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

Area of Science:

• Urology
• Oncology
• Medical Informatics

Background:

• Accurate preoperative prediction of lymph node invasion (LNI) in prostate cancer is crucial for surgical planning.
• Existing nomograms struggle to balance the risks of missing LNI and unnecessary lymph node dissections.
• Machine learning (ML) offers enhanced predictive capabilities for complex clinical data.

Purpose of the Study:

• To develop and evaluate ML models for predicting LNI in prostate cancer.
• To compare the performance of ML models against traditional nomograms.
• To enhance model interpretability using SHapley Additive exPlanations (SHAP).

Main Methods:

• Developed ML models using clinicopathologic features to predict LNI.
• Applied Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance.
• Trained and evaluated four ML algorithms (k-NN, Random Forest, SVM, XGBoost) using 10-fold cross-validation.
• Assessed performance via accuracy, sensitivity, specificity, and AUC; utilized SHAP for interpretability.

Main Results:

• The Random Forest model exhibited the highest predictive performance.
• Key predictors included PSA density, clinical stage, and cribriform pattern.
• SHAP analysis provided visualization of feature contributions.
• Random Forest and XGBoost demonstrated superior discrimination compared to existing nomograms.

Conclusions:

• ML models show potential to surpass traditional nomograms in predicting prostate cancer LNI.
• Dataset balancing and explainability tools like SHAP are vital for ML model efficacy.
• External validation and incorporating additional features can improve model generalizability.