Prostate cancer diagnosis using sensitive and sophisticated machine learning classifiers based on non-invasive...

Area of Science:

• Urology
• Oncology
• Biomarkers
• Machine Learning

Background:

• Prostate cancer (PCa) is a leading global malignancy in men.
• Current screening methods like PSA tests and DRE lack specificity, leading to overdiagnosis and overtreatment.
• There is a critical need for accurate, non-invasive diagnostic tools for PCa.

Purpose of the Study:

• To develop and validate a novel diagnostic framework, PCASSO, for accurate prostate cancer detection.
• To integrate machine learning (ML) with urinary RNA biomarkers for improved diagnostic performance.
• To assess the utility of this framework in the clinically ambiguous PSA gray zone.

Main Methods:

• Collected 163 urine samples (112 PCa, 51 benign prostatic hyperplasia [BPH]).
• Analyzed 20 urinary RNA biomarkers (lncRNAs, fusion gene, miRNAs) using quantitative PCR.
• Evaluated six ML classifiers, focusing on a Gradient Boosting model with an optimized 9-biomarker panel.

Main Results:

• The Gradient Boosting model achieved a high diagnostic accuracy (AUC: 0.99).
• Robust cross-validation results (Stratified-K-Fold: 0.912; LOOCV: 0.890) confirmed model stability.
• The classifier demonstrated high accuracy in patients within the PSA gray zone (3-10 ng/mL).

Conclusions:

• ML-based classification of DRE-free urinary RNA biomarkers offers a promising non-invasive approach for prostate cancer diagnosis.
• The PCASSO framework shows potential to enhance diagnostic accuracy and aid clinical decision-making.
• Further validation studies are warranted to establish this method in clinical practice.