Prostate cancer (PCa) exhibits pronounced racial disparities in incidence, with Black men in the United States demonstrating an incidence rate approximately 1.8-fold greater than their White counterparts, according to recent epidemiological data. Current diagnostic approaches relying on prostate-specific antigen (PSA) screening remain critically limited by insufficient specificity, while many emerging machine learning approaches overlook the critical influence of race-specific variations in molecular profiles for predictive modeling.

This study introduces a race-aware prostate cancer (PCa) detection framework designed to enhance diagnostic accuracy and equity through optimized feature selection. Leveraging RNA-seq (STAR-aligned count data) and clinical phenotype data from The Cancer Genome Atlas (TCGA) cohort (n = 554 patients), we developed a multi-stage feature selection pipeline integrating: 1. Differential Gene Expression (DGE) analysis, 2. Receiver Operating Characteristic (ROC) curve evaluation, and 3. Gene Set Enrichment Analysis (GSEA). This pipeline identified a 9-gene biomarker panel strongly enriched in prostate carcinogenesis pathways. To address racial disparities in model generalizability, the framework was trained on White population data and rigorously validated on a Black patient subset, employing synthetic minority oversampling (SMOTE) and cost-sensitive learning for dataset balancing.

The optimal model, leveraging a 9-gene biomarker panel, demonstrated 95% diagnostic accuracy in the White cohort and 96.8% accuracy in the Black cohort. Despite the Black population exhibiting marginally higher classification performance, a 4% fairness gap was identified through Demographic Parity fairness metrics, highlighting disparities in equitable outcomes. These results underscore the clinical validity of race-specific biomarkers, illustrating how biologically driven feature selection not only enhances diagnostic performance but also advances computational interpretability and resource efficiency.

This study introduces a race-aware prostate cancer (PCa) detection framework that enhances diagnostic precision through biologically informed biomarker selection. By mitigating misclassification risks inherent in demographically invariant models, the framework underscores the necessity of incorporating race-specific transcriptional signatures into machine learning-driven diagnostics. Beyond improving early detection, the approach facilitates personalized therapeutic stratification, aligning with broader goals to advance equitable precision oncology in prostate cancer management.