This systematic review and meta-analysis evaluated the diagnostic performance of artificial intelligence (AI) models that analyze preoperative prostate MRI images in conjunction with clinical parameters for predicting extraprostatic extension (EPE) in prostate cancer. A comprehensive search of PubMed, Embase, and Web of Science up to July 2025 identified 14 eligible studies involving 2,131 patients. The pooled analysis demonstrated that integrated radiomics-clinical models achieved high diagnostic performance, with a sensitivity of 0.83 (95% CI: 0.78-0.87), specificity of 0.82 (95% CI: 0.77-0.86), and an area under the curve (AUC) of 0.89 (95% CI: 0.86-0.92). The diagnostic odds ratio (DOR) was 19.82 (95% CI: 12.33-31.86), indicating robust discrimination between EPE-positive and EPE-negative cases. Subgroup analysis suggested models using deep learning algorithms had marginally higher accuracy (DOR: 24.6) than those using traditional machine learning (DOR: 17.3), though the difference was not statistically significant. Heterogeneity among studies stemmed from variations in MRI protocols, segmentation methods, and modeling approaches. No significant publication bias was detected. The results affirm that integrating radiomic features from multiparametric MRI (e.g., T2-weighted, diffusion-weighted imaging) with clinical variables (e.g., PSA, Gleason score) significantly outperforms conventional assessments for preoperative EPE prediction, demonstrating excellent diagnostic accuracy and supporting its potential clinical application in risk stratification. This supports the potential of combined models to enhance risk stratification and guide personalized surgical planning. Future research should prioritize standardized radiomics workflows, external validation, and multi-center collaborations to facilitate clinical adoption.
Wu et al. (Mon,) studied this question.