Machine learning-integrated tumor markers and mpMRI-derived scores to construct a differentiation model for prostate cancer and benign prostatic hyperplasia in men with tPSA of 4-10 ng/mL
10.12483/j.issn.1009-8291.2026.04.008
- VernacularTitle:机器学习整合肿瘤标志物与核磁评分构建tPSA位于4～10ng/mL时前列腺癌与良性前列腺增生的鉴别模型
- Author:
Pengyu WANG
1
;
Shan HE
1
;
Zhongqing WEI
1
Author Information
1. Department of Urology, The Second Affiliated Hospital of Nanjing Medical University, Jiangsu Key Laboratory of Urological Disease Prevention and Treatment, Nanjing 210011, China
- Publication Type:Journal Article
- Keywords:
prostate cancer;
prostate-specific antigen;
machine learning;
logistic regression model;
classification and regression tree
- From:
Journal of Modern Urology
2026;31(4):339-344
- CountryChina
- Language:Chinese
-
Abstract:
Objective To develop and validate a machine learning model integrating serum tumor markers with multiparametric magnetic resonance imaging(mpMRI)-derived scores for discriminating prostate cancer(PCa)from benign prostatic hyperplasia(BPH)in men with a total prostate-specific antigen(tPSA)of 4-10 ng/mL, so as to optimize prostate biopsy decisions. Methods A retrospective analysis was conducted on the clinical data of 85 PCa patients and 204 BPH patients treated in our hospital during Jan. 2022 and May 2025. All patients underwent prostate biopsy to obtain the pathological results. The data collected were analyzed using univariate and multivariate logistic regression to determine the independent risk factors for PCa. Model performance was assessed in terms of discrimination, calibration, and clinical utility using receiver operating characteristic(ROC)curves, calibration plots, and decision curve analysis(DCA). A classification and regression tree(CART)algorithm was further developed to derive biopsy decision rules. The area under the ROC curve(AUC)of the logistic and CART models were compared to select the optimal predictive model, which was subsequently visualized using SHAP values. Results Age, free prostate-specific antigen(fPSA), fPSA/tPSA(f/tPSA)ratio, and Prostate Imaging-Reporting and Data System(PI-RADS)scores differed significantly between the PCa and BPH groups(P＜0.05). Univariate and multivariate logistic regression identified age, f/tPSA ratio, and PI-RADS scores as independent influencing factors(P＜0.05). A logistic model incorporating these variables achieved an AUC of 0.785(95%CI:0.724-0.846, P＜0.05), with a sensitivity of 70.6% and specificity of 77.9%. The model showed excellent calibration. DCA indicated a positive net benefit when the risk threshold exceeded 7%. The CART model yielded an AUC of 0.769(95%CI:0.717-0.821), with the sensitivity of 63.5% and specificity of 79.9%. Within threshold probabilities of 5%-56%, the CART model provided higher net benefit. When the risk threshold reached 55%, the number of diagnosed cases of PCa began to converge with the actual number. Conclusion Compared with the CART model, the logistic model based on f/tPSA, PI-RADS score and age has a higher diagnostic efficacy for PCa and is the optimal model.
