Application Value of Radiomics-Based Machine Learning for Preoperative Risk Stratification of Bladder Cancer: Systematic Review and Meta-Analysis

Background: Some researchers have explored the application of radiomics-based machine learning to detect preoperative muscle invasion, high-grade tumors, human epidermal growth factor receptor 2 expression, and other risk factors for bladder cancer. However, systematic evidence proving its effectiveness remains lacking. Objective: This study aimed to evaluate the performance of radiomics-based machine learning in preoperative risk stratification for patients with bladder cancer. These findings could contribute to advancing the development or updating of intelligent risk assessment tools for bladder cancer. Methods: The Embase, Cochrane Library, PubMed, and Web of Science databases were systematically retrieved for publicly available studies on the effectiveness of radiomics-based machine learning (ML) in the preoperative risk stratification of bladder cancer up to October 17, 2025. The risk of bias in the included studies was evaluated using the Prediction Model Risk of Bias Assessment Tool for Artificial Intelligence. The overall quality of the studies was quantified using the Radiomics Quality Scoring tool. The certainty of the evidence was graded using the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) framework. Subgroup analyses were conducted according to the type of imaging source and modeling method. Results: This meta-analysis ultimately incorporated 57 studies with a total of 11,933 participants. These studies primarily used radiomics-based ML to identify muscle invasion (n=34) and high-grade tumors (n=16). Additionally, the methodology was used to evaluate human epidermal growth factor receptor 2 positive expression (n=3), Ki-67 expression (n=2), and lymph node staging (n=2) preoperatively in bladder cancer. In the validation sets, the pooled area under the receiver operating characteristic curve (AUROC) for identifying muscle invasion was 0.893 (95% CI 0.840-0.948), 0.916 (95% CI 0.891-0.942), and 0.840 (95% CI 0.737-0.958) for computed tomography (CT)-, magnetic resonance imaging (MRI)-, and ultrasound-based radiomics, respectively. The AUROC was 0.874 (95% CI 0.852-0.896) and 0.921 (95% CI 0.867-0.979) for models integrating clinical features with CT- or MRI-based radiomics, respectively. The pooled AUROC for diagnosing high-grade tumors was 0.874 (95% CI 0.775-0.985), 0.846 (95% CI 0.663-1.000), and 0.750 (95% CI 0.636-0.884) for CT-, MRI-, and ultrasound-based radiomics, respectively. Furthermore, the AUROC was 0.919 (95% CI 0.774-1.000) for MRI-based radiomics combined with clinical features. Conclusions: This is the first systematic review to comprehensively evaluate the role of radiomics in preoperative risk stratification for bladder cancer. It provides evidence to inform the development and refinement of future ML-based tools for image analysis in this setting. However, this evidence faces significant challenges, including methodological shortcomings and a high risk of bias and low GRADE level, which preclude its readiness for clinical translation. Future studies should standardize the methodological workflows in radiomics, conduct multicenter research, and thoroughly evaluate and discuss the validity of external validation.