Artificial intelligence in diagnostic, prognostic, and predictive genomic biomarkers for prostate cancer: Ready for prime time?

INTRODUCTION

Until recently, the widespread use of genetic markers in prostate cancer (PCa) has been limited by the complexities and cost of genomic data analysis. Artificial intelligence (AI), due to its ability to process large volumes of unstructured data, holds the potential to play a transformative role in the future of medical genetics.

METHODS

We conducted a systematic literature review using the Medline citation database and the Google Scholar search engine to evaluate the feasibility of AI applications in PCa diagnostics and disease progression prediction. We selected articles that presented data on the use of AI to identify genetic markers and/or their association with clinical data in patients with confirmed or suspected prostate cancer, without applying any time restriction. In total, 15 articles were included in the final analysis.

RESULTS

Studies investigating the application of AI in prostate cancer diagnosis have demonstrated that machine learning (ML) methods can be effectively used to identify novel cancer-related genes from genetic databases. Additionally, ML algorithms have shown potential in predicting clinical risk in PCa. By analyzing miRNAs, mRNAs, lncRNAs, and patterns of gene upregulation and alteration, AI has been able to predict adverse clinical outcomes such as metastatic progression, biochemical recurrence following radical prostatectomy, reduced survival, and elevated serum PSA levels. Moreover, AI tools have been utilized to predict genitourinary complications after radiation therapy through genome-wide data analysis, to identify cell line phenotypes resistant to antiandrogen therapy and to detect novel signaling pathways that may be targeted by emerging systemic treatments.

CONCLUSION

AI-based methods appear to be promising tools for the identification of new genetic biomarkers in PCa, offering potential for improvements in disease detection, prognosis, and prediction of treatment response, including the identification of actionable therapeutic targets. However, their clinical implementation remains limited due to a lack of clinical validation and practical benefit uncertainties.