Scholarlink

Objective <p>This study aimed to develop, validate, and compare robust machine learning (ML) models that integrate readily available clinical data for predicting prostate cancer (PCa) and clinically significant PCa (csPCa) in patients within the prostate-specific antigen (PSA) gray zone from a multicenter cohort in Saudi Arabia.</p> Methods <p>A retrospective multicenter study was conducted using data from 704 patients in Riyadh, Saudi Arabia. Two predictive models were developed: Model 1 for any PCa (<i>N</i> = 240/704) and Model 2 for csPCa (<i>N</i> = 154/704) patients. The input features included age, PSA density, prostate volume, and mpMRI risk. The dataset was split into training (<i>N</i> = 563) and testing (<i>N</i> = 141) sets. To ensure a robust evaluation, k-fold cross-validation (k = 5) was performed on the training set to assess model stability before the final evaluation on the independent test set. Eight ML algorithms were evaluated, with performance assessed using Accuracy, F1-score, and the Area Under the Receiver Operating Characteristic Curve (AUC) on the test set.</p> Results <p>For clinically relevant Model 2 (csPCa prediction), the GradientBoost classifier demonstrated superior performance, achieving an AUC of 0.88 (95% CI: 0.83–0.93) and an accuracy of 0.82. For Model 1 (any PCa), GradientBoost also performed best, with an AUC of 0.897 and an accuracy of 0.83. Analysis of importance revealed that prostate volume was the most influential predictor in both models. Explainable AI (XAI) techniques, specifically SHAP (SHapley Additive exPlanations), were incorporated to provide transparency into the model’s decision-making process.</p> Conclusion <p>The developed machine learning model, particularly the GradientBoost classifier, is a promising tool for risk stratification of csPCa in the PSA gray zone. This model has the potential to aid in patient selection for prostate biopsy, thereby reducing unnecessary procedures and associated morbidity in the Saudi population. Further prospective validation is required to confirm its clinical impact. Integrating this model into clinical software could offer clinicians real-time, data-driven risk assessments, thereby optimizing personalized patient care.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predictive models for prostate cancer in patients with prostate-specific antigen levels in the gray zone using machine learning: a multi-center study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月12日

查看全文
  • Karim Farhat,
  • Mostafa Arafa,
  • Nesma Lotfy,
  • Farrukh Khan,
  • Abdulaziz Althunayan,
  • Alaa Mokhtar,
  • Waleed Altaweel,
  • Sultan Al-Khateeb,
  • Sami Azhari,
  • Danny Rabah

摘要

Objective

This study aimed to develop, validate, and compare robust machine learning (ML) models that integrate readily available clinical data for predicting prostate cancer (PCa) and clinically significant PCa (csPCa) in patients within the prostate-specific antigen (PSA) gray zone from a multicenter cohort in Saudi Arabia.

Methods

A retrospective multicenter study was conducted using data from 704 patients in Riyadh, Saudi Arabia. Two predictive models were developed: Model 1 for any PCa (N = 240/704) and Model 2 for csPCa (N = 154/704) patients. The input features included age, PSA density, prostate volume, and mpMRI risk. The dataset was split into training (N = 563) and testing (N = 141) sets. To ensure a robust evaluation, k-fold cross-validation (k = 5) was performed on the training set to assess model stability before the final evaluation on the independent test set. Eight ML algorithms were evaluated, with performance assessed using Accuracy, F1-score, and the Area Under the Receiver Operating Characteristic Curve (AUC) on the test set.

Results

For clinically relevant Model 2 (csPCa prediction), the GradientBoost classifier demonstrated superior performance, achieving an AUC of 0.88 (95% CI: 0.83–0.93) and an accuracy of 0.82. For Model 1 (any PCa), GradientBoost also performed best, with an AUC of 0.897 and an accuracy of 0.83. Analysis of importance revealed that prostate volume was the most influential predictor in both models. Explainable AI (XAI) techniques, specifically SHAP (SHapley Additive exPlanations), were incorporated to provide transparency into the model’s decision-making process.

Conclusion

The developed machine learning model, particularly the GradientBoost classifier, is a promising tool for risk stratification of csPCa in the PSA gray zone. This model has the potential to aid in patient selection for prostate biopsy, thereby reducing unnecessary procedures and associated morbidity in the Saudi population. Further prospective validation is required to confirm its clinical impact. Integrating this model into clinical software could offer clinicians real-time, data-driven risk assessments, thereby optimizing personalized patient care.