CT-based clinical deep learning radiomics nomogram for preoperative differentiation of pyelocaliceal upper tract urothelial carcinoma and xanthogranulomatous pyelonephritis
Abstract
ObjectiveThis study sought to create a CT-based clinical deep learning radiomics (CDLR) model to noninvasively differentiate between pyelocaliceal upper tract urothelial carcinoma (UTUC) and xanthogranulomatous pyelonephritis (XGP) preoperatively, aiding in personalized treatment decisions.MethodsA retrospective analysis included 161 patients with pathologically confirmed pyelocaliceal UTUC or XGP following surgery. Patients were randomly divided into a training cohort (n=112) and a validation cohort (n=49). Radiomics (Rad) features and deep learning (DL) features were extracted from the corticomedullary phase CT images. Combined with clinical features, and after feature selection, a logistic regression (LR) classifier was employed to construct the clinical model, the deep learning radiomics (DLR) model, and the clinical deep learning radiomics (CDLR) model. Model performance was evaluated using receiver operating characteristic (ROC) curves and decision curve analysis (DCA), with interpretability assessed by the shapley additive explanations (SHAP) method. In addition, the diagnostic performance of the models was compared with that of a radiologist with 5 years of experience in abdominal CT diagnosis.ResultsThe CDLR model achieved AUCs of 0.955 in the training cohort and 0.930 in the validation cohort. In the validation cohort, the CDLR model significantly outperformed the radiologist’s assessment (AUC: 0.930 vs. 0.695, P = 0.0006) and yielded numerically higher AUCs than the clinical model (0.823) and the DLR model (0.882). DCA indicated that the CDLR model provided the highest net clinical benefit. SHAP analysis demonstrated good model interpretability.ConclusionThe CDLR model integrating clinical, Rad, and DL features serves as an effective noninvasive tool for preoperative differentiation of pyelocaliceal UTUC and XGP, offering valuable reference for clinical decision-making.
Related
- Xanthogranulomatous Pyelonephritis in a Horseshoe Kidney: Effective Conservative Therapy in an Immunocompromised Patient
- Diagnostic and Management Challenges in Xanthogranulomatous Pyelonephritis: A Case Report
- Bilateral Xanthogranulomatous Pyelonephritis in a Patient With Spina Bifida and Neurogenic Bladder: A Case Report
- An Extremely High Preoperative CA19-9 Cannot Exclude the Diagnosis of Xanthogranulomatous Cholecystitis: A Case Report and Literature Review
- Expression of NECTIN-4 and TROP-2 in Resected Upper Tract Urothelial Carcinoma with and Without Preoperative Chemotherapy
BibTeX
@article{feng2026ctbased,
title = {{CT-based clinical deep learning radiomics nomogram for preoperative differentiation of pyelocaliceal upper tract urothelial carcinoma and xanthogranulomatous pyelonephritis}},
author = {Yi Feng and Yiheng Cheng and Jian Zeng and Junyu Lin and Hao Qin and Yizheng Yan and Zile Huang and Yongxian Wu and Xiao Liu and Fuxiang Fang and Linjian Mo and Wei Xu and Chengyang Li},
journal = {Frontiers in Oncology},
year = {2026},
month = nov,
doi = {10.3389/fonc.2026.1880690},
url = {https://www.frontiersin.org/articles/10.3389/fonc.2026.1880690},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - CT-based clinical deep learning radiomics nomogram for preoperative differentiation of pyelocaliceal upper tract urothelial carcinoma and xanthogranulomatous pyelonephritis
AU - Yi Feng
AU - Yiheng Cheng
AU - Jian Zeng
AU - Junyu Lin
AU - Hao Qin
AU - Yizheng Yan
AU - Zile Huang
AU - Yongxian Wu
AU - Xiao Liu
AU - Fuxiang Fang
AU - Linjian Mo
AU - Wei Xu
AU - Chengyang Li
PY - 2026
DA - 2026/11/02
JO - Frontiers in Oncology
DO - 10.3389/fonc.2026.1880690
UR - https://www.frontiersin.org/articles/10.3389/fonc.2026.1880690
AB - ObjectiveThis study sought to create a CT-based clinical deep learning radiomics (CDLR) model to noninvasively differentiate between pyelocaliceal upper tract urothelial carcinoma (UTUC) and xanthogranulomatous pyelonephritis (XGP) preoperatively, aiding in personalized treatment decisions.MethodsA retrospective analysis included 161 patients with pathologically confirmed pyelocaliceal UTUC or XGP following surgery. Patients were randomly divided into a training cohort (n=112) and a validation cohort (n=49). Radiomics (Rad) features and deep learning (DL) features were extracted from the corticomedullary phase CT images. Combined with clinical features, and after feature selection, a logistic regression (LR) classifier was employed to construct the clinical model, the deep learning radiomics (DLR) model, and the clinical deep learning radiomics (CDLR) model. Model performance was evaluated using receiver operating characteristic (ROC) curves and decision curve analysis (DCA), with interpretability assessed by the shapley additive explanations (SHAP) method. In addition, the diagnostic performance of the models was compared with that of a radiologist with 5 years of experience in abdominal CT diagnosis.ResultsThe CDLR model achieved AUCs of 0.955 in the training cohort and 0.930 in the validation cohort. In the validation cohort, the CDLR model significantly outperformed the radiologist’s assessment (AUC: 0.930 vs. 0.695, P = 0.0006) and yielded numerically higher AUCs than the clinical model (0.823) and the DLR model (0.882). DCA indicated that the CDLR model provided the highest net clinical benefit. SHAP analysis demonstrated good model interpretability.ConclusionThe CDLR model integrating clinical, Rad, and DL features serves as an effective noninvasive tool for preoperative differentiation of pyelocaliceal UTUC and XGP, offering valuable reference for clinical decision-making.
SN - 2234-943X
ER -