Comparatıve evaluatıon of the brıgantı nomogram, partın nomogram, mskcc nomogram and a machıne learnıng-based model ın predıctıng lymph node ınvasıon before radıcal prostatectomy.

INTRODUCTION

Accurate preoperative prediction of lymph node invasion (LNI) is crucial for deciding on extended pelvic lymph node dissection (ePLND) in radical prostatectomy. Traditional nomograms such as Briganti, Partin, and MSKCC are widely used, but machine learning (ML)–based models may improve predictive accuracy. MATERIALS AND

METHODS

Data from 471 prostate cancer patients were analyzed, including demographic, clinical, and histopathological variables and scores from Briganti, Partin, and MSKCC nomograms. Eleven ML algorithms were evaluated, with performance assessed by AUC-ROC, accuracy, sensitivity, specificity, and F1-score. Class imbalance was addressed with resampling techniques, and feature importance analyses were performed.

RESULTS

LNI was present in 97 patients (20.6%). Significant predictors included MSKCC and Partin scores, PSA, ISUP grade, PIRADS score, lymphovascular invasion, and age. Neural Network (AUC: 0.81) and Random Forest (AUC: 0.77) showed similar performance to the nomograms (MSKCC: 0.79; Briganti: 0.77; Partin: 0.78) when considering the AUC values and their 95% confidence intervals.Decision tree analysis highlighted negative core count, ISUP grade, prostate density, PSA, BMI, and age as key variables. Combining nomogram scores with ML models resulted in numerically slightly higher AUC values; however, these differences remained within a similar performance range and did not indicate a clinically meaningful improvement.

CONCLUSION

Although the AUC values of the ML models appear numerically comparable to, or slightly higher than, those of traditional nomograms, the inherent limitations of the study preclude demonstrating a clinically superior or reliably advantageous performance; therefore, multicenter prospective validation studies are warranted.