Artificial neural network–based prediction of lymph node metastasis in prostate cancer patients undergoing radical prostatectomy

Abstract Background Accurate preoperative prediction of Lymph Node Metastasis (LNM) is essential for selecting prostate cancer patients who may benefit from extended pelvic lymph node dissection (ePLND) during radical prostatectomy (RP). Conventional nomograms have limited ability to capture complex, nonlinear interactions among clinical variables. This study aimed to develop and internally validate a machine learning (ML)–based model for preoperative LNM prediction and to compare its performance with that of established nomogram-based risk estimation. Materials and methods We retrospectively analyzed 110 consecutive prostate cancer patients who underwent RP with concomitant ePLND between January 2017 and January 2021. Preoperative demographic, clinical, biopsy, and multiparametric magnetic resonance imaging (mpMRI) variables were collected. Multiple ML algorithms were trained and evaluated, including logistic regression, decision tree, support vector machines, naïve Bayes, boosted tree, and artificial neural networks (ANN). Model performance was assessed using accuracy and area under the receiver operating characteristic curve (AUC). Internal validation was performed using a training–validation–test split for ANN, while five-fold cross-validation was applied for the remaining algorithms. Results ANN demonstrated the highest diagnostic performance, achieving an accuracy of 94.1% and an AUC of 0.962 in the independent test dataset. Other ML models showed accuracies ranging from 79% to 86%, with AUC values exceeding 0.70. ANN outperformed logistic regression and showed superior discrimination compared with Briganti nomogram–based stratification. Conclusions An ANN-based model may improve preoperative identification of patients at risk for LNM. ML-assisted decision support may complement existing risk stratification tools; however, its clinical utility remains to be established. External validation in larger multicenter cohorts is warranted.