Non-invasive evaluation of muscle invasion and survival prognosis in bladder cancer using enhanced CT-based deep learning radiomics: a multi-center real-world cohort study.
Authors
Affiliations (9)
Affiliations (9)
- Department of Urology, Xiangya Hospital, Central South University, Changsha 410008, China.
- National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha 410008, China.
- Department of Urology, the Second Affiliated Hospital, Guizhou Medical University, Guiyang 550000, China.
- Department of Pathology, the Third Xiangya Hospital of Central South University, Changsha 410013, China.
- Department of Urology, the Second Hospital of University of South China, Hengyang 421001, Hunan, China.
- Department of Imaging, the First People's Hospital of Kaili, Kaili 556000, Guiyang, China.
- Department of Imaging, the Second Affiliated Hospital, Guizhou Medical University, Kaili 556000, Guiyang, China.
- Department of Urology, Xiangya Boai Rehabilitation Hospital, Changsha 410146, China.
- Department of Urology, Hunan Provincial People's Hospital/the First Affiliated Hospital of Hunan Normal University, Changsha 410005, China.
Abstract
Bladder cancer (BLCA) is a prevalent malignancy characterized by high recurrence and poor prognosis, particularly muscle-invasive bladder cancer (MIBC). Histopathology, the gold standard for assessing muscle invasion, often suffers from sampling errors and operator dependency, underscoring the need for non-invasive, accurate preoperative assessment methods. This study aimed to develop and validate a hybrid artificial intelligence (AI) model based on computed tomography (CT) radiomics and deep learning (DL) to predict MIBC and overall survival (OS) preoperatively in BLCA patients. A total of 1370 patients from 6 academic medical centers were retrospectively included. Preoperative contrast-enhanced CT scans were analyzed to extract handcrafted radiomic features using PyRadiomics and DL features using ResNet101, followed by machine learning (ML)-based modeling for prediction. A hybrid model combining radiomic and DL features was constructed and validated in internal and external cohorts. Model performance was evaluated using metrics such as the area under the curve (AUC) and Cox proportional hazards analysis for OS prediction. The DL radiomics nomogram (DLRN) model demonstrated superior diagnostic performance, achieving an AUC of 0.807 in the internal validation cohort and 0.783 in the external multi-center validation cohort for predicting muscle invasion. The DLRN generated an imaging-derived risk score (DLRN score), which was subsequently incorporated as one covariate into a multivariable Cox proportional hazards model together with clinicopathological variables to evaluate OS. Using this approach, patients were effectively stratified into high- and low-risk groups for OS, showing robust generalizability across diverse clinical settings. AI-assisted diagnostics significantly improved the sensitivity and accuracy of urologists, particularly among less experienced clinicians. The DLRN model provides a reliable, non-invasive tool for preoperative assessment of muscle invasion and prognosis in BLCA. Addressing histopathology limitations, it offers valuable insights for personalized treatment strategies, paving the way for precision oncology in real-world clinical applications.