Open Access

ARTICLE

Machine learning-based comparison of transperineal vs. transrectal biopsy for prostate cancer diagnosis: evaluating procedural effectiveness

Mostafa Ahmed Arafa^1,2, Karim Hamda Farhat^1,*, Nesma Lotfy³, Farrukh Kamel Khan¹, Alaa Mokhtar⁴, Abdulaziz Mohammed Althunayan¹, Waleed Al-Taweel⁴, Sultan Saud Al-Khateeb⁴, Sami Azhari^1,5, Danny Munther Rabah^1,4

1 The Cancer Research Chair, Surgery Department, College of Medicine, King Saud University, Riyadh, 11472, Saudi Arabia
2 Department of Epidemiology, High Institute of Public Health, Alexandria University, Alexandria, 21521, Egypt
3 Department of Biostatistics, High Institute of Public Health, Alexandria University, Alexandria, 21521, Egypt
4 Department of Urology, King Faisal Specialist Hospital and Research Center, Riyadh, 11211, Saudi Arabia
5 College of Medicine, Alfaisal University, Riyadh, 11533, Saudi Arabia

* Corresponding Author: Karim Hamda Farhat. Email:

Canadian Journal of Urology 2025, 32(3), 173-180. https://doi.org/10.32604/cju.2025.066016

Received 27 March 2025; Accepted 30 May 2025; Issue published 27 June 2025

Abstract

Background: Transrectal (TR) and transperineal (TP) biopsies are commonly used methods for diagnosing prostate cancer. However, their comparative effectiveness in conjunction with machine learning (ML) techniques remains underexplored. This study aimed to evaluate the predictive accuracy of ML algorithms in detecting prostate cancer using data derived from TR and TP biopsies. Methods: The clinical records of patients who underwent prostate biopsy at King Saud University Medical City and King Faisal Specialist Hospital and Research Centerin Riyadh, Saudi Arabia, between 2018 and 2025 were analyzed. Data were used to train and test ML models, including eXtreme Gradient Boosting (XGBoost), Decision Tree, Random Forest, and Extra Trees. Results: The two datasets are comparable. The models demonstrated exceptional performance, achieving accuracies of up to 96.49% and 95.56% on TP and TR biopsy datasets, respectively. The area under the curve (AUC) values were also high, reaching 0.9988 for TP and 0.9903 for TR biopsy predictions. Conclusion: These findings highlight the potential of ML to enhance the diagnostic accuracy of prostate cancer detection irrespective of the biopsy method. However, TP biopsy data showed marginally higher accuracy, possibly because of the lower risk of contamination. While ML holds great promise for transforming prostate cancer care, further research is needed to address limitations. Collaboration between clinicians, data scientists, and researchers is crucial to ensure the clinical relevance and interpretability of ML models.

---

Keywords

---