Development of multiphasic CT-based delta-radiomics model for predicting postoperative recurrence risk in bladder cancer.
Authors
Affiliations (8)
Affiliations (8)
- Department of Radiology, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Jiefang Avenue #1277, Wuhan, 430022, China.
- Hubei Provincial Clinical Research Center for Precision Radiology & Interventional Medicine, Wuhan, 430022, China.
- Hubei Key Laboratory of Molecular Imaging, Wuhan, 430022, China.
- Institute of Research and Clinical Innovations, Neusoft Medical Systems Co. Ltd, Shenyang, 110167, China.
- Department of Urology, Tongji Medical College, Union Hospital, Huazhong University of Science and Technology, Wuhan, 430022, China. [email protected].
- Department of Radiology, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Jiefang Avenue #1277, Wuhan, 430022, China. [email protected].
- Hubei Provincial Clinical Research Center for Precision Radiology & Interventional Medicine, Wuhan, 430022, China. [email protected].
- Hubei Key Laboratory of Molecular Imaging, Wuhan, 430022, China. [email protected].
Abstract
This study aimed to develop a CT-based delta-radiomics model for personalized prediction of postoperative prognosis in bladder cancer patients and identification of differentially expressed genes associated with tumour recurrence. This retrospective study included 316 patients with bladder cancer from Wuhan Union Hospital who underwent preoperative unenhanced and arterial-phase CT before surgical resection. Patients were randomly divided into training and internal validation cohorts at an 8:2 ratio. Radiomic features were extracted from tumor volumes of interest on both CT phases, and delta-radiomic features were calculated as arterial-phase features minus unenhanced-phase features. Feature selection was performed within the training cohort using statistical filtering, correlation analysis, and LASSO regression. Multiple machine-learning classifiers were evaluated, and a combined clinico-radiomic model was constructed by integrating the delta-radiomics score with selected clinical predictors. Model performance was assessed using ROC analysis, calibration, decision curve analysis, DeLong testing, and repeated stratified random splitting. An additional TCIA cohort of 33 patients was used for exploratory transcriptomic analysis. After feature selection, nine delta-radiomic features derived from paired unenhanced and arterial-phase CT images were retained for model construction. The LightGBM-based delta-radiomics model achieved AUCs of 0.837 and 0.822 in the training and validation cohorts, respectively. The combined clinical-radiomics model, constructed by integrating the delta-radiomics Rad-score with muscle invasion as a clinical predictor, achieved the best discrimination, with AUCs of 0.860 and 0.861 in the training and validation cohorts, respectively, and showed stable validation performance in 100 repeated splits. High radiomics scores were associated with significantly shorter recurrence-free survival. Exploratory transcriptomic analysis showed enrichment of immune-related pathways in the low-risk radiomic subgroup. Multiphasic CT-based delta-radiomics provides a non-invasive approach for predicting postoperative recurrence and stratifying prognosis in bladder cancer. A combined clinico-radiomic model incorporating delta-radiomics and muscle invasion achieved superior predictive performance and may support individualized postoperative risk assessment.