Non-invasive prediction of lymph node involvement in prostate cancer via machine learning on whole-prostate MRI.

Background: Assessment of lymph node status is essential for guiding surgical decisions, prognosis evaluation, and recurrence risk estimation in prostate cancer (PCa). Therefore, this research seeks to establish and corroborate an integrated prediction model that leverages T2-weighted MRI scans to accurately identify lymph node involvement (LNI) in PCa patients. Methods: A retrospective cohort of 339 prostate cancer patients who underwent preoperative whole-prostate MRI followed by radical prostatectomy and extended pelvic lymph node dissection (ePLND) was evaluated. To non-invasively predict LNI, we developed a composite machine learning model integrating MRI-derived radiomics, deep learning features, and clinical parameters based on the 2012 Briganti nomogram. Model performance and clinical utility were assessed using the area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA). Results: The DRBN model demonstrated excellent predictive power in both the training and validation sets, with ROC-AUC scores of 0.963 and 0.920, and PR-AUC values of 0.950 and 0.921, correspondingly. Additionally, the DRBN model exhibited satisfactory calibration and clinical utility in both cohorts. Conclusion: Integrating clinical, radiomic, and deep learning features from whole-prostate MRI provides a feasible non-invasive approach for predicting LNI.