Frontiers in oncology · 2026
Radiomics-based interpretable machine learning model from multiphasic CT imaging for predicting pathological grade in upper tract urothelial carcinoma: a multicenter study.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Yuan Zhanpeng, Mei Yuhua, Peng Xiang, Wei Zongjie, Xv Yingjie, Xiao Bangxin, Xiao Mingzhao
Аннотация
OBJECTIVE: This study aims to develop a non-invasive machine learning (ML) model based on radiomic features extracted from computed tomography urography (CTU) images to predict the pathological grade of upper tract urothelial carcinoma (UTUC), and aims to assist in preoperative diagnosis.
METHODS: A retrospective multicenter study was conducted with 338 patients who underwent radical nephroureterectomy (RNU) and preoperative CTU between June 2015 and June 2024. Radiomic features were extracted from non-contrast, arterial and venous CTU phases using the PyRadiomics. A total of 3,864 features were initially extracted, with feature selection performed using intraclass correlation coefficient, Spearman correlation, t-tests and LASSO regression. Six machine learning models (XGBoost, ExtraTrees, RandomForest, Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), Light Gradient Boosting Machine (LGBM)) were employed to construct predictive models. Model performance was evaluated using receiver operating characteristic (ROC) curves, AUC, sensitivity, specificity and other metrics. SHAP method was used for model interpretability.
RESULTS: In the training set, LGBM demonstrated the highest discriminative performance, with an AUC of 0.945, and showed balanced sensitivity (84.5%) and specificity (91.2%). In the test set, LGBM maintained robust generalizability with an AUC of 0.829 (sensitivity: 71.8%, specificity: 77.4%), and outperformed other models, including SVM and MLP. SHAP analysis revealed that the most influential features for prediction were related to the venous and arterial CTU phases, enhancing the model's interpretability.
CONCLUSION: The developed radiomics-based ML model demonstrates excellent potential in predicting pathological grade for UTUC preoperatively. It provides a non-invasive, clinically interpretable tool for improving diagnostic accuracy and aiding in personalized treatment planning.
METHODS: A retrospective multicenter study was conducted with 338 patients who underwent radical nephroureterectomy (RNU) and preoperative CTU between June 2015 and June 2024. Radiomic features were extracted from non-contrast, arterial and venous CTU phases using the PyRadiomics. A total of 3,864 features were initially extracted, with feature selection performed using intraclass correlation coefficient, Spearman correlation, t-tests and LASSO regression. Six machine learning models (XGBoost, ExtraTrees, RandomForest, Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), Light Gradient Boosting Machine (LGBM)) were employed to construct predictive models. Model performance was evaluated using receiver operating characteristic (ROC) curves, AUC, sensitivity, specificity and other metrics. SHAP method was used for model interpretability.
RESULTS: In the training set, LGBM demonstrated the highest discriminative performance, with an AUC of 0.945, and showed balanced sensitivity (84.5%) and specificity (91.2%). In the test set, LGBM maintained robust generalizability with an AUC of 0.829 (sensitivity: 71.8%, specificity: 77.4%), and outperformed other models, including SVM and MLP. SHAP analysis revealed that the most influential features for prediction were related to the venous and arterial CTU phases, enhancing the model's interpretability.
CONCLUSION: The developed radiomics-based ML model demonstrates excellent potential in predicting pathological grade for UTUC preoperatively. It provides a non-invasive, clinically interpretable tool for improving diagnostic accuracy and aiding in personalized treatment planning.