The combined clinical-intratumoral-peritumoral-peritumoral radiomics model, alongside its SHAP visualization tool, enhanced the accuracy (ACC) of non-invasive preoperative LVSI prediction, demonstrating certain potential for clinical application.
Abstract
Rationale
AND
Objectives
To develop and validate a machine learning model combining multiparametric Magnetic Resonance Imaging (MRI) radiomics and clinical indicators for predicting lymphovascular space invasion (LVSI) in endometrial cancer (EC).
Materials And Methods
This retrospective study enrolled EC patients who underwent preoperative MRI at two centers. Of 567 initially screened patients, 408 were included per inclusion/exclusion criteria, divided into training and validation sets by hospital. Clinical risk factors and intratumoral/peritumoral radiomic features were identified. Six machine learning algorithms were used to build models; the one with the highest validation Area Under the Curve (AUC) was optimal. Five additional models were developed, and performance was evaluated via AUC, calibration curves, and decision curve analysis (DCA).
Results
Logistic regression identified CA125 and tumor diameter as independent LVSI risk factors. Six machine learning models were built with CA125, tumor diameter, Rad_Score1 and Rad_Score2; the NeuralNetwork performed best (validation AUC=0.803). The combined clinical-intratumoral-peritumoral radiomics model achieved the highest AUC(training AUC = 0.863, validation AUC = 0.803), with good calibration (Hosmer-Lemeshow test, P > 0.05) and favorable net clinical benefit (threshold 0.1-0.7). SHapley Additive exPlanations (SHAP) analysis enhanced model interpretability.
Conclusion
This study systematically compared the predictive performance of six machine learning models for LVSI in EC, identifying the NeuralNetwork model as the most optimal. The combined clinical-intratumoral-peritumoral radiomics model, alongside its SHAP visualization tool, enhanced the accuracy (ACC) of non-invasive preoperative LVSI prediction, demonstrating certain potential for clinical application.
Introduction: MRI is the gold-standard imaging modality for rectal cancer (RC) local staging, but the ability to determine tumor invasion (pT) and nodal status (pN) remains limited in clinical practice. The main objective of this study is to evaluate the performance of different machine learning (ML) models based on clinical–radiological and radiomic variables for predicting these categories from preoperative MRI. Methods: A retrospective observational study was conducted involving 152 patients with RC (70 without neoadjuvant therapy and 82 with neoadjuvant therapy). Radiomic features were extracted from high-resolution T2 sequences using two independent segmentations: tumor and tumor + mesorectum. Twenty-three ML algorithms were evaluated using cross-validation to predict pT and pN. For each combination of outcome, cohort, data source and segmentation, an optimal model was selected based on the area under the curve (AUC). Results: Models based on clinical and radiological variables showed the most consistent performance, particularly in the overall cohort, with AUCs of 0.767 for pT and 0.764 for pN. The radiomic and combined models achieved a moderate and heterogeneous performance, with maximum AUCs of 0.770 for pT and 0.732 for pN. Conclusions: The clinico-radiological variables analyzed using ML showed a predictive performance similar to that of a radiologist. Radiomics did not show significant improvement in this setting.
Marta García Cerezo, David López Cornejo, Alba Ortigosa-Palomo et al.· Applied Sciences· 0 citations
Background Lymphovascular space invasion (LVSI) is an important pathological feature associated with tumor aggressiveness and adverse prognosis in cervical cancer. However, reliable preoperative prediction of LVSI remains a challenge. This study aimed to develop and validate an MRI-based radiomics model incorporating intratumoral and peritumoral features for LVSI prediction and systematically evaluate the optimal peritumoral extent. Materials and methods In this single-center study comprising both retrospective and prospective cohorts, 204 patients with pathologically confirmed cervical cancer were randomly divided into a training cohort (n = 142) and a test cohort (n = 62). Radiomics features were extracted from the intratumoral and peritumoral regions with expansion distances of 1, 3, and 5 mm. Feature selection was performed using intraclass correlation coefficient (ICC) analysis, minimum redundancy maximum relevance (mRMR), and least absolute shrinkage and selection operator (LASSO) analysis. Six machine learning algorithms, including logistic regression [LR], support vector machine, random forest, ExtraTrees, LightGBM, and multilayer perceptron, were used to construct the predictive models. The model performance was evaluated using receiver operating characteristic analysis, calibration curves, and decision curve analysis. SHapley Additive Explanations (SHAP) were applied to interpret the optimal models. Results Among all ROI configurations, the Intra+P1 model demonstrated the best overall performance, particularly when it was combined with LR. The LR-based Intra+P1 model achieved an AUC of 0.866 (95% CI: 0.807–0.926) in the training cohort and 0.843 (95% CI: 0.731–0.955) in the test cohort respectively. Increasing the peritumoral expansion from 1 mm to 3 mm or 5 mm did not further improve predictive performance. The addition of clinical variables (tumor diameter and SCC level) did not provide a significant incremental predictive value. Calibration and decision curve analyses demonstrated good agreement and favorable clinical utility of the model. SHAP analysis showed that both intratumoral and peritumoral features contributed substantially to the model prediction, with texture features playing a dominant role. Conclusion This interpretable MRI-based radiomics model facilitates accurate preoperative prediction of LVSI in cervical cancer. A narrowly defined 1-mm peritumoral region provides the most informative complementary information, underscoring the importance of the tumor-invading front. This approach offers a noninvasive tool for risk stratification and may support individualized treatment decision-making.
Xian-Yan Wu, Chuan-Fang Xu, Sha Shi et al.· Frontiers in Oncology· 0 citations
RATIONALE AND OBJECTIVES
This study aimed to develop and validate an interpretable model using pretreatment multiparametric magnetic resonance imaging (mpMRI) radiomics and clinical data to predict neoadjuvant chemotherapy (NAC) sensitivity and recurrence-free survival (RFS) in breast cancer. The model was interpreted using the SHapley Additive exPlanations (SHAP) framework to enhance clinical transparency and support individualized treatment planning.
MATERIALS AND METHODS
In this multicenter retrospective study, 373 patients with pretreatment mpMRI (T2, dynamic contrast-enhanced, diffusion-weighted imaging) were enrolled. Patients were classified as NAC-insensitive (Miller-Payne grades 1-3) or sensitive (grades 4-5). Radiomic features from manual segmentations were integrated with clinical variables. A combined model was built using random forest and interpreted via SHapley Additive exPlanations (SHAP). A nomogram score from the model was assessed for RFS using Cox regression.
RESULTS
The combined model achieved an AUC of 0.859 (95% CI: 0.764-0.954) in the internal validation set, outperforming the clinical (AUC 0.643) and radiomics (AUC 0.844) models, with similar performance in external validation (AUC 0.862). SHAP analysis identified sigma_5_0_mm_3D_glcm_DifferenceEntropy_T2 as the most influential feature. A nomogram score ≥95 predicted high sensitivity probability. Lower nomogram scores were independently associated with worse RFS.
CONCLUSION
The interpretable radiomics-clinical model based on pretreatment mpMRI effectively predicts NAC sensitivity and RFS, offering potential to aid personalized treatment decisions. By providing transparent, individualized predictions via SHAP, this tool shows potential for optimizing personalized treatment strategies and reducing unnecessary toxicity.
ABSTRACT Purpose To explore the value of constructing a radiomics combined model based on preoperative dynamic contrast‐enhanced magnetic resonance imaging (DCE‐MRI) for predicting lymphovascular invasion (LVI) in breast cancer. Materials and Methods Retrospective data collection was performed from December 2022 to November 2025, involving 912 patients with pathologically confirmed breast cancer who underwent DCE‐MRI examinations at two medical centers. Data from 757 patients at Center 1 (Guangdong Provincial Maternal and Child Health Hospital) were used as the model development set. Through stratified random sampling, this set was divided into a training set (n = 529) and an internal validation set (n = 228) at a 7:3 ratio. The 155 patients from Center 2 (The First Affiliated Hospital of Jinan University) formed an independent external test set. Three‐dimensional regions of interest (ROIs) were manually delineated on DCE‐MRI images, and 1197 radiomics features were extracted using the PyRadiomics software. Within the training set, key features were selected through univariate analysis (p < 0.05) using Pearson correlation analysis (|r| < 0.9) and LASSO regression (10‐fold cross‐validation). Key features were evaluated using logistic regression (LR), support vector machine (SVM), K‐nearest neighbors (KNN), random forest (RF), extreme trees (ET), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), gradient boosting machine (GBM), adaptive boosting (AdaBoost), and multilayer perceptron (MLP) to construct radiomics models, identifying the model with the highest predictive performance. Simultaneously, univariate and multivariate logistic regression analyses were employed to identify independent risk factors for breast cancer LVI from clinical and pathological characteristics, and a clinical model was constructed. A combined model was constructed by integrating the independent risk factors for breast cancer LVI with the optimal radiomics model, and a nomogram was developed. The predictive performance and clinical utility of each model for breast cancer LVI were evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Results Based on DCE‐MRI images, 18 key features were ultimately selected for constructing the radiomics model. Among these, the ExtraTrees radiomics model demonstrated the highest predictive performance, achieving an area under the curve (AUC) of 0.812 (95% CI: 0.7768–0.8480) on the training set and 0.653 (95% CI: 0.5977–0.7081), with accuracy superior to the other nine machine learning models in the validation set. Independent risk factors for breast cancer LVI included sentinel lymph node status, ER, and TIC, which were used to construct a clinical model. The AUC values for the training set were 0.700 (95% CI: 0.6557–0.7440), 0.752 (95% CI: 0.7101–0.7933), and 0.796 (95% CI: 0.7582–0.8341). The corresponding AUC values on the validation set were 0.642 (95% CI: 0.5695–0.7145), 0.664 (95% CI: 0.5901–0.7374), and 0.704 (95% CI: 0.6339–0.7745). The AUC values for the test set were 0.603 (95% CI: 0.5131–0.6923), 0.692 (95% CI: 0.6097–0.7739), and 0.703 (95% CI: 0.6187–0.7866). Calibration curves demonstrated good agreement between the combined model's predictions and actual observed outcomes. DCA indicated that the model provided greater clinical net benefit across a broad range of threshold probabilities. Conclusions The combined model constructed based on DCE‐MRI effectively predicts LVI in breast cancer, demonstrating superior performance compared to standalone radiomics or clinical models. It exhibits excellent calibration and clinical utility, offering promising potential to support preoperative individualized treatment decisions.
Hon-Gen Li, Qing-Wen Xiao, Yi-Hui Zeng et al.· Cancer Reports· 0 citations
The interpretable MRI-based XGBoost model showed favorable performance for identifying the vascular dissemination phenotype in HCC, with SHAP-based interpretability and exploratory transcriptomic context.
Junhan Pan, Cong Zhang, Yitian Wu et al.· Journal of Hepatocellular Ca...· 0 citations
Purpose To develop an interpretable ensemble learning model integrating multiple machine learning algorithms, intratumoral and peritumoral CT radiomics, and serum biomarkers for predicting poorly differentiated esophageal squamous cell carcinoma (ESCC). Methods This retrospective study enrolled 261 ESCC patients, who were randomly allocated to training (n=183) and validation (n=78) cohorts. Radiomics features were extracted from intratumor, peritumoral 0.3 cm and intra-peritumoral 0.3 cm areas of enhanced CT arterial phase. Serum neutrophils (NEU) and alkaline phosphatase (ALP) were collected. Following feature dimensionality reduction, ensemble radiomics score (ENs) was calculated for each region. Seven individual machine learning models and an ensemble model (ENML) were subsequently constructed. Model performance was assessed using the area under the curve (AUC), confidence interval (CI) and decision curve analysis (DCA), while interpretability was evaluated via SHAP, correlation, and restricted cubic spline (RCS) analyses. Results In the validation cohort, ENML achieved the highest AUC (0.799), F1 score (0.810), recall (0.870), and Brier score (0.166), and demonstrated clinical net benefit across threshold probabilities ranging from 10% to 75%. Bootstrap resampling with 1,000 iterations confirmed its stable performance in the full cohort (AUC: 0.849, 95% CI: 0.790–0.891). Spearman correlation analyses revealed strong positive associations between intratumoral and peritumoral radiomic features (r = 0.54, 0.49, and 0.51, respectively; all P < 0.001). RCS analyses identified significant nonlinear relationships between intratumor, intra-peritumoral 0.3 cm, ENs, and ALP levels and the increased incidence of poorly differentiated ESCC (Overall P < 0.05; Nonlinear P < 0.05). Additionally, both ALP and NEU exhibited significant inflection point effects in predicting poorly differentiated ESCC. SHAP analyses identified the IntraPeri3mm_wavelet.LLL_glszm_SmallAreaLowGrayLevelEmphasis and SVM as the primary contributors to the fusion model. Conclusions The interpretable ENML model exhibits favorable discriminative capability for predicting poorly differentiated ESCC; however, further multicenter external validation is warranted to confirm its generalizability.
Jun Chen, Ji-Qiang He, Xiao-Jiao Zhang et al.· Frontiers in Oncology· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
