Lymph Node Invasion Prediction in Prostate Cancer: A Comparative Machine-Learning Study

Accurate preoperative prediction of lymph node invasion (LNI) in prostate cancer is critical for guiding lymph node dissection. Current nomograms often fail to optimall balance the risks of missing metastatic cases and unnecessary dissections. Machine-learning models can provide improved predictive performance through more flexible modeling of complex clinical data.

The authors developed machine-learning models using clinicopathologic features to predict LNI. Due to a significant class imbalance between LNI-positive and LNI-negative cases, a synthetic minority oversampling technique (SMOTE) was applied to balance the dataset. Four machine-learning algorithms (k-Nearest Neighbors, Random Forest, Support Vector Machine, and Extreme Gradient Boosting [XGBoost]) were trained using 10-fold cross-validation. Model performance was evaluated using accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve. SHapley Additive exPlanations (SHAP) analysis was performed for interpretability.

The Random Forest model demonstrated the highest predictive performance. Key predictive features included prostate-specific antigen (PSA) density, clinical stage, and presence of the cribriform pattern. Use of SHAP analysis enabled visualization of individual feature contributions. Compared with existing nomograms, Random Forest and XGBoost achieved superior discrimination performance.

Machine-learning models may outperform traditional nomograms in predicting LNI in prostate cancer, especially when trained on balanced datasets and combined with explainability tools such as SHAP. Further external validation and inclusion of additional features can improve model generalizability.