinoma. Nature. 2013;497(7447):67–73. Arafa M, Shebl AM, Salama A, ElZahaf E, Ashamallah SA, Foda AA, et al. Correlation of PD-L1 immunohistochemical expression with microsatellite instability and p53 status in endometrial carcinoma. Eur J Obstet Gynecol Reprod Biol X. 2022;16:100172. Ribeiro-Santos P, Martins Vieira C, Viana Veloso GG, Vieira Giannecchini G, Parenza Arenhardt M, Müller Gomes L et al. Tailoring Endometrial Cancer Treatment Based on Molecular Pathology: Current Status and Possible Impacts on Systemic and Local Treatment. Int J Mol Sci 2024 July 15;25(14):7742. Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016;278(2):563–77. Bera K, Braman N, Gupta A, Velcheti V, Madabhushi A. Predicting cancer outcomes with radiomics and artificial intelligence in radiology. Nat Rev Clin Oncol. 2022;19(2):132–46. Di Donato V, Kontopantelis E, Cuccu I, Sgamba L, Golia D’Augè T, Pernazza A et al. Magnetic resonance imaging-radiomics in endometrial cancer: a systematic review and meta-analysis. Int J Gynecol Cancer Off J Int Gynecol Cancer Soc 2023 July 3;33(7):1070–6. Huang G, Du S, Gao S, Guo L, Zhao R, Bian X, et al. Molecular subtypes of breast cancer identified by dynamically enhanced MRI radiomics: the delayed phase cannot be ignored. Insights Imaging. 2024;15(1):127. Gao R, Pang J, Lin P, Wen R, Wen D, Liang Y et al. Identification of clear cell renal cell carcinoma subtypes by integrating radiomics and transcriptomics. Heliyon 2024 June 15;10(11):e31816. Soneji ND, Bharwani N, Ferri A, Stewart V, Rockall A. Pre-operative MRI staging of endometrial cancer in a multicentre cancer network: can we match single centre study results? Eur Radiol. 2018;28(11):4725–34. Meyer HJ, Renatus K, Höhn AK, Hamerla G, Schopow N, Fakler J, et al. Texture analysis parameters derived from T1-and T2-weighted magnetic resonance images can reflect Ki67 index in soft tissue sarcoma. Surg Oncol. 2019 Sept;30:92–7. Arcos-García Á, Álvarez-García JA, Soria-Morillo LM. Deep neural network for traffic sign recognition systems: An analysis of spatial transformers and stochastic optimisation methods. Neural Netw Off J Int Neural Netw Soc. 2018;99:158–65. Wu J, Xia Y, Wang X, Wei Y, Liu A, Innanje A, et al. uRP: An integrated research platform for one-stop analysis of medical images. Front Radiol. 2023;3:1153784. He H, Bai Y, Garcia EA, Li S. ADASYN: Adaptive synthetic sampling approach for imbalanced learning. In: 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence) [Internet]. 2008 [cited 2024 Dec 15]. pp. 1322–8. Available from: https://ieeexplore.ieee.org/document/4633969 Tian S, Liu A, Zhu W, Li Y, Chen L, Chen A, et al. Difference in Diffusion-Weighted Magnetic Resonance Imaging and Diffusion Tensor Imaging Parameters Between Endometrioid Endometrial Adenocarcinoma and Uterine Serous Adenocarcinoma: A Retrospective Study. Int J Gynecol Cancer Off J Int Gynecol Cancer Soc. 2017;27(8):1708–13. Hodneland E, Dybvik JA, Wagner-Larsen KS, Šoltészová V, Munthe-Kaas AZ, Fasmer KE, et al. Automated segmentation of endometrial cancer on MR images using deep learning. Sci Rep. 2021;11:179. Kurata Y, Nishio M, Moribata Y, Kido A, Himoto Y, Otani S, et al. Automatic segmentation of uterine endometrial cancer on multi-sequence MRI using a convolutional neural network. Sci Rep. 2021 July;14:11:14440. Lin Z, Wang T, Li H, Xiao M, Ma X, Gu Y, et al. Magnetic resonance-based radiomics nomogram for predicting microsatellite instability status in endometrial cancer. Quant Imaging Med Surg. 2023;13(1):108–20. Jia Y, Hou L, Zhao J, Ren J, Li D, Li H, et al. Radiomics analysis of multiparametric MRI for preoperative prediction of microsatellite instability status in endometrial cancer: a dual-center study. Front Oncol. 2024;14:1333020. Wu J, Hicks C. Breast Cancer Type Classification Using Machine Learning. J Pers Med. 2021;11(2):61. Chu LC, Park S, Kawamoto S, Fouladi DF, Shayesteh S, Zinreich ES, et al. Utility of CT Radiomics Features in Differentiation of Pancreatic Ductal Adenocarcinoma From Normal Pancreatic Tissue. AJR Am J Roentgenol. 2019;213(2):349–57. Chen D, Afzal N, Sohn S, Habermann EB, Naessens JM, Larson DW, et al. Postoperative bleeding risk prediction for patients undergoing colorectal surgery. Surgery. 2018;164(6):1209–16. Wang X, Wu K, Li X, Jin J, Yu Y, Sun H. Additional Value of PET/CT-Based Radiomics to Metabolic Parameters in Diagnosing Lynch Syndrome and Predicting PD1 Expression in Endometrial Carcinoma. Front Oncol [Internet]. 2021 May 12 [cited 2024 Dec 30];11. Available from: https://www.frontiersin.org/journals/oncology/articles/ 10.3389/fonc.2021.595430/full van Griethuysen JJ, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res. 2017;77(21):e104–7. Bilbao C, Lara PC, Ramírez R, Henríquez-Hernández LA, Rodríguez G, Falcón O, et al. Microsatellite instability predicts clinical outcome in radiation-treated endometrioid endometrial cancer. Int J Radiat Oncol Biol Phys. 2010;76(1):9–13. Bae H, Rha SE, Kim H, Kang J, Shin YR. Predictive Value of Magnetic Resonance Imaging in Risk Stratification and Molecular Classification of Endometrial Cancer. Cancers. 2024;16(5):921. Abbasian Ardakani A, Bureau NJ, Ciaccio EJ, Acharya UR. Interpretation of radiomics features-A pictorial review. Comput Methods Programs Biomed. 2022;215:106609. Zong L, Sun Z, Mo S, Lu Z, Yu S, Xiang Y, et al. PD-L1 expression in tumor cells is associated with a favorable prognosis in patients with high-risk endometrial cancer. Gynecol Oncol. 2021 Sept;162(3):631–7. McConechy MK, Talhouk A, Li-Chang HH, Leung S, Huntsman DG, Gilks CB, et al. Detection of DNA mismatch repair (MMR) deficiencies by immunohistochemistry can effectively diagnose the microsatellite instability (MSI) phenotype in endometrial carcinomas. Gynecol Oncol. 2015;137(2):306–10. Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterial.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 16 Apr, 2026 Reviews received at journal 11 Jan, 2026 Reviews received at journal 05 Jan, 2026 Reviewers agreed at journal 03 Jan, 2026 Reviewers agreed at journal 30 Dec, 2025 Reviewers invited by journal 23 Dec, 2025 Editor invited by journal 21 Nov, 2025 Editor assigned by journal 21 Nov, 2025 Submission checks completed at journal 21 Nov, 2025 First submitted to journal 13 Nov, 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-8106086","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":564665251,"identity":"dbb3e693-7b98-4f2e-b536-2e10a975262c","order_by":0,"name":"Ke Wang","email":"","orcid":"","institution":"Peking University People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ke","middleName":"","lastName":"Wang","suffix":""},{"id":564665252,"identity":"ba7acef9-1967-4149-aa3f-864d7cd7fe27","order_by":1,"name":"Xiaoli Song","email":"","orcid":"","institution":"Second Hospital of Shanxi Medical University","correspondingAuthor":false,"prefix":"","firstName":"Xiaoli","middleName":"","lastName":"Song","suffix":""},{"id":564665253,"identity":"a4f02c46-9f61-4743-a1d4-28bf6a806323","order_by":2,"name":"Xiaoyi Liu","email":"","orcid":"","institution":"Peking University People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xiaoyi","middleName":"","lastName":"Liu","suffix":""},{"id":564665255,"identity":"f165e1e2-15d6-4f3f-8005-d2f2e4e3f620","order_by":3,"name":"Xuqing Lin","email":"","orcid":"","institution":"Peking University People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xuqing","middleName":"","lastName":"Lin","suffix":""},{"id":564665262,"identity":"631009c0-926c-4516-a0e9-e18a30f5f757","order_by":4,"name":"Hongjian Luo","email":"","orcid":"","institution":"The Third Affiliated Hospital of Zunyi Medical University, The First People’s Hospital of Zunyi)","correspondingAuthor":false,"prefix":"","firstName":"Hongjian","middleName":"","lastName":"Luo","suffix":""},{"id":564665264,"identity":"4118dd58-24b7-4284-baef-f27446e013ff","order_by":5,"name":"Xinyi Gou","email":"","orcid":"","institution":"Peking University People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xinyi","middleName":"","lastName":"Gou","suffix":""},{"id":564665266,"identity":"36686248-4453-4044-9743-aa5e464b4050","order_by":6,"name":"Nan Hong","email":"","orcid":"","institution":"Peking University People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Nan","middleName":"","lastName":"Hong","suffix":""},{"id":564665268,"identity":"543cfb1f-ee7a-4d1c-874c-0a6af9b607c3","order_by":7,"name":"Yi Wang","email":"","orcid":"","institution":"Peking University People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yi","middleName":"","lastName":"Wang","suffix":""},{"id":564665270,"identity":"3d6c79ac-5758-4eff-aab6-ab92a75b287a","order_by":8,"name":"Rong Zhou","email":"","orcid":"","institution":"Peking University People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Rong","middleName":"","lastName":"Zhou","suffix":""},{"id":564665271,"identity":"65266cce-3d02-4faf-9ef0-06cb94037ab0","order_by":9,"name":"Jin Cheng","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABAElEQVRIiWNgGAWjYDACCRDBAyR5GBiYGSqAHHYGBsYG4rWcAdLMRGkBawNqYWwjQgv/7OZjD7/IWOSZ85wxe1w4zyZxPzPzwYczGOzkdHHok7hzLN1Yhkei2LK3x9x45ra0xB5mtmTDDQzJxmYHsGsxkMgxk5bgkUjccJ7HTJp322GgFh4zyQcMBxK34dSS/w1JyxyitOSwSX4AaTnbA9TSANWyAY8WiRtpZtLAQE7c2XOs3JjnWJpxz2GgX2YY4PYL/4zkZ5I/e+oSt/Mkb3vMU2Mj297efPBhT4WdHC4tIMDM2wN0IQMDG7KDcSsHAcYfPzC0jIJRMApGwShAAADkUFTW/0OfRQAAAABJRU5ErkJggg==","orcid":"","institution":"Peking University People’s Hospital","correspondingAuthor":true,"prefix":"","firstName":"Jin","middleName":"","lastName":"Cheng","suffix":""}],"badges":[],"createdAt":"2025-11-13 13:23:06","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8106086/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8106086/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":99217655,"identity":"b241c49e-b316-484b-9abd-0cf0d1f53abf","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":94223,"visible":true,"origin":"","legend":"","description":"","filename":"AuthorManuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/09fe391e0e4959366892d357.docx"},{"id":99318169,"identity":"c932208e-5f43-4117-becf-410127d55525","added_by":"auto","created_at":"2025-12-31 16:31:45","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25610,"visible":true,"origin":"","legend":"","description":"","filename":"Table.docx","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/3c9774b8101fd33110e46218.docx"},{"id":99318266,"identity":"a89e0d10-65f4-4649-b348-ea77a642ae84","added_by":"auto","created_at":"2025-12-31 16:32:30","extension":"tiff","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2101324,"visible":true,"origin":"","legend":"","description":"","filename":"Fig3.tiff","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/82da6cfd445cddac9cc679df.tiff"},{"id":99217657,"identity":"73759e34-fc4a-4e86-8649-8cf3b07ad9fd","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"json","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":11147,"visible":true,"origin":"","legend":"","description":"","filename":"bd834e6d8088452198e6fb93e9c2e6e1.json","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/d0427f0cae4ff2713e47d92f.json"},{"id":99217658,"identity":"3339325f-bc3d-4e8b-9d9f-dd966678525d","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"docx","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1037790,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/25b0a574d98b885725d6d584.docx"},{"id":99319803,"identity":"83a45c71-0df9-4c0f-ac34-6e51cef65059","added_by":"auto","created_at":"2025-12-31 16:37:52","extension":"xml","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":161142,"visible":true,"origin":"","legend":"","description":"","filename":"bd834e6d8088452198e6fb93e9c2e6e11enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/ed0c0d5eee4b9737b534a679.xml"},{"id":99317150,"identity":"627331af-bb6a-4c45-b035-02681ebe7110","added_by":"auto","created_at":"2025-12-31 16:29:42","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":733024,"visible":true,"origin":"","legend":"","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/0c53c0587a8e6e2640ceb39a.png"},{"id":99318059,"identity":"dccc810f-901c-403b-b534-42d9cfd163ad","added_by":"auto","created_at":"2025-12-31 16:31:18","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1874656,"visible":true,"origin":"","legend":"","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/3805e0b5337953e0af0c34f6.png"},{"id":99217662,"identity":"0d038aa0-4bd1-4e31-884f-d120d6aede02","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"tiff","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2101324,"visible":true,"origin":"","legend":"","description":"","filename":"Fig3.tiff","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/f0b9b5a51ea2374d9c9cfc89.tiff"},{"id":99318261,"identity":"d9017fbd-e05d-47c2-86bf-8f981666b27b","added_by":"auto","created_at":"2025-12-31 16:32:26","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2916095,"visible":true,"origin":"","legend":"","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/402edebe46025c5ce8730dc1.png"},{"id":99217678,"identity":"f837be83-e2b2-4d95-b24c-1733422fca9d","added_by":"auto","created_at":"2025-12-30 09:11:26","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5711058,"visible":true,"origin":"","legend":"","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/8171d80a713ba4cb5f193ed7.png"},{"id":99217674,"identity":"f776c53e-5f00-4155-89be-1204071ff068","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3533018,"visible":true,"origin":"","legend":"","description":"","filename":"Fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/195a1696263b2e517ae74092.png"},{"id":99217676,"identity":"ede2b175-e0e8-41d3-ab74-8a8fc83c9aa2","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":184156,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig1.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/b147cb59afbe0c0c96adf522.png"},{"id":99317248,"identity":"7c6e4e8f-febe-4ee7-909b-292d2eece73d","added_by":"auto","created_at":"2025-12-31 16:29:49","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":508895,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig2.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/e95e8b728909df97923c0b0b.png"},{"id":99217673,"identity":"c8b59c5c-eddb-46d5-a277-c8a6cecdbaf6","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":562211,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig4.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/a43321ac2512bf52c3218edd.png"},{"id":99217677,"identity":"ae409e0d-a5ee-41e2-b78a-f96752ad063c","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1606609,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig5.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/413f49e01bdb74824f6d15e2.png"},{"id":99217670,"identity":"8400fa3e-b1e8-4cda-88dd-444186546485","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":519960,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig6.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/b2338819118206d85a6b5661.png"},{"id":99217663,"identity":"e70dbd25-0ad5-4d38-807e-69dcdf38168e","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"xml","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":159228,"visible":true,"origin":"","legend":"","description":"","filename":"bd834e6d8088452198e6fb93e9c2e6e11structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/5694dc9a37893d886aed1f30.xml"},{"id":99318420,"identity":"9338e0d4-da18-4650-8849-f548f472704c","added_by":"auto","created_at":"2025-12-31 16:33:10","extension":"html","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":172873,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/d00ef08b204fa6a50c98cbd9.html"},{"id":99217653,"identity":"ef7204f0-4360-4d8d-93da-541db6211443","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":733024,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of patient selection and study design. CE-T1WI = contrast-enhanced T1-weighted imaging; CV = cross-validation; DSC = Dice similarity coefficient; DWI = diffusion-weighted imaging; MSI = microsatellite instability; MSS = microsatellite stability; T2WI = T2-weighted imaging.\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/df080a014699174c48395c09.png"},{"id":99318258,"identity":"7e717297-335c-49e0-867c-36d332272455","added_by":"auto","created_at":"2025-12-31 16:32:25","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1874656,"visible":true,"origin":"","legend":"\u003cp\u003eThe workflow of the DL and radiomic study.\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/3070fe4e207f7d7d07e3b63a.png"},{"id":99317442,"identity":"187cdb03-f685-4e18-b07e-80af360ce4df","added_by":"auto","created_at":"2025-12-31 16:30:13","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":232427,"visible":true,"origin":"","legend":"\u003cp\u003eOriginal V-Net network architecture and cascaded framework used in this study.\u003c/p\u003e","description":"","filename":"OnlineFig3.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/00c03dc007d51140a83ec8ca.png"},{"id":99217659,"identity":"0a2530a5-e957-43a7-ac0b-b639d1851d2b","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":2916095,"visible":true,"origin":"","legend":"\u003cp\u003eExamples of manual and automated segmentation in microsatellite stability (MSS) or microsatellite instability (MSI) endometrial cancer. Manually segmented regions are shown in red; automatically segmented regions are shown in purple. (\u003cstrong\u003eTop row\u003c/strong\u003e) Axial T2-weighted imaging (T2WI). (\u003cstrong\u003eMiddle row\u003c/strong\u003e) Axial diffusion-weighted imaging (DWI) (b = 1000 s/mm²). (\u003cstrong\u003eBottom row\u003c/strong\u003e) Axial contrast-enhanced T1-weighted imaging (CE-T1WI).\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/8dab705abbb0ce617b544639.png"},{"id":99318003,"identity":"7b36f1d3-31d8-4214-ab79-54a3c6d04025","added_by":"auto","created_at":"2025-12-31 16:31:08","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":5711058,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation heatmap of selected radiomics features.\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/910055729624969be8922672.png"},{"id":99217666,"identity":"3e14e720-4b47-4114-93e7-cae23bdcb48c","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":3533018,"visible":true,"origin":"","legend":"\u003cp\u003eReceiver operating characteristic (ROC) curves and Decision Curve Analysis (DCA) results for support vector machine (SVM), random forest (RF), and logistic regression (LR) models trained using features extracted from automated and manual segmentation results (reference standard) for predicting microsatellite instability or microsatellite stability status in patients with endometrial cancer. (\u003cstrong\u003ea\u003c/strong\u003e) The mean ROC curve for the internal CV cohort using SVM, RF, and LR. (\u003cstrong\u003eb, c\u003c/strong\u003e) ROC curves for SVM, RF, and LR in the external testing cohorts for manual (\u003cstrong\u003eb\u003c/strong\u003e) and automated (\u003cstrong\u003ec\u003c/strong\u003e) segmentation. \u003cstrong\u003e(d-f\u003c/strong\u003e) DCA results for the internal CV cohort (\u003cstrong\u003ed\u003c/strong\u003e) and external testing cohorts using manual (\u003cstrong\u003ee\u003c/strong\u003e) and automated (\u003cstrong\u003ef\u003c/strong\u003e) segmentation.\u003c/p\u003e","description":"","filename":"Fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/55219db707b3ec07a255e162.png"},{"id":99323901,"identity":"a3d8ff13-3f2d-44aa-8fca-e700cf556699","added_by":"auto","created_at":"2025-12-31 16:46:39","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":18648141,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/decf106b-1cb4-4260-916c-1e9c01b867c2.pdf"},{"id":99217652,"identity":"2d7acd60-1e57-4d31-b6c8-2cb0734a081c","added_by":"auto","created_at":"2025-12-30 09:11:25","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":1037790,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-8106086/v1/7d25ff35b8bc5fb490531bfd.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Deep Learning-Based MRI Segmentation for Non-Invasive Prediction of Microsatellite Instability in Endometrial Cancer: A Multicenter Study","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAmong gynecologic cancers, endometrial cancer (EC) ranks highest in prevalence, with approximately 3% of women facing a lifetime risk of its development. The increasing prevalence of risk factors such as obesity has contributed to increases in both incidence and mortality rates for EC.(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) Historically, risk stratification was based on histomorphological features like grade and histological type, yet even seasoned gynecological pathologists often find these characteristics to be inconsistently reproducible.(\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) This clinical issue highlights the pressing need for dependable, noninvasive biomarkers to enhance preoperative risk evaluation and support personalized treatment strategies.\u003c/p\u003e \u003cp\u003eThe DNA mismatch repair (MMR) system corrects errors that occur during DNA replication, recombination, or even after therapy-induced damage. A deficiency in MMR genes leads to the accumulation of mutations in microsatellite regions, which in turn results in microsatellite instability (MSI).(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) Among all tumor types, EC exhibits the highest proportion of MMR deficiency (MMR-d), with up to 25% to 31% of EC cases classified as MSI.(\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e) In 2013, The Cancer Genome Atlas (TCGA) categorized MSI as a distinct molecular subtype of EC.(\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e) MSI status serves as an important prognostic biomarker and is crucial for treatment planning, as MSI tend to show enhanced responsiveness to programmed death ligand 1 (PD-L1) immunotherapies compared with microsatellite-stable (MSS) cases.(\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e)\u003c/p\u003e \u003cp\u003eCurrently, immunohistochemistry (IHC) analysis of tumor specimens is the standard method for evaluating MSI mutation status, with cases of EC classified as either MSI or MSS.(\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e) While postoperative IHC remains the diagnostic gold standard, current MSI assessment protocols necessitate a 2\u0026ndash;4 week waiting period for pathological specimen processing. This delay is particularly critical for patients with advanced EC, as early initiation of immunotherapy in MSI cases can significantly improve progression-free survival. A preoperative noninvasive approach could enable timely immunotherapy planning, reduce unnecessary surgeries in low-risk cases, and mitigate risks associated with invasive biopsies in frail patients. Radiomic analysis has recently emerged as a promising technique. This method involves extracting high-dimensional and quantitative features from medical images, thereby providing detailed insights into lesion morphology and heterogeneity. Such quantitative metrics have shown potential as noninvasive biomarkers for staging diseases, evaluating treatment response, and assessing prognosis.(\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e) A recent systematic review and meta-analysis found that MRI-based radiomic analysis performed preoperatively in EC patients can reliably predict tumor grade, myometrial invasion (MI), lymph node metastasis (LNM), and lymphovascular space invasion (LVSI).(\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e) Studies have shown that radiomic features may correlate with tumor molecular biomarkers and that these features can be used to identify the molecular subtype of the disease preoperatively.(\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e)\u003c/p\u003e \u003cp\u003eHowever, traditional manual segmentation is not only labor-intensive but also time-consuming, particularly when large datasets are involved. Consequently, deep learning\u0026ndash;based automated segmentation techniques have garnered significant attention as a means to overcome these limitations.(\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e) In particular, developing an automated segmentation algorithm using multicenter data is critical for addressing inter-institutional variability and improving model robustness.\u003c/p\u003e \u003cp\u003eIn this study, we aimed to develop and validate a multiparametric MRI-based deep learning automated segmentation model for patients with EC. Furthermore, by comparing automated segmentation's performance with that of manual segmentation, we explored the feasibility of integrating deep learning techniques and radiomic analysis into clinical practice for the prediction of MSI molecular subtypes in EC.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003ePatient Selection\u003c/h2\u003e \u003cp\u003e This retrospective study was approved by the Institutional Review Boards of Peking University People's Hospital (Approval No. 2025PHB057-001; Date: 2025-01-25). Given the retrospective nature of this study, the requirement for informed consent was waived. The study included 593 patients with EC who underwent preoperative pelvic MRI scans at Peking University People's Hospital (Center 1) between May 2007 and September 2023, 181 patients from Second Hospital of Shanxi Medical University (Center 2) between May 2017 and September 2022, and 93 patients from The Third Affiliated Hospital of Zunyi Medical University (Center 3) between May 2019 and November 2022. Patients from Center 1 served as the internal cross-validation (internal CV) set for training machine learning models to identify MSI status in patients with EC. Patients from Centers II and III comprised the external testing cohort. Detailed inclusion and exclusion criteria, along with a comprehensive study flowchart, are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eDetermination of MSI Status Using IHC Analysis\u003c/h3\u003e\n\u003cp\u003eAll patients underwent a total hysterectomy, bilateral salpingo-hysterectomy, and pelvic lymph node sampling. The expression of four MMR gene proteins (MLH1, MSH2, MSH6, and PMS2) were determined by IHC, which served as the reference standard to establish MSI status. Two pathologists (each with \u0026gt;\u0026thinsp;10 years of experience) reviewed and confirmed the results. Patients were subsequently classified into two groups: those with all four MMR proteins expressed (designated as MSS) and those missing one or more MMR proteins (designated as MSI).\u003c/p\u003e\n\u003ch3\u003eMRI Acquisition\u003c/h3\u003e\n\u003cp\u003eEach patient received a pelvic MRI scan encompassing the entire tumor region. The imaging protocol included several sequences: Axial T2-weighted imaging (T2WI); Axial diffusion-weighted imaging (DWI) with b values of 0 and 1000 s/mm\u0026sup2;; and Axial contrast-enhanced T1-weighted imaging (CE-T1WI) performed with gadodiamide at a dosage of 0.1 mmol/kg. All patients underwent CE-T1WI scans, and to minimize motion artifacts, 20 mg of hyoscine butylbromide was administered intramuscularly prior to imaging in all patients. The detailed MRI acquisition parameters are shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMRI acquisition parameters\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026times;\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameter\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eScanner\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSequence\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTR/TE (ms)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFOV (mm)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSlice Thickness (mm)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMatrix\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e \u003cp\u003eCenter1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003ePhilips ingenia 3.0\u0026nbsp;T\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eT2WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3739/90.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e360\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e480\u0026thinsp;\u0026times;\u0026thinsp;348\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDWI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4900/58.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e380\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e128\u0026times;204\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCE-T1WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.9/1.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e320\u0026times;208\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eGE discovery 750 3.0\u0026nbsp;T\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eT2WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3825/89.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e380\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e384\u0026thinsp;\u0026times;\u0026thinsp;224\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDWI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4875/58.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e380\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e160\u0026thinsp;\u0026times;\u0026thinsp;160\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCE-T1WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.0/1.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e288\u0026thinsp;\u0026times;\u0026thinsp;224\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eCenter2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eGE discovery 750\u0026nbsp;W 3.0\u0026nbsp;T\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eT2WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e5526/71.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e380\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e288\u0026thinsp;\u0026times;\u0026thinsp;288\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDWI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6500/68.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e380\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e128\u0026thinsp;\u0026times;\u0026thinsp;130\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCE-T1WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6.2/2.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e256\u0026thinsp;\u0026times;\u0026thinsp;256\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eSiemes uMR588 3.0\u0026nbsp;T\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eT2WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4360/88.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e340\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e320\u0026times;256\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCenter3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDWI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e5600/60.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e340\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e128\u0026times;78\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCE-T1WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.8/1.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026times;\" colname=\"c7\"\u003e \u003cp\u003e320\u0026thinsp;\u0026times;\u0026thinsp;203\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e\n\u003ch3\u003ePreprocessing and Segmentation\u003c/h3\u003e\n\u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows the workflow of the DL and radiomic study. To reduce bias and eliminate intensity differences from various imaging protocols, all raw MR images were preprocessed before extracting features. Specifically, volumes of interest (VOIs) were normalized using z-score gray-level standardization and resampled to a uniform resolution (1 mm \u0026times; 1 mm \u0026times; 1 mm) via B-spline interpolation. The intensity range was limited to \u0026micro;\u0026thinsp;\u0026plusmn;\u0026thinsp;3σ to lessen the impact of contrast and brightness variations. (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e)\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eRadiologists with at least six years of pelvic MRI experience from each institution manually outlined the three-dimensional tumor VOIs on axial oblique slices from T2WI, CE-T1WI, and DWI sequences. These contours were then reviewed by a senior radiologist with over 20 years of experience. To evaluate the consistency of manual segmentation, two radiologists from different institutions jointly and randomly selected T1WI, T2WI, and DWI images from 150 patients for repeated segmentation. Although the reviewer knew the patients had endometrial carcinoma, they were blinded to the detailed clinical information and histopathological findings.\u003c/p\u003e\n\u003ch3\u003eDeep Learning–Based Automated Segmentation\u003c/h3\u003e\n\u003cp\u003eThe automated segmentation framework was built on the V-Net architecture, which is effective for volumetric medical image segmentation and uses skip connections to improve information flow between layers. Given the high resolution, large field of view (FOV) of pelvic MRI, and challenges from multi-center data differences, a cascaded lightweight V-Net structure was used. The first V-Net processed down-sampled, large FOV image blocks to roughly localize the tumor, while the second V-Net refined the segmentation using high-resolution data. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e shows the original V-Net alongside its cascaded framework for clarity.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTraining and testing of the V-Net were conducted in two steps: internal CV and external testing. For internal CV, data from 593 patients at Center 1 were used with a 5-fold cross-validation scheme to train, test, and evaluate the segmentation method. External testing was carried out on data from 274patients at Center 2 and Center 3 using the best model from internal CV.\u003c/p\u003e \u003cp\u003eThe network underwent training for 1,200 iterations. A learning rate of 10⁻⁵ and a batch size of 16 were used, with cross-entropy employed as the loss function. Optimization was achieved using the Adam optimizer (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). All experiments were conducted on the uAI Research Portal (version 20240730) platform, and the V-Net network was implemented using the Pytorch framework.(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e)\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eFeature Extraction and Selection\u003c/h2\u003e \u003cp\u003eRadiomic features from VOIs across different MRI sequences were automatically extracted using the PyRadiomics software package (version 3.0.1). The extracted features were categorized as first-order (n\u0026thinsp;=\u0026thinsp;18), shape-based (n\u0026thinsp;=\u0026thinsp;14), and higher-order (n\u0026thinsp;=\u0026thinsp;75), all standardized using z scores to remove scale differences.\u003c/p\u003e \u003cp\u003eA three-step feature selection process was used to identify the most predictive features. First, data from 100 randomly selected patients were reviewed by two radiologists from different centers (each with at least six years of experience), and the intraclass correlation coefficient (ICC) analysis was applied to retain features with ICC\u0026thinsp;\u0026gt;\u0026thinsp;0.85. Next, Spearman\u0026rsquo;s rank correlation test was used to remove redundant features (correlation coefficient\u0026thinsp;\u0026ge;\u0026thinsp;0.9).Finally,the least absolute shrinkage and selection operator (LASSO) method\u0026mdash;with 5-fold CV and an optimized penalty parameter\u0026mdash;was applied to keep features with nonzero coefficients that were relevant to MSI status in EC patients.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eClassification Model Construction\u003c/h3\u003e\n\u003cp\u003eDataset 1 from Center 1, containing 593 patients (497 with MSS and 96 with MSI), was used to test the feasibility of radiomic analysis for predicting MSI status in EC. Model training and feature selection were performed using 5-fold CV, with the adaptive synthetic (ADASYN) sampling technique applied in each training fold to address class imbalance.(\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e) Based on the selected radiomic features, three classifiers\u0026mdash;support vector machine (SVM), random forest (RF), and logistic regression (LR)\u0026mdash;were developed using the scikit-learn library in Python (version 3.6). These machine learning algorithms were optimized via grid search and 5-fold CV. External testing, conducted on data from Centers 2 and 3 (totaling 230 MSS and 44 MSI cases), compared the classification performance of models based on manual versus automated segmentation, using the feature sets derived from internal CV.\u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eFor the automated segmentation model, performance metrics including the Dice similarity coefficient (DSC), sensitivity, and specificity were calculated for the entire tumor, tumor core, and enhanced tumor regions. The DSC measured the overlap between the automated segmentation and the ground truth relative to the total area. Recall was defined as the ratio of correctly detected tumor voxels to total tumor voxels, and precision as the ratio of correctly detected tumor voxels to all voxels identified by the automated segmentation.\u003c/p\u003e \u003cp\u003eFor radiomics classification models, receiver operating characteristic (ROC) curves were generated to assess performance, with optimal cutoff points determined using the Youden index on the training cohort and then applied to two independent validation cohorts. Model performance was evaluated using metrics such as AUC, accuracy, sensitivity, and specificity, and the DeLong test was used to compare AUCs among the three models. All statistical tests were two-tailed, with a P value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 considered statistically significant.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eClinical Characteristics\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e summarizes the clinical characteristics of patients. In the internal CV cohort, the prevalence of MSI was 16.19% (96 out of 593 patients), while in the external testing cohort it was 16.06% (44 out of 274 patients), which suggests a comparable distribution (P\u0026thinsp;=\u0026thinsp;0.942). Notably, in the internal CV cohort, lymphovascular space invasion (LVSI) differed markedly, while tumor grading was significantly distinct in the external testing group.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eClinicopathologic Characteristics of Patients With MSI and MSS Tumors in the Internal CV and External Validation Cohorts\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eClinicopathologic Characteristic\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eSubgroup\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c5\" namest=\"c3\"\u003e \u003cp\u003eInternal CV Cohort\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c8\" namest=\"c6\"\u003e \u003cp\u003eExternal Validation Cohort\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMSI (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;96)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMSS (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;497)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e Value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMSI (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;44)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMSS (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;230)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e Value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean age\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (y)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e58.4\u0026thinsp;\u0026plusmn;\u0026thinsp;6.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e57.2\u0026thinsp;\u0026plusmn;\u0026thinsp;9.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.372\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e56.9\u0026thinsp;\u0026plusmn;\u0026thinsp;8.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e58.9\u0026thinsp;\u0026plusmn;\u0026thinsp;7.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.266\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eHistologic Grade, \u003cem\u003en\u003c/em\u003e (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e33 (34.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e202 (41.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.451\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e12 (27.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e103 (44.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.043\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMedium\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e39 (40.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e187 (37.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e23 (52.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e77 (33.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e24 (25.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e104 (21.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e9 (20.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e50 (21.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eFIGO stage, \u003cem\u003en\u003c/em\u003e (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e76(79.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e418(84.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.227\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e34(77.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e187(81.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.334\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eII\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e8(7.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e34(6.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4(9.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e25(10.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIII\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11(11.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e36(7.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6(13.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e17(7.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1(1.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5(1.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0(0.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1(0.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eDeep Myometrial invasion, \u003cem\u003en\u003c/em\u003e (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePresent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20 (20.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e107 (21.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.986\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e14 (31.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e58 (25.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.469\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAbsent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e74 (77.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e390 (78.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e30 (68.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e172 (74.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLymph vascular space invasion, \u003cem\u003en\u003c/em\u003e (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePresent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e19 (19.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e54 (10.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.023\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e11 (25.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e39 (17.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.293\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAbsent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e77 (80.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e443 (89.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e33 (75.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e191 (83.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLymph node metastasis, \u003cem\u003en\u003c/em\u003e (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePresent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10 (10.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e35 (7.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.351\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e7 (15.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e21 (9.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.316\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAbsent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e86 (89.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e462 (93.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e37 (84.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e201 (90.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMaximum tumor diameter, \u003cem\u003en\u003c/em\u003e (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;5.0 cm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e59 (61.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e335 (67.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.312\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e25 (56.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e159 (76.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.156\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;5.0 cm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e37 (38.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e162 (32.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e19 (43.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e71 (30.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"8\"\u003eCV\u0026thinsp;=\u0026thinsp;cross-validation; MSI\u0026thinsp;=\u0026thinsp;microsatellite instability; MSS\u0026thinsp;=\u0026thinsp;microsatellite stability.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eInter-observer Agreement in Manual Segmentation\u003c/h2\u003e \u003cp\u003eHigh inter-observer consistency was demonstrated by DSC of 89.4% \u0026plusmn; 16.6% (T2WI), 91.2% \u0026plusmn; 5.4% (DWI), and 92.2% \u0026plusmn; 7.9% (CE-T1WI) between the two radiologists.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eDeep Learning\u0026ndash;Based Automated Segmentation\u003c/h2\u003e \u003cp\u003eIn the external testing set, the mean DSC\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation values were 76.2% \u0026plusmn; 17.3% (95% confidence interval [CI], 72.8%-78.5%) for T2WI, 85.8% \u0026plusmn; 13.3% (95% CI, 83.8%-87.3%) for DWI, and 80.5% \u0026plusmn; 14.6% (95% CI, 76.2%-81.7%) for CE-TIWI (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e illustrates typical cases of lesion segmentation for T2WI, DWI, and CE-TIWI images.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance of Automated Segmentation in the External Validation Cohort\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eSequence\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eDice Similarity Coefficient\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;Standard Deviation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMEDIAN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;Standard Deviation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;Standard Deviation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT2WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e76.2% \u0026plusmn; 17.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e77.9%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e85.4% \u0026plusmn; 17.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e71.9% \u0026plusmn; 6.3%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDWI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e85.8% \u0026plusmn; 13.3%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e87.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e83.3% \u0026plusmn; 14.2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e78.6% \u0026plusmn; 8.1%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCE-T1WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e \u003cp\u003e80.5% \u0026plusmn; 14.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e84.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e82.93% \u0026plusmn; 16.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e73.4% \u0026plusmn; 15.0%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003eCE-T1WI\u0026thinsp;=\u0026thinsp;contrast-enhanced T1-weighted imaging; DWI\u0026thinsp;=\u0026thinsp;diffusion-weighted imaging; T2WI\u0026thinsp;=\u0026thinsp;T2-weighted imaging.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eRadiomics Analysis\u003c/h2\u003e \u003cp\u003eAmong the 5,652 radiomic features extracted from T2WI, DWI, and CE-TIWI, features with ICC values\u0026thinsp;\u0026lt;\u0026thinsp;0.8 were first excluded, reducing the number of radiomic features to 1,347. Subsequently, 409 features were retained through Spearman correlation testing (correlation coefficient\u0026thinsp;\u0026gt;\u0026thinsp;0.9), followed by univariate logistic regression selection. Ultimately, 29 features were selected through LASSO regression, including 7 from T2WI, 8 from DWI, and 14 from CE-TIWI, to establish the radiomic model. Analysis of the selected radiomic features revealed no significant correlations among them, as illustrated by Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn the internal CV cohort, the mean AUCs obtained through 5-fold CV using the automated segmentation model were 0.876 (95% CI, 0.842\u0026ndash;0.906) with SVM, 0.937 (95% CI, 0.911\u0026ndash;0.960) with RF, and 0.812 (95% CI, 0.765\u0026ndash;0.853) with LR. For the external testing cohort, the AUCs achieved by the automated segmentation model were 0.876 (95% CI, 0.814\u0026ndash;0.927) with SVM, 0.867 (95% CI, 0.780\u0026ndash;0.942) with RF, and 0.800 (95% CI, 0.701\u0026ndash;0.887) with LR, and the AUCs achieved by the manual segmentation model were 0.915 (95% CI, 0.853\u0026ndash;0.962) with SVM, 0.890 (95% CI, 0.829\u0026ndash;0.946) with RF, and 0.821 (95% CI, 0.751\u0026ndash;0.883) with LR (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eClassification Performance of Different Classifiers Based on Different Segmentation Methods for Predicting the MSI Status of the External Validation Set\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCohort\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClassifier\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSegmentation Method\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eAUC (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003eS\u003c/em\u003e\u003c/sub\u003e Value\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003eC\u003c/em\u003e\u003c/sub\u003e Value\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eInternal CV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eManual\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.897\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.926\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.937 (0.911\u0026ndash;0.960)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e3.34E-11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eManual\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.694\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.865\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.541\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.812 (0.765\u0026ndash;0.853)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e3.04E-25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eManual\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.804\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.792\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.807\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.876 (0.842\u0026ndash;0.906)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.183\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e \u003cp\u003eExternal validation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAutomated\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.874\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.794\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.882\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.867 (0.780\u0026ndash;0.942)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.167\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e8.10E-29\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eManual\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.928\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.568\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.989\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.915 (0.853\u0026ndash;0.962)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e2.68E-26\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAutomated\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.767\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.833\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.800 (0.701\u0026ndash;0.887)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.294\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e6.55E-17\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eManual\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.764\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.886\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.627\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.821 (0.751\u0026ndash;0.883)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e1.27E-16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAutomated\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.831\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.778\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.876 (0.814\u0026ndash;0.927)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.077\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.844\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eManual\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.798\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.864\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.787\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.890 (0.829\u0026ndash;0.946)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.561\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"9\"\u003eAUC\u0026thinsp;=\u0026thinsp;area under the curve; CI\u0026thinsp;=\u0026thinsp;confidence interval; CV\u0026thinsp;=\u0026thinsp;cross-validation; LR\u0026thinsp;=\u0026thinsp;logistic regression; RF\u0026thinsp;=\u0026thinsp;random forest; SVM\u0026thinsp;=\u0026thinsp;support vector machine.\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"9\"\u003e\u003csup\u003ea\u003c/sup\u003eSignificance level for comparisons between different segmentation methods (automated vs manual) with the same classifiers.\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"9\"\u003e\u003csup\u003eb\u003c/sup\u003eSignificance level for comparisons between RF and LR, LR and SVM, and RF and SVM in each cohort.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIn the internal CV cohort, RF was significantly superior to SVM and LR. However, in the external testing set, there was no significant difference between RF and SVM with manual segmentation (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.561) or with automated segmentation (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.844), and both were significantly better than LR. For models based on the same classifiers (SVM, RF, and LR) but different segmentation methods, the differences in AUCs were not significant (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.077\u0026ndash;0.294). Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e presents the ROC curves and DCA results for the classification models.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this investigation, we employed an automated segmentation approach based on a V-Net cascading model to delineate lesions on multimodal MR images of EC, and we then developed a radiomic classification framework to evaluate MSI/MSS status. The segmentation demonstrated strong performance, achieving DSCs in the external testing cohort of 85.8% \u0026plusmn; 13.3% with DWI, 76.2% \u0026plusmn; 17.3% with T2WI, and 80.5% \u0026plusmn; 14.6% with CE-T1WI. Radiomics analysis using automated segmentation exhibited performance comparable to manual segmentation, with consistent accuracy across multi-center datasets. These results underscore the efficacy and generalizability of our automated approach despite inherent heterogeneity in multi-Center 1maging data.\u003c/p\u003e \u003cp\u003eThe inherent complexity and pronounced heterogeneity characteristic of EC imaging data pose significant challenges to accurate tumor segmentation, particularly in multicenter settings.(\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e) Current research in this field demonstrates notable limitations: Hodneland et al.'s pioneering study(\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e) employed contrast-enhanced T1-weighted imaging (CE-T1WI) as the sole input for machine learning segmentation, achieving median DSC of 0.84 and 0.77 across two readers. However, this single-center investigation (n\u0026thinsp;=\u0026thinsp;34) restricted its analysis to CE-T1WI, thereby constraining the generalizability of its findings. Kurata et al.(\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e) advanced the methodology through multisequence integration (T2WI, DWI, and ADC maps) combined with ensemble learning, attaining a mean DSC of 0.806 with sensitivity and positive predictive value (PPV) of 0.816 and 0.834, respectively. Nevertheless, the single-institution origin of these data raises concerns regarding model robustness against protocol variability. Our cascaded V-Net addresses multicenter variability through coarse-to-fine localization, mimicking radiologists\u0026rsquo; workflow. The proposed cascaded V-Net model demonstrated high segmentation consistency across multi-center MRI data, external validation across three institutions with heterogeneous MRI protocols demonstrates unprecedented generalizability. Yet segmentation failures still occurred in certain cases (Supplementary Material Fig.\u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). First, the low contrast between tumors and adjacent tissues on T2WI sequences may hinder precise boundary delineation, particularly in cases with infiltrative tumors or coexisting adenomyosis. Second, variations in image quality across different scanners (e.g., resolution, signal-to-noise ratio, or artifact levels) could destabilize the model. For instance, low-field MRI images in the external test set exhibited local detail loss due to noise interference, thereby compromising segmentation accuracy. Furthermore, segmentation errors for small lesions disproportionately reduced DSC scores, as the limited voxels in minute tumors amplify the impact of minor segmentation inaccuracies. Finally, interference from adjacent pathologies (e.g., concurrent uterine fibroids) might lead to misidentification of tumor regions, especially when malignant and benign lesions exhibit overlapping imaging features. Despite these limitations, the observed failure cases remained sporadic, and the model's overall performance in external validation proved comparable to manual segmentation, thereby confirming its clinical utility.\u003c/p\u003e \u003cp\u003eComparative analysis with existing literature underscores our model's translational superiority. Lin et al.(\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e) d developed a model to predict MSI status in endometrial cancer using T2WI and CE-T1WI from 296 patients, the classification models achieved AUC values of 0.752 in the training cohort and 0.723 in the validation cohort. Jia et al.(\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e) implemented a multiparametric MRI radiomics framework (T2WI, CE-T1WI, ADC) using SVM, attaining exceptional AUCs of 0.905 (training), 0.875 (internal test), and 0.862 (external test). In our investigation, both automated and manual segmentation-based classifiers demonstrated robust MSI classification performance. While RF exhibited superior training performance (AUC\u0026thinsp;=\u0026thinsp;0.85 vs. SVM\u0026thinsp;=\u0026thinsp;0.80), their external test accuracies showed no statistically significant difference (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05). This observation aligns with established machine learning principles - RF's ensemble methodology reduces feature/sample dependency to prevent overfitting, whereas SVM optimizes hyperplane determination for minimal misclassification. Both algorithms have been extensively validated in oncological applications including genomic profiling and survival prediction.(\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e) The comparatively lower LR performance (AUC\u0026thinsp;=\u0026thinsp;0.72) likely reflects its limited capacity to model complex radiomic feature interactions, suggesting LR\u0026rsquo;s linear decision boundaries likely failed to capture nonlinear interactions among textural features, whereas ensemble methods like RF inherently model complex hierarchies.(\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e)\u003c/p\u003e \u003cp\u003eClass imbalance in machine learning model development can significantly compromise classifier performance. To address this challenge, our study employed the Adaptive Synthetic Sampling (ADASYN) technique, which dynamically generates synthetic samples of the minority class within each cross-validation fold of the training set to enhance model learning. Crucially, the application of ADASYN strictly adheres to a core principle in machine learning: oversampling is exclusively applied during the training phase, while validation and external test sets retain their original data distributions. Specifically, in the 5-fold cross-validation framework, ADASYN was independently implemented on the training subset of each fold, whereas validation subsets remained untouched by synthetic sample generation. This design effectively prevents data leakage and ensures unbiased model evaluation. Furthermore, we mitigated potential noise sensitivity induced by oversampling through LASSO regression for discriminative feature selection, coupled with regularization strategies such as kernel function constraints in SVM and out-of-bag error evaluation in RF. Notably, the model demonstrated consistent classification performance between internal validation (AUC\u0026thinsp;=\u0026thinsp;0.876) and external testing cohorts, with no significant difference compared to manual segmentation (P\u0026thinsp;\u0026gt;\u0026thinsp;0.05). This alignment underscores the method\u0026rsquo;s robust generalizability in real-world scenarios. Collectively, the synergistic integration of rigorous data isolation protocols and regularization techniques successfully balances the challenges of class imbalance and overfitting risks, establishing a reliable technical framework for radiomics-based molecular subtyping research.\u003c/p\u003e \u003cp\u003eNotably, more than half of the radiomic features used in our approach were textural (20/29), with GLSZM being the most prevalent (9/29), suggesting that GLSZM may be more closely associated with MMR genes. Radiomic feature analysis has demonstrated significant differences in GLCM entropy values between Lynch syndrome patients (a hereditary tumor syndrome caused by germline mutations in MMR genes) and the control cohort.(\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e).As an extension of the GLCM methodology, the GLSZM analyzes the distribution characteristics of spatially connected pixel clusters, with its regional entropy parameter effectively quantifying the uncertainty in both zone size and grayscale value distribution.(\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). The GLSZM features were higher in the MSI group compared to the MSS group overall (Supplementary Material Fig.S2). This imaging-based analytical approach enables precise characterization of tumor microstructural heterogeneity, where elevated entropy values suggest increased randomness in textural patterns. The observed feature elevation in MSI tumors aligns with their inherently disordered growth patterns. Notably, previous clinical studies have confirmed that endometrial carcinomas with MSI positivity are typically associated with higher tumor grades, deeper myometrial invasion, and increased lymph node metastasis rates, consistent with the more heterogeneous microenvironment reflected by these quantitative imaging features.(\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e) These pathological characteristics correlate with histopathological evidence of enhanced intratumoral architectural complexity observed in mismatch repair-deficient lesions. Importantly, radiomics-based spatial heterogeneity analysis provides a non-invasive means to characterize these microstructural spatial variations, demonstrating remarkable consistency with histomorphological findings.(\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e)\u003c/p\u003e \u003cp\u003eThe determination of MSI status in EC carries substantial clinical relevance, as it facilitates Lynch syndrome screening and informs prognosis and therapeutic strategies, particularly regarding immunotherapy response.(\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e) Conventional tissue-based methods, such as PCR and IHC, are constrained by their invasiveness, cost, time requirements, and dependence on sufficient tissue samples, which may delay clinical decisions. Our radiomics framework offers a rapid, non-invasive preoperative alternative that enables molecular subtyping and enhances risk stratification accuracy, thereby addressing a critical unmet clinical need. Furthermore, there were no significant differences between the segmentation strategies (manual vs. automated) across the three classifiers. These findings suggest the automated segmentation method performs robustly in radiomic analysis. Although the underlying pathological mechanisms remain unclear, our results imply that MSI tumors may exhibit greater histological heterogeneity than MSS tumors\u0026mdash;a difference that radiomic classifiers appear capable of capturing.\u003c/p\u003e \u003cp\u003eThis study had several limitations. First, the number of positive samples was relatively small. Although oversampling was used to mitigate the data distribution imbalance to some extent, a larger and more uniformly distributed dataset, especially from different centers, would facilitate the training of more robust segmentation and radiomic models. Nevertheless, the outcomes were statistically reliable given the current sample size. Although IHC is a reliable and cost-effective technique for assessing MMR proteins and determining MSI status, PCR is still considered the gold standard for MSI evaluation. A previous study found that the consistency between PCR and IHC in determining MSI status is nearly 92%.(\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e) Finally, this study primarily focused on evaluating MSI status as a single indicator of EC. Tumor occurrence, development, and treatment response are influenced by various biological characteristics; therefore, further optimization of the experimental design is needed to conduct comprehensive analyses of multiple biological characteristics.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn this multicenter study, we demonstrated that a deep learning-based automated segmentation model achieved performance comparable to manual segmentation, and when combined with radiomic analysis, provided a robust, non-invasive method for the preoperative prediction of MSI status in endometrial cancer By integrating our framework into preoperative MRI workflows, clinicians could stratify EC patients into MSI/MSS subgroups before surgery. Preoperative identification of MSI status enables early enrollment in neoadjuvant immunotherapy trials, particularly for patients with high surgical risk or those seeking fertility preservation. Conversely, MSS patients may avoid costly immunotherapy with limited benefit. Future efforts should prioritize multi-center validation, integration of multi-modal biomarkers, and prospective trials to evaluate therapeutic impact. Enhancing model interpretability and addressing ethical challenges will be critical for real-world translation. These advances could streamline precision oncology workflows in EC management.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eADASYN\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAdaptive Synthetic Sampling approach for imbalanced learning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eArea Under the Curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eCE-T1WI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eContrast-Enhanced T1-weighted Imaging\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eCV\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCross - Validation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eDSC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDice Similarity Coefficient\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eDWI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDiffusion-Weighted Imaging\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eEC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eEndometrial Cancer\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eICC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eIntraclass Correlation Coefficient\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eIHC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eImmunohistochemistry\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eLASSO\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLeast Absolute Shrinkage and Selection Operator\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eLR\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eLNM\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLymph Node Metastasis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eLVSI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLymphovascular Space Invasion\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eMI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMyometrial Invasion\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eMMR\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDNA Mismatch Repair\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eMSS\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMicrosatellite Stability\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eMSI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMicrosatellite Instability\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eMRI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMagnetic Resonance Imaging\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003ePPV\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePositive Predictive Value\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eRF\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eROC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eReceiver Operating Characteristic\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eSVM\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSupport Vector Machine\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eT2WI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eT2-weighted Imaging\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eVOI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eVolume of Interest\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":" \u003cp\u003e \u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e \u003cp\u003e This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Institutional Review Boards of Peking University People's Hospital (Approval No. 2025PHB057-001; Date: 2025-01-25). Due to the retrospective nature of this study involving the analysis of existing data, the requirement for informed consent was waived by the Institutional Review Boards of Peking University People's Hospital.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e \u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis work was supported by the National Key Research and Development Program of China (Grant Number: 2022YFC2704405).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eKe Wang and Xiaoli Song were responsible for data collection and analysis and were the main contributors to writing the manuscript. Xinyi Gou and Xuqing Lin analyzed the images. Hongjian Luo examined the patients and provided images. Yi Wang and Nan Hong analyzed the data. Rong Zhou and Jin Cheng were responsible for the study design, and the final approval of the version to be published. All authors read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e \u003cp\u003eNot applicable..\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets generated and/or analysed during the current study are not publicly available due to privacy or ethical restrictions, but are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eCrosbie EJ, Kitson SJ, McAlpine JN, Mukhopadhyay A, Powell ME, Singh N. Endometrial cancer. Lancet. 2022;399(10333):1412\u0026ndash;28.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoribata Y, Kurata Y, Nishio M, Kido A, Otani S, Himoto Y, et al. Automatic segmentation of bladder cancer on MRI using a convolutional neural network and reproducibility of radiomics features: a two-center study. Sci Rep. 2023;13:628.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeyer LA, Broaddus RR, Lu KH. Endometrial cancer and Lynch syndrome: clinical and pathologic considerations. Cancer Control J Moffitt Cancer Cent. 2009;16(1):14\u0026ndash;22.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRyan NaJ, Glaire MA, Blake D, Cabrera-Dandy M, Evans DG, Crosbie EJ. The proportion of endometrial cancers associated with Lynch syndrome: a systematic review of the literature and meta-analysis. Genet Med. 2019;21(10):2167.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCancer Genome Atlas Research Network, Kandoth C, Schultz N, Cherniack AD, Akbani R, Liu Y, et al. Integrated genomic characterization of endometrial carcinoma. Nature. 2013;497(7447):67\u0026ndash;73.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArafa M, Shebl AM, Salama A, ElZahaf E, Ashamallah SA, Foda AA, et al. Correlation of PD-L1 immunohistochemical expression with microsatellite instability and p53 status in endometrial carcinoma. Eur J Obstet Gynecol Reprod Biol X. 2022;16:100172.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRibeiro-Santos P, Martins Vieira C, Viana Veloso GG, Vieira Giannecchini G, Parenza Arenhardt M, M\u0026uuml;ller Gomes L et al. Tailoring Endometrial Cancer Treatment Based on Molecular Pathology: Current Status and Possible Impacts on Systemic and Local Treatment. Int J Mol Sci 2024 July 15;25(14):7742.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016;278(2):563\u0026ndash;77.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBera K, Braman N, Gupta A, Velcheti V, Madabhushi A. Predicting cancer outcomes with radiomics and artificial intelligence in radiology. Nat Rev Clin Oncol. 2022;19(2):132\u0026ndash;46.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDi Donato V, Kontopantelis E, Cuccu I, Sgamba L, Golia D\u0026rsquo;Aug\u0026egrave; T, Pernazza A et al. Magnetic resonance imaging-radiomics in endometrial cancer: a systematic review and meta-analysis. Int J Gynecol Cancer Off J Int Gynecol Cancer Soc 2023 July 3;33(7):1070\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang G, Du S, Gao S, Guo L, Zhao R, Bian X, et al. Molecular subtypes of breast cancer identified by dynamically enhanced MRI radiomics: the delayed phase cannot be ignored. Insights Imaging. 2024;15(1):127.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGao R, Pang J, Lin P, Wen R, Wen D, Liang Y et al. Identification of clear cell renal cell carcinoma subtypes by integrating radiomics and transcriptomics. Heliyon 2024 June 15;10(11):e31816.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSoneji ND, Bharwani N, Ferri A, Stewart V, Rockall A. Pre-operative MRI staging of endometrial cancer in a multicentre cancer network: can we match single centre study results? Eur Radiol. 2018;28(11):4725\u0026ndash;34.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeyer HJ, Renatus K, H\u0026ouml;hn AK, Hamerla G, Schopow N, Fakler J, et al. Texture analysis parameters derived from T1-and T2-weighted magnetic resonance images can reflect Ki67 index in soft tissue sarcoma. Surg Oncol. 2019 Sept;30:92\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArcos-Garc\u0026iacute;a \u0026Aacute;, \u0026Aacute;lvarez-Garc\u0026iacute;a JA, Soria-Morillo LM. Deep neural network for traffic sign recognition systems: An analysis of spatial transformers and stochastic optimisation methods. Neural Netw Off J Int Neural Netw Soc. 2018;99:158\u0026ndash;65.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu J, Xia Y, Wang X, Wei Y, Liu A, Innanje A, et al. uRP: An integrated research platform for one-stop analysis of medical images. Front Radiol. 2023;3:1153784.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHe H, Bai Y, Garcia EA, Li S. ADASYN: Adaptive synthetic sampling approach for imbalanced learning. In: 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence) [Internet]. 2008 [cited 2024 Dec 15]. pp. 1322\u0026ndash;8. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://ieeexplore.ieee.org/document/4633969\u003c/span\u003e\u003cspan address=\"https://ieeexplore.ieee.org/document/4633969\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTian S, Liu A, Zhu W, Li Y, Chen L, Chen A, et al. Difference in Diffusion-Weighted Magnetic Resonance Imaging and Diffusion Tensor Imaging Parameters Between Endometrioid Endometrial Adenocarcinoma and Uterine Serous Adenocarcinoma: A Retrospective Study. Int J Gynecol Cancer Off J Int Gynecol Cancer Soc. 2017;27(8):1708\u0026ndash;13.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHodneland E, Dybvik JA, Wagner-Larsen KS, Šolt\u0026eacute;szov\u0026aacute; V, Munthe-Kaas AZ, Fasmer KE, et al. Automated segmentation of endometrial cancer on MR images using deep learning. Sci Rep. 2021;11:179.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKurata Y, Nishio M, Moribata Y, Kido A, Himoto Y, Otani S, et al. Automatic segmentation of uterine endometrial cancer on multi-sequence MRI using a convolutional neural network. Sci Rep. 2021 July;14:11:14440.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin Z, Wang T, Li H, Xiao M, Ma X, Gu Y, et al. Magnetic resonance-based radiomics nomogram for predicting microsatellite instability status in endometrial cancer. Quant Imaging Med Surg. 2023;13(1):108\u0026ndash;20.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJia Y, Hou L, Zhao J, Ren J, Li D, Li H, et al. Radiomics analysis of multiparametric MRI for preoperative prediction of microsatellite instability status in endometrial cancer: a dual-center study. Front Oncol. 2024;14:1333020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu J, Hicks C. Breast Cancer Type Classification Using Machine Learning. J Pers Med. 2021;11(2):61.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChu LC, Park S, Kawamoto S, Fouladi DF, Shayesteh S, Zinreich ES, et al. Utility of CT Radiomics Features in Differentiation of Pancreatic Ductal Adenocarcinoma From Normal Pancreatic Tissue. AJR Am J Roentgenol. 2019;213(2):349\u0026ndash;57.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen D, Afzal N, Sohn S, Habermann EB, Naessens JM, Larson DW, et al. Postoperative bleeding risk prediction for patients undergoing colorectal surgery. Surgery. 2018;164(6):1209\u0026ndash;16.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang X, Wu K, Li X, Jin J, Yu Y, Sun H. Additional Value of PET/CT-Based Radiomics to Metabolic Parameters in Diagnosing Lynch Syndrome and Predicting PD1 Expression in Endometrial Carcinoma. Front Oncol [Internet]. 2021 May 12 [cited 2024 Dec 30];11. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.frontiersin.org/journals/oncology/articles/\u003c/span\u003e\u003cspan address=\"https://www.frontiersin.org/journals/oncology/articles/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fonc.2021.595430/full\u003c/span\u003e\u003cspan address=\"10.3389/fonc.2021.595430/full\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan Griethuysen JJ, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res. 2017;77(21):e104\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBilbao C, Lara PC, Ram\u0026iacute;rez R, Henr\u0026iacute;quez-Hern\u0026aacute;ndez LA, Rodr\u0026iacute;guez G, Falc\u0026oacute;n O, et al. Microsatellite instability predicts clinical outcome in radiation-treated endometrioid endometrial cancer. Int J Radiat Oncol Biol Phys. 2010;76(1):9\u0026ndash;13.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBae H, Rha SE, Kim H, Kang J, Shin YR. Predictive Value of Magnetic Resonance Imaging in Risk Stratification and Molecular Classification of Endometrial Cancer. Cancers. 2024;16(5):921.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbbasian Ardakani A, Bureau NJ, Ciaccio EJ, Acharya UR. Interpretation of radiomics features-A pictorial review. Comput Methods Programs Biomed. 2022;215:106609.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZong L, Sun Z, Mo S, Lu Z, Yu S, Xiang Y, et al. PD-L1 expression in tumor cells is associated with a favorable prognosis in patients with high-risk endometrial cancer. Gynecol Oncol. 2021 Sept;162(3):631\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcConechy MK, Talhouk A, Li-Chang HH, Leung S, Huntsman DG, Gilks CB, et al. Detection of DNA mismatch repair (MMR) deficiencies by immunohistochemistry can effectively diagnose the microsatellite instability (MSI) phenotype in endometrial carcinomas. Gynecol Oncol. 2015;137(2):306\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bmim","sideBox":"Learn more about [BMC Medical Imaging](http://bmcmedimaging.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bmim/default.aspx","title":"BMC Medical Imaging","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Endometrial cancer, Microsatellite instability, Deep Learning, Radiomics, Magnetic Resonance Imaging, Segmentation, Prediction Model","lastPublishedDoi":"10.21203/rs.3.rs-8106086/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8106086/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003ePreoperative determination of microsatellite instability (MSI) status in endometrial cancer (EC) is crucial for guiding immunotherapy, but currently relies on invasive tissue sampling. This study aimed to develop and validate a fully automated, non-invasive framework using deep learning-based MRI segmentation and radiomic analysis for preoperative prediction of MSI status in EC.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eIn this retrospective multicenter study, 867 EC patients from three institutions were enrolled (593 for model development, 274 for external testing). MSI status was determined by immunohistochemistry. A cascaded V-Net model was developed for automated tumor segmentation on T2-weighted, diffusion-weighted, and contrast-enhanced T1-weighted images. Radiomic features were extracted and selected based on interobserver reliability, redundancy removal, and least absolute shrinkage and selection operator regression. Three classifiers\u0026mdash;support vector machine (SVM), random forest (RF), and logistic regression (LR)\u0026mdash;were trained and externally validated. Segmentation performance was assessed using the Dice similarity coefficient (DSC), sensitivity, and specificity. Classification performance was evaluated by the area under the curve (AUC), accuracy, sensitivity, and specificity, with group comparisons using DeLong tests.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eIn the external testing cohort, automated segmentation achieved mean DSCs of 85.8%\u0026plusmn;13.3% for DWI, 76.2%\u0026plusmn;17.3% for T2WI, and 80.5%\u0026plusmn;14.6% for CE-T1WI. The radiomic models based on automated segmentation yielded AUCs of 0.876 (SVM), 0.867 (RF), and 0.800 (LR) for predicting MSI status, which were not significantly different from the performance of models based on manual segmentation (AUCs: 0.915, 0.890, 0.821; all P\u0026thinsp;\u0026gt;\u0026thinsp;0.05).\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eDeep learning-based automated MRI segmentation can reliably delineate EC tumors, and when combined with radiomic analysis, provides an effective, non-invasive method for preoperative prediction of MSI status. This automated framework demonstrates performance comparable to manual segmentation, supporting its potential integration into clinical workflows to aid preoperative decision-making.\u003c/p\u003e","manuscriptTitle":"Deep Learning-Based MRI Segmentation for Non-Invasive Prediction of Microsatellite Instability in Endometrial Cancer: A Multicenter Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-30 09:11:20","doi":"10.21203/rs.3.rs-8106086/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-16T07:05:45+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-01-11T11:59:52+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-01-05T08:33:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"45862967254074641459803498185205596419","date":"2026-01-03T14:15:45+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"7676520383077279233292750315336203316","date":"2025-12-30T12:24:31+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-12-23T11:48:18+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-11-21T20:33:37+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-11-21T10:14:55+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-11-21T10:13:29+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Imaging","date":"2025-11-13T13:08:04+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":"9bf3acc8-29af-4874-99f9-2121316f988e","owner":[],"postedDate":"December 30th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-12T13:28:19+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-30 09:11:20","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8106086","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8106086","identity":"rs-8106086","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}